Prepared for

Agentic AI

as a Managed BPO

Your next BPO isn't a BPO. It's an AI agent workforce that works alongside your team — owned, operated, and optimised by Norfolk AI. Voice, WhatsApp, and Chat. Giving your people superpowers.

April 2026 · Confidential

Cross-Functional

CX · Support · Technical · Sales

One agent workforce covering customer operations, technical support, account management, and sales qualification simultaneously. We spin up agent swarms across every department — not just one. Cirion's human specialists stop handling Tier-1 volume entirely.

Cross-Channel

Voice · WhatsApp · Chat

A customer who calls, then follows up on WhatsApp, then opens a chat session gets a single continuous conversation. Full context carries across every channel — no re-verification, no repeated explanations. One agent swarm, three channels, zero context loss.

Agentic AI

Reasoning, not scripts

Every human escalation is an edge case codification event. The resolution becomes a guardrail update — expanding the resolution envelope so the same situation is handled autonomously next time. The system compounds. It does not depreciate.

Local LATAM

Austin TX · São Paulo · Trilingual

Headquartered in Austin, Texas, with on-the-ground presence in São Paulo, Brazil. The entire team operates in English, Spanish, and Portuguese — and has done so since before Norfolk AI existed. WhatsApp Business API governance, LATAM data residency, and local escalation protocols are built into the operating model from day one. Not bolted on.

Prepared For

Cirion Technologies

Prepared By

Norfolk AI

Date

April 2026

Version

v1.0

Section 01

Executive Summary

The strategic case for agentic AI as a managed CX and support capability for Cirion

The core thesis: Cirion Technologies operates hyperscaler-grade infrastructure across Latin America — colocation, cloud connectivity, SD-WAN, and enterprise broadband — but its customer operations still run on human-only teams at significant cost and inconsistent quality. Norfolk AI deploys purpose-built agent swarmsA group of AI agents deployed as an operational unit. Swarms scale elastically — more agents during peak volume, fewer during quiet periods, and specialised swarms for specific task types and intensity levels. across Voice, WhatsApp Business API, and Chat to work alongside Cirion's human specialists — handling routine Tier-1 interactions autonomously and giving the human team superpowers on everything else. This is not a scripted bot, and it is not a replacement for your people. It is a managed AI agent workforce that absorbs volume, prepares context, drafts responses, and recommends next actions — so your team spends their time on the decisions that actually require a person. When an interaction falls outside the agent's guardrail-bounded autonomyAgents operate freely within defined policy, compliance, and behavioral guardrails. When an interaction exceeds those boundaries, the agent pauses and escalates with full context. The human decision is then codified back into the guardrail library., it pauses, routes to a human with full context, and codifies that decisionThe three-step protocol when an agent reaches the boundary of its guardrail-bounded autonomy: (1) Pause — stop the interaction and preserve full context. (2) Route — hand off to a human specialist with the complete conversation loaded. (3) Codify — convert the human's resolution decision into a reusable guardrail rule, expanding the resolution envelope for next time. into the guardrail library. Every escalation expands the resolution envelopeThe boundary of what an agent swarm can resolve autonomously. Every escalation that gets codified into the guardrail library expands this envelope — the swarm handles more, humans handle less.. Every human decision becomes compounding intelligenceThe system-level effect where every human escalation decision is codified into the agent's guardrail library, expanding the resolution envelope. Each interaction makes the next one better — intelligence compounds over time.. The result: Cirion's team focuses on judgment calls that genuinely require a person, while agent swarms absorb the volume — with CRM-grade auditability on every interaction.

40–60%

Routine Tier-1 interactions handled autonomously ²

~25%

Handle time reduction when AI assists human agents ³

2.4×

First contact resolution improvement vs. scripted bots

6–12 wks

Time-to-value: first live customer conversations ⁴

What This Means for Cirion

Cirion's human specialists spend far less time on password resets, service status checks, and ticket routing. Agent swarms absorb that routine volume and provide real-time assistance on more complex interactions — surfacing context, drafting responses, and recommending next actions. The human team focuses on what they do best: complex escalations, high-value account reviews, and the edge cases that build institutional knowledgeThe accumulated resolution patterns, edge case decisions, and workflow expertise that compounds inside the guardrail library over time. Unlike human memory, institutional knowledge in Norfolk AI's system is persistent, searchable, and automatically applied to future interactions.. Every exception a human resolves is codified back into the guardrail library — an edge case codificationThe process of converting a human escalation decision into a reusable guardrail rule. Once codified, the same edge case is handled autonomously next time — no repeat escalation required. event — expanding the resolution envelope for next time. The system compounds — it does not depreciate.

Business Outcomes at Scale

24/7 availability, zero missed calls, sub-60s response

24/7

Agent availability

No shift patterns, no holidays, no sick days — agent swarms scale elastically with demand

Missed calls

Agent swarms scale elastically to meet demand peaks — no call queues during normal operations

≤60s

First response time

From inbound contact to agent reply, across Voice, WhatsApp, and Chat

<3 min

Avg. Tier-1 resolution

For standard service interactions — with full CRM write-back on every case

Three Channels — One Managed Service

Voice, WhatsApp Business API, and Chat — configured for Cirion's workflows

Norfolk AI deploys AI agents across Cirion's three primary customer engagement channels — each configured for the interaction patterns, compliance requirements, and CRM workflows specific to that channel.

Inbound

Voice Channel

Inbound call handling for customer operations and technical support. Norfolk AI agents handle routine Tier-1 interactions — service status checks, password resets, ticket creation — and provide real-time support on more complex calls by surfacing context, drafting responses, and recommending next actions. When a call requires human judgment, the specialist inherits the conversation mid-stride with full context already loaded — no re-verification, no repeated explanations.

LATAM-Native

WhatsApp Business API

Compliant, template-governed messaging for service alerts, onboarding milestones, proactive issue notifications, and support follow-ups. The dominant enterprise engagement channel in LATAM — 3B+ monthly active users globally ⁵, 500M+ active users across Latin America ¹¹ — managed entirely by Norfolk AI, including WhatsApp Business Platform setup, BSP partnership where applicable, opt-in governance, and message-template approval workflows.

Web + Portal

Chat Channel

Website and portal chat for inbound service resolution, FAQ containment, and live-agent escalation. When a conversation requires human judgment, the full thread — every message, every system action, every CRM state change — is passed to the human agent in a single handoff packetThe structured context bundle an agent assembles before routing to a human specialist. Includes conversation history, customer profile, attempted resolutions, and the specific reason for escalation — so the human never has to ask the customer to repeat themselves.. Zero context loss. The human picks up exactly where the AI left off.

Section 02

Business Case

Why the shift from conversational AI to agentic AI changes Cirion's cost-to-serve equation

The Shift

From Answering Questions
to Getting Work Done

Conversational AI answers questions. Agentic AI gets work done. The distinction is not semantic — it determines whether Cirion's investment produces deflection metrics or genuine customer experience outcomes. Norfolk AI's agents don't just respond; they resolve issues, create tickets, update CRM records, trigger escalation workflows, and hand off to human agents with full context. Critically, this is not a model built on replacing human agents — it is built on giving them operating leverageThe ability to handle more interactions without proportionally increasing headcount. Agent swarms deliver operating leverage by resolving high-frequency, repeatable workflows autonomously.. AI handles the volume; humans handle the judgment. The result is a measurable improvement in first contact resolution (FCR), customer satisfaction (CSAT), and a compounding reduction in cost-to-serve ⁸. By 2029, Gartner predicts agentic AI will handle 80% of common customer operations issues, leading to a 30% reduction in operational costs ⁷ ⁸. Norfolk AI's approach ensures your human team isn't sidelined by this shift — they're amplified by it, focusing on the complex cases where their judgment creates the most value.

Traditional Conversational AI

Scripted flows and decision trees

Responds to explicit questions only

No ticket write-back — context lost on escalation

Requires human follow-up for every unresolved issue

Containment rate: 10–20% of interactions

Degrades without continuous internal investment

Norfolk AI Agentic Workforce

Goal-directed agent swarms with reasoning engines

Proactive service alerts and issue notifications

Bidirectional ticketing + CRM integration — full auditability

Autonomous resolution of Tier-1 and Tier-2 issues at scale

Autonomous resolution rate: 40–60% — expanding quarter over quarter ²

Edge cases route to humans with full context — zero re-verification

Edge case codification: every human decision expands the resolution envelope

Compounding intelligence — each interaction makes the next one better

Section 03

What We Deliver

CX, support, and service workflow coverage across Cirion's customer operations

Managed Outcomes

Six Customer Workflows.
Fully Managed.

Norfolk AI does not deliver software. It deploys agent swarms — purpose-built, continuously optimised, and operationally accountable for Cirion's six highest-value customer workflows across service, support, and account management. Each workflow maps to a measurable CX or efficiency outcome, not a feature set. The swarms scale elastically: more agents spin up during peak volume, fewer during quiet periods. Cirion pays for outcomes, not idle capacity.

Revenue Workflow	Engagement Channel	Agent Action	CRM Output	Business Outcome
Inbound Customer Operations	Voice + Chat	Tier-1 resolution, FAQ containment, ticket creation	Ticket created, status updated	FCR improvement + CSAT
Technical Support (Tier-1)	Voice + Chat	Diagnostic scripts, knowledge base retrieval, escalation routing	Ticket created, context logged	MTTR reduction + agent workload relief
Proactive Service Alerts	WhatsApp	Outage notifications, maintenance windows, service updates	Notification sent, response logged	Customer satisfaction + reduced inbound volume
Onboarding Follow-up	WhatsApp + Chat	Milestone check-ins, document collection, activation confirmation	Onboarding stage updated	Churn reduction + improved activation rates
Renewal & Account Management	WhatsApp	Contract renewal reminders, usage alerts, account health check-ins	Renewal flag, CSM alert	Retention + upsell identification
Inbound Sales Enquiries	Voice + Chat	Product qualification, pricing guidance, meeting booking	Lead status + meeting logged	Pipeline acceleration (inbound only)

WhatsApp Business API compliance: All WhatsApp interactions are governed by Meta's Business Messaging Policy. Norfolk AI manages template registration, opt-in flows, and the 24-hour session window as part of the managed service — Cirion does not need to manage this complexity internally. Technical details are available in Appendix A3.

Section 04

Why Norfolk AI

The managed CX operations model vs. platform vendors and consulting engagements

Operating Model

Why Norfolk AI — Not a Platform, Not a Consultancy.
A Managed Differentiator.

The buyer choice is not "which AI vendor." It is which operating model. Platform vendors — the Sierras, the PolyAIs — sell tooling. The customer staffs, trains, and maintains the AI. Consulting firms implement stacks and hand the keys back on day one. Norfolk AI delivers customer experience outcomes as a fully managed service: we own the agents, operate them, and continuously expand their resolution envelope. The fundamental design principle is guardrail-bounded autonomyAgents operate freely within defined policy, compliance, and behavioral guardrails. When an interaction exceeds those boundaries, the agent pauses and escalates with full context. The human decision is then codified back into the guardrail library., not human replacement. Agent swarms handle the volume — the repeatable, high-frequency interactions that consume human capacity without requiring human judgment. Edge cases, complex disputes, and high-stakes decisions route to Cirion's human specialists with the full interaction context already loaded — no re-verification, no repeated explanations, no context loss. Every human decision is codified into the guardrail library by Norfolk AI's in-house agent operations team — Cirion does not need to hire prompt engineers, maintain a model ops team, or manage the feedback lo The resolution envelope expands. The system compounds — it does not depreciate.

Dimension	Platform Vendors	Consulting Firms	Norfolk AI BPO
What you get	Software licence	Implementation project	Managed AI agent workforce — we own and operate the agents
Who operates it	Your team	Your team (post-project)	Norfolk AI team
Ongoing optimisation	Manual — your responsibility	Out of scope	Continuous — included
Time-to-value	Months (build + QA)	Months (design + build)	6–12 weeks to live agent swarms
Risk model	Customer bears all risk	Customer bears all risk	Norfolk AI operationally accountable for outcomes
LATAM / WhatsApp	Partial	Limited	Native
Human-AI Collaboration	Customer staffs oversight	Customer staffs oversight	Guardrail-bounded autonomy + human oversight, Norfolk AI managed
Outcome accountability	None	Delivery only	Outcome-backed performance

Agent Design & Deployment

We Build It

Norfolk AI designs, trains, and deploys every agent swarm from the ground up. Cirion provides the business context — workflows, CRM schema, escalation rules, and the edge cases that matter most. We do the engineering, the integration, and the QA. Cirion gets a live agent workforce without writing a single line of code or hiring a single prompt engineer.

Day-to-Day Management

We Operate It

Our team monitors agent swarm performance, handles incidents, and manages the operational layer 24/7. When an agent behaves unexpectedly, we catch it before Cirion's customers do. Model drift, prompt degradation, hyperscaler API changes — that is our problem, not Cirion's. We are operationally accountable.

Continuous Improvement

We Optimise It

Every edge case that reaches a human specialist is reviewed, codified, and fed back into the guardrail library — expanding the resolution envelopeThe boundary of what an agent swarm can resolve autonomously. Every escalation that gets codified into the guardrail library expands this envelope — the swarm handles more, humans handle less.. Monthly knowledge base updates, prompt refinement, and workflow expansion mean the system delivers compounding intelligenceThe system-level effect where every human escalation decision is codified into the agent's guardrail library, expanding the resolution envelope. Each interaction makes the next one better — intelligence compounds over time.. The autonomous resolution rate goes up quarter over quarter — without additional integration work from Cirion.

Guardrail-Bounded Autonomy

Human-AI Collaboration

Norfolk AI agent swarms work alongside Cirion's team — not instead of them. Routine interactions are handled within guardrail-bounded autonomyAgents operate freely within defined policy, compliance, and behavioral guardrails. When an interaction exceeds those boundaries, the agent pauses and escalates with full context. The human decision is then codified back into the guardrail library.. When a case falls outside the agent's defined envelope, it pauses, routes to a human specialist with the full conversation context already loaded, and codifies that resolution into the guardrail library. Every human decision becomes compounding intelligence. Your team shifts to higher-value oversight — and the resolution envelope expands with every escalation.

Bidirectional Data Contracts

Deep CRM Integration

Agent swarms read context from and write outcomes to Cirion's CRM in real time. No data loss on handoff. Every interaction produces a structured CRM record — qualified lead, support ticket, booked meeting, or renewal risk flag. When a human takes over, they inherit the full interaction history, not a summary. The CRM becomes the single source of truth for both human and AI operations.

Spanish + Portuguese + WhatsApp

LATAM-Native

Bilingual agents, WhatsApp Business API governance, and deep familiarity with LATAM enterprise interaction patterns. BSP compliance, opt-in governance, and message-template approval workflows are built into the operating model — not bolted on. The dominant channel in the region, managed entirely by Norfolk AI from day one.

Market Validation · Agentic AI at Enterprise Scale

Proven at Scale — Before You Sign Anything

The managed agentic AI model Norfolk AI brings to Cirion is not a hypothesis. AEGEA — Brazil's largest private water and sanitation utility, serving 31 million customers across 16 states — deployed a WhatsApp-first AI agent workforce and measured the results. The platform vendor in that case was Cognigy; the operating model was self-managed. Norfolk AI delivers the same outcomes as a fully managed BPO service — Cirion does not need to staff, train, or maintain the AI.

Source

Cognigy · AEGEA Case Study ↗

87%

AI retention rate

Customers resolved without human escalation

80%

Reduction in CSR escalations

Human specialists freed for complex cases

1.1M

Conversations / month

Peak throughput on WhatsApp Business API

31M

Customers served

Across 16 Brazilian states

Why this matters for Cirion: Cognigy is a platform vendor — they sell the tooling; the customer staffs, trains, and operates the AI. That model is not available to Cirion as a managed service. Norfolk AI delivers the same category of outcome — WhatsApp-first, high-volume, enterprise-grade agentic resolution — as a fully owned and operated BPO capability. Cirion gets the results without the operational overhead.

Deep Integration Partner

10.123 Ventures

AI Deep Integration Specialists

When the integration goes deep, so does the team.

For engagements requiring custom-built connectors, non-standard CRM configurations, or advanced workflow automation beyond the standard Factory deployment, Norfolk AI partners with 10.123 Ventures — a specialist AI deep integration firm. Their engineers work alongside Norfolk AI's in-house team on multi-system data pipelines and bespoke API orchestration. Cirion gets the combined capability without managing two vendor relationships.

Local Enterprise Partner · Brazil

SM Services

São Paulo, Brazil

No remote-only go-to-market. Local relationships from day one.

Norfolk AI's commercial expansion in Brazil is anchored by SM Services — a São Paulo-based enterprise sales partner with 20+ years prospecting IT and Telecom accounts across Latin America. Their Metodologia de Inteligência Comercial maps target accounts, qualifies opportunities, and builds the pipeline that connects Norfolk AI's managed AI workforce to Cirion-scale buyers in the region.

04.5

Human-AI Collaboration Model

How AI agents and Cirion's team work together — and where the handoff happens

Norfolk AI agent swarms handle the high-volume, repeatable interactions within guardrail-bounded autonomyAgents operate freely within defined policy, compliance, and behavioral guardrails. When an interaction exceeds those boundaries, the agent pauses and escalates with full context. The human decision is then codified back into the guardrail library. — freeing Cirion's team for the work that genuinely requires human judgment, empathy, and authority. The model is not AI replacing humans. It is AI amplifying them. Every escalation codifies a new edge case. Every codified edge case expands the resolution envelopeThe boundary of what an agent swarm can resolve autonomously. Every escalation that gets codified into the guardrail library expands this envelope — the swarm handles more, humans handle less.. The system delivers compounding intelligenceThe system-level effect where every human escalation decision is codified into the agent's guardrail library, expanding the resolution envelope. Each interaction makes the next one better — intelligence compounds over time..

Interaction Flow: AI-Handled vs. Human-Escalated

Customer Contact

Inbound interaction arrives

Voice · WhatsApp · Chat

Norfolk AI Agent

Agent handles the interaction

Retrieves context · Reasons · Acts · Logs to CRM

Routine — Resolved

AI handles routine interactions

Account lookup & status

Ticket creation & routing

FAQ & policy answers

Appointment scheduling

Proactive notifications

CRM record written

Ticket · Activity · Status update

Edge Case — Escalated

Routes to human agent

Complex disputes & exceptions

High-value contract decisions

Regulatory / compliance issues

Emotionally sensitive situations

Novel scenarios outside policy

Human resolves with full context

Full conversation history passed

↺

Edge cases become training data

Every human resolution is reviewed and codified — the agent learns from each exception, expanding its autonomous resolution envelope over time.

The Team Behind the Service

Built by People Who Have Done This Before

Norfolk AI is headquartered in Austin, Texas, with on-the-ground presence in São Paulo, Brazil. The leadership team brings direct enterprise software and operations experience across Latin America — not as a market entry strategy, but as the founding context from which the company was built.

Ricardo Cidale

Founder & CEO · Norfolk AI

Ricardo Cidale has scaled revenue units at Dell, HP, and RealNetworks, managed P&Ls exceeding $250M, and built teams of 300+ people aimed at measurable outcomes. He founded and exited LabOne Systems — growing it from 4 employees to 220 people across 18 countries and $60M ARR, with Telefónica as the anchor customer at $11M+ ARR over a decade, before exiting to Agile Content at nearly $100M valuation. He brings an expansive professional network spanning the Valley, Seattle, Miami, Mexico City, São Paulo, Madrid, and Barcelona — documented in LinkedIn and CRM systems. Mentor at MassChallenge and Capital Factory. Venture Partner at Synapse Ventures. Published author (McGraw-Hill). Triple citizenship: U.S., Brazil, Italy. Trilingual: English, Spanish, and Portuguese.

View LinkedIn Profile

LATAM Enterprise Heritage

LabOne Systems · 18 Countries

Several Norfolk AI team members are alumni of LabOne Systems — a company Ricardo Cidale built from 4 employees to 220 people across 18 countries, competing head-to-head against Microsoft and Accenture. LabOne's largest customer was Telefónica at $11M+ ARR over a decade. The company also served Claro and similar LATAM multinationals, providing enterprise content management platforms at the same scale and operational complexity that Cirion operates in today. No venture funding — every dollar came from customer value. That institutional knowledgeThe accumulated resolution patterns, edge case decisions, and workflow expertise that compounds inside the guardrail library over time. Unlike human memory, institutional knowledge in Norfolk AI's system is persistent, searchable, and automatically applied to future interactions. is what Norfolk AI brings to this engagement.

Trilingual Team

English · Spanish · Portuguese

Every member of the Norfolk AI team operates fluently in English, Spanish, and Portuguese. Customer conversations, vendor negotiations, CRM documentation, and escalation protocols run in the language of the interaction — without translation lag, without cultural friction, and without the coordination overhead that comes from routing between language-specific teams.

Super Conversations™

Proprietary Methodology · Norfolk Consulting Group

Super Conversations™ is a B2B engagement methodology developed by Ricardo Cidale over five years of studying how exceptional communicators create genuine connection. After reading Thaler and Sunstein's Nudge, Ricardo recognized he had been architecting choice environments for decades. Super Conversations™ codifies those patterns so they can be taught — and now, encoded into AI agents. The result: agents that don't just respond, but engage — making the right path the easy one for the customer.

Seven Pillars

CuriosityPersonalizationActive ListeningLong-Term RelationshipsStructure & AdaptabilityDifferentiationFriction Reduction

Technology Partners

Best-in-Class Stack. No Lock-In.

Norfolk AI assembles and operates the leading enterprise AI infrastructure for each capability layer — agent orchestration, voice intelligence, large language models, and voice synthesis. The entire stack runs on hyperscalerLarge-scale cloud infrastructure providers — Google Cloud, Microsoft Azure — that provide the compute, storage, and networking backbone for enterprise AI deployments. compute (Google Cloud, Microsoft Azure). For complex integrations and custom engineering work, Norfolk AI teams up with 10.123 Ventures, a specialist AI deep integration firm. Cirion gets the best available technology at each layer without managing vendor relationships, API contracts, or model upgrades. We own the integration. We own the operations. The technology compounds.

Section 05

Service Tiers & Investment

Three tiers aligned to Cirion's operational maturity and growth ambitions

Pricing

Service Tiers — Start Where It Makes Sense.
Scale When It Works.

Every tier is a fully managed service — Norfolk AI handles deployment, optimisation, and operations. The difference is scope: how many channels, how many workflows, and how much dedicated support.

Foundation

$10,000/month

Single-channel deployment for one primary CX or support workflow

1 AI agent (Voice or WhatsApp Business API)

1 workflow (e.g., inbound customer operations or technical support)

Ticketing system + CRM integration

Standard knowledge base setup

Ongoing agent management & guardrail updates by Norfolk AI's in-house team

Monthly performance report

Email support

Enterprise cloud hosting (Azure) & monitoring

RECOMMENDED

Growth

$20,000/month

Multi-channel deployment across 2–3 CX, support, and service workflows

3 AI agents (Voice + WhatsApp + Chat)

3 workflows (customer operations + technical support + onboarding)

Full CRM + ticketing integration (HubSpot / Salesforce / Freshdesk)

Advanced knowledge base + RAG

Weekly performance dashboards

Dedicated success manager

Enterprise cloud hosting (Azure) + monitoring

Priority support with dedicated escalation path

Enterprise

$30,000/month

Full AI agent workforce working alongside your team across all channels and workflows

Unlimited agents across all channels

Full workflow coverage (all 6 CX and support workflows)

Agentforce + HubSpot + Salesforce + Freshdesk + ERP

Custom knowledge base + continuous training

Real-time dashboards + executive reporting

Dedicated engineering + success team

Enterprise cloud hosting (Azure) & monitoring

24/7 priority support with dedicated engineering team

White-label option available

Capability Breakdown

Deliverables Matrix

All tiers include custom software development, hosting, security controls, and integrations to knowledge bases and CRM systems. The difference is scope, depth, and the sophistication of the workflows Norfolk AI operates on Cirion's behalf.

Capability Area	Level 1 — Foundation	Level 2 — Growth	Level 3 — Enterprise
Channels	Inbound calls; web widget (voice & text); WhatsApp inbound	Adds WhatsApp compliant proactive sequences	Full omnichannel with multi-agent handoffs
CRM Integration	Contact/lead creation, activity logging, summaries	Bi-directional enrichment, deal/opportunity workflows	Deep agentic orchestration; Agentforce action graphs
Analytics + QA	Volume, containment, transfer, qualified leads	QA regression suite; monthly optimization; QBR pack	Continuous experimentation; cohort analysis; attribution
Security + Hosting	Enterprise cloud hosting (Azure) + monitoring	Expanded audit logging + governance	Dedicated environments; optional lab pattern
Support	Brazil-based bilingual L3 support	Priority support + expanded hours	Fastest response times; dedicated delivery pod

White-Label BPO Option

An optional White-Label BPO addendum enables Cirion to resell AI agents under Cirion branding to enterprise customers across the region, while Norfolk AI provides L3 operations, platform development, and governance support. Cirion controls end-customer-facing branding; Norfolk AI maintains platform integrity and security baselines.

Revenue Opportunity: Cirion can offer "Cirion-branded AI Agents" as a managed service to its enterprise customer base — creating a new recurring revenue stream without building or operating the underlying technology. Norfolk AI handles all platform operations; Cirion handles Tier 1/2 customer success.

Section 06

Onboarding & Deployment

A structured 10–12 week rollout designed for enterprise-grade safety, CRM integration depth, and geographic scale

Deployment Timeline

Live in 10–12 Weeks

Norfolk AI's onboarding process is structured for enterprise complexity. The six-step rollout moves from discovery to live conversations in 10–12 weeks, with a controlled soft launch phase that limits risk before full deployment. Timelines may extend based on CRM integration depth, compliance requirements, and geographic deployment scope.

Weeks 1–3

Discovery & Scoping

Kickoff workshop with Cirion stakeholders. Map workflows, CRM schema, escalation rules, and integration requirements. Define agent mission, behavioral guardrails, and success metrics.

Weeks 3–6

Agent Design

Configure conversation logic, grounded knowledge base, and tool integrations. Build and test agent personas, escalation paths, and CRM write-back logic in staging environment.

Weeks 6–9

Integration & Testing

Connect to production CRM, telephony, and messaging APIs. End-to-end testing with real data. UAT with Cirion team. Compliance review for WhatsApp Business API templates.

Weeks 9–12

Soft Launch

Deploy to 10–20% of live traffic. Monitor autonomous resolution rate, escalation rates, and CRM data quality. Iterate on agent responses based on real conversation data.

Months 4–6

Full Deployment

Scale to 100% of target traffic. Activate all channels and workflows. Establish weekly performance review cadence with Cirion success team.

Month 7+

Continuous Optimisation

Monthly knowledge base updates, prompt refinement, and workflow expansion. Quarterly business reviews with operating leverage reporting and roadmap planning.

Service Commitments

Enterprise

Hosting

Azure-hosted, monitored 24/7

Priority

Incident Response

Dedicated escalation path

Continuous

Optimisation

Monthly knowledge base updates

60 days

Cancellation

Notice period, any time

Commercial Terms

Contract & Engagement Terms

Standard terms for the Norfolk AI managed service engagement

Term & Cancellation

Initial term: 36 months from go-live date

Cancellation: 60 days' written notice, at any time

No lock-in penalty — exit with 60 days' notice

Pricing fixed for the full 36-month term

Renewal terms agreed in writing before expiry

Data Ownership

Cirion retains full ownership of all customer data

Conversation transcripts belong to Cirion

CRM data written by agents is Cirion's property

Norfolk AI holds no data rights post-termination

Data deletion within 45 days of contract end

Transition Assistance

30-day handover period on contract end

Full export of agent configurations and knowledge bases

CRM integration documentation provided

Conversation history export in standard formats

Optional: 30-day parallel running with successor vendor

Annual Pricing Review

Pricing fixed for full 36-month term

Review aligned to CPI + performance metrics at renewal

Volume discounts available at 36-month renewal

Scope expansion priced on a per-workflow basis

White-label option: available on Growth and Enterprise tiers

Telephony & SIP — Excluded from BPO Scope

Telephony infrastructure costs — including SIP trunking, CPaaS platform fees (e.g. Twilio, Vonage/Ericsson), PSTN termination, DID provisioning, and per-minute voice charges — are not included in the Norfolk AI BPO fee. Cirion retains full control of its telephony vendor selection and commercial terms. Norfolk AI integrates with Cirion's chosen telephony provider via standard SIP/WebRTC interfaces at no additional integration cost.

Infrastructure & Tech Stack — Included in BPO Fee

All AI infrastructure and technology stack costs are fully covered by the Norfolk AI BPO fee. This includes hyperscaler compute (Google Cloud, Microsoft Azure), LLM inference costs, vector database hosting, agent orchestration platform, monitoring and observability tooling, knowledge base infrastructure, and all software licences required to operate the Agent Factory. Cirion bears no separate infrastructure, compute, or platform costs.

Governing Law & Liability: This engagement is governed by the laws of the applicable jurisdiction agreed in the Master Services Agreement. Norfolk AI's total liability is capped at three months' fees. Cirion's data is processed in accordance with applicable data protection regulations (LGPD, GDPR-equivalent). Full terms available in the MSA on request.

AI Intellectual Property: All AI agent intellectual property developed under this engagement — including but not limited to agent prompting, system prompts, context architectures, agent programming, API integrations, voice configurations, conversation workflows, orchestration logic, knowledge base structures, guardrail libraries, and the agents and sub-agents themselves — remains the exclusive property of Norfolk AI. Cirion receives a fully managed service and a perpetual, non-exclusive licence to use the agent outputs and CRM-written data. No source code, model weights, prompt templates, or system architectures are transferred to Cirion under this agreement.

Currency: All amounts, pricing, and fees quoted in this proposal are denominated in United States Dollars (USD). Invoicing and payment are conducted in USD unless otherwise agreed in writing in the Master Services Agreement.

Service Delivery: All services under this engagement are delivered remotely from Norfolk AI's operating locations (Austin, TX / São Paulo, BR). Norfolk AI's delivery obligation is fulfilled upon electronic transmission of the managed service. Risk of service interruption due to network conditions, third-party infrastructure, or force majeure events beyond Norfolk AI's points of presence is governed by the service schedule in the Master Services Agreement. Service commitments and remedies are set forth therein.

Confidentiality & Non-Disclosure: Both parties agree to treat all proprietary information exchanged during this engagement as strictly confidential. This includes, without limitation, business strategies, customer data, technical architectures, pricing structures, agent configurations, workflow designs, and any materials marked or reasonably understood to be confidential. Neither party shall disclose confidential information to third parties without prior written consent, except as required by law or regulation. Confidentiality obligations survive termination of the agreement for a period of three (3) years. Detailed non-disclosure terms are set forth in the mutual NDA executed alongside the Master Services Agreement.

Section 07

Competitive Positioning

Cross-functional · Cross-channel · Agentic · Local — the combination no one else offers in LATAM

The Unfair Advantage

Competitive Landscape — Four Things at Once. Nobody Else Does That.

There are plenty of WhatsApp BPO providers. There are plenty of AI platforms. There are consulting firms that will implement either. What does not exist — anywhere in the LATAM market — is a single managed service that deploys cross-functional agent swarmsA coordinated group of AI agents that handles multiple business functions simultaneously — CX, technical support, customer operations, and sales — within a single operational workforce. Unlike single-purpose bots, cross-functional swarms share context across functions, resolve handoffs internally, and scale elastically across all channels without siloed tooling. (CX, technical support, customer operations, and sales handled by one workforce), across all channels (Voice, WhatsApp, and Chat unified under one operational layer), with agentic reasoningGoal-directed problem-solving by an AI agent — not scripted responses, but dynamic planning. The agent observes context, reasons about the best path to a defined goal, takes action, evaluates the outcome, and updates its approach. This enables agents to handle novel situations that fall within their guardrail boundaries without explicit programming for every scenario. (goal-directed, not scripted), locally operated in LATAM, and backed by a dedicated enterprise sales partner on the ground in Brazil. That combination is Norfolk AI's unfair advantage — and it is the reason the comparison below looks the way it does.

Analyst Intelligence · Gartner, 2025

"By 2029, agentic AI will handle 80% of common customer service issues, leading to a 30% reduction in operational costs." Norfolk AI ensures your team leads this shift — not gets replaced by it.

Source: Gartner Newsroom Press Release · Daniel O'Sullivan, Senior Director Analyst

Reference ⁷

Reading the Landscape

The competitive landscape for AI-powered customer operations in LATAM splits into four camps, none of which solves Cirion's problem on its own. Global AI platforms (Sierra, PolyAI) have strong agentic technology but are built primarily for North American and European enterprise patterns — they require Cirion to staff and run operations internally, and their infrastructure runs on hyperscaler compute that Cirion would need to manage. Consulting firms (Slalom) deliver implementations but hand the keys back to the client on day one — despite their growing LATAM delivery presence in Mexico and Colombia, they are not a managed operations provider. LATAM messaging and AI platform specialists (Blip, Weni by VTEX) have deep local market knowledge and WhatsApp fluency, but are primarily platform vendors — they require Cirion to build and manage its own agent workflows rather than operating as a managed service. LATAM AI startups (Odisseia AI) are building toward agentic capabilities but remain primarily focused on pre-sales lead qualification and early-stage CX via WhatsApp — not the full range of cross-functional, multi-channel service operations that Cirion requires.

The table below scores each provider across the six dimensions that actually matter for Cirion's deployment. The pattern is consistent: every competitor is strong in one or two areas and weak in the rest. Norfolk AI is the only provider that scores across all six — because it was built to be all six at once.

Dimension	Norfolk AI Managed BPO	Sierra AI Platform	PolyAI Voice AI	Slalom Consulting	Blip LATAM Messaging	Weni by VTEX ex-Weni, LATAM Messaging AI	Odisseia AI LATAM Pre-Sales	Regional BPO Legacy Contact Centre
Agentic AI Goal-directed reasoning, tool use, and escalation — not scripted flows	High	High	Medium	Medium	Low	Medium	Medium	None
Managed Operations Vendor runs and owns day-to-day agent operations	High	Low	Medium	Low	Medium	Low	Low	High
Continuous Optimisation Ongoing improvement included — not a separate project	High	Limited	Limited	Low	Limited	Limited	Limited	Low
KEYLATAM Native Bilingual (ES/PT) · local ops team · LATAM enterprise market depth	Native	Partial	Limited	Partial	Strong	Strong	Strong	Strong
Cross-Functional CX, technical support, customer operations, and sales in one agent workforce	High	Medium	Medium	Medium	Low	Low	Low	Low
Cross-Channel Voice, WhatsApp, and Chat unified under one operational layer	High	Low	Medium	Low	Medium	Medium	Low	Low
Time-to-Value Time from kickoff to live customer conversations	6–12 wks	3–6 mo	2–4 mo	4–8 mo	4–8 wks	4–8 wks	2–4 wks	4–8 wks
Outcome Accountability Vendor is operationally accountable for performance	Yes	No	No	No	No	No	No	Partial
KEYPartner Ecosystem Named deep integration partner for complex custom engineering (multi-system pipelines, bespoke API orchestration)	10.123 Ventures	None	None	Ad-hoc	None	None	None	None
KEYLocal Sales Presence SM Services (São Paulo) — dedicated enterprise sales & market prospecting partner with on-the-ground Brazil network	Native	None	None	None	Strong	Strong	Strong	None

Norfolk AI's Unfair Advantage

Every competitor in this table is good at one thing. Blip and Weni by VTEX know LATAM messaging. Odisseia AI knows pre-sales qualification. PolyAI knows voice. Sierra knows agentic AI. Regional BPOs know operations. Norfolk AI is the only provider that deploys cross-functional agent swarmsA coordinated group of AI agents that handles multiple business functions simultaneously — CX, technical support, customer operations, and sales — within a single operational workforce. Unlike single-purpose bots, cross-functional swarms share context across functions, resolve handoffs internally, and scale elastically across all channels without siloed tooling. (one workforce for CX, support, and service), across all channels (Voice, WhatsApp, and Chat unified), with agentic reasoningGoal-directed problem-solving by an AI agent — not scripted responses, but dynamic planning. The agent observes context, reasons about the best path to a defined goal, takes action, evaluates the outcome, and updates its approach. This enables agents to handle novel situations that fall within their guardrail boundaries without explicit programming for every scenario. (goal-directed, not scripted), and locally operated in LATAM — all at once, as a single managed service. That is not a feature list. That is a structural advantage that delivers compounding intelligenceThe system-level effect where every human escalation decision is codified into the agent's guardrail library, expanding the resolution envelope. Each interaction makes the next one better — intelligence compounds over time..

Cross-Functional

One agent swarm workforce handles CX, technical support, customer operations, and sales inquiries. No siloed tools, no handoff gaps, no separate vendor contracts.

Cross-Channel

Voice, WhatsApp, and Chat operate as a single unified service. The same agent swarm context, the same CRM write-back, the same service commitment — regardless of channel.

Locally Operated

Norfolk AI's operations team is based in LATAM. Bilingual (Spanish/Portuguese), culturally fluent, and accountable to Cirion under a LATAM-governed service agreement.

Local Sales Presence

SM Services — Norfolk AI's São Paulo-based enterprise sales partner — provides dedicated market prospecting, account qualification, and pipeline management across Brazil and Latin America. No competitor in this table has a named, specialist local sales partner operating in Brazil.

Section 08

Next Steps

How Cirion and Norfolk AI move from proposal to live conversations

The Ask

Start with a Discovery Call

A 45-minute working session to map Cirion's highest-value workflows, confirm CRM integration scope, and define which agent swarms to deploy first. No obligation — just a clear picture of what's possible.

Discovery Call

45-minute working session to map workflows, confirm CRM scope, and align on success metrics. We come prepared — you leave with a clear picture.

Deployment Scoping

Norfolk AI produces a detailed deployment scope: agent swarm design, integration requirements, workflow configuration, and a 30-day readiness plan. We define which swarms to spin up first and how they scale.

Foundation Launch

Deploy a single-channel, single-workflow agent swarm on the Foundation tier. Live conversations in 6–12 weeks depending on integration complexity. Measure, learn, expand the resolution envelopeThe boundary of what an agent swarm can resolve autonomously. Every escalation that gets codified into the guardrail library expands this envelope — the swarm handles more, humans handle less., and scale.

Schedule Discovery Call

[email protected] · norfolk.ai · Confidential — Prepared for Cirion Technologies

Appendix

AI Agent Factory

Technical reference — architecture, agent taxonomy, pipeline, and integration details

The Norfolk AI Agent Factory — five-stage orchestration pipeline from discovery to continuous optimisation

Appendix

What Is the Agent Factory?

The Norfolk AI Agent FactoryNorfolk AI's production environment for building, deploying, and operating AI agent swarms at enterprise scale. Not a SaaS platform — a managed infrastructure layer with full operational ownership. is a production environment for building, deploying, and operating AI agent swarms at enterprise scale. It is not a scripted-bot builder or a no-code drag-and-drop platform. It is a managed infrastructure layer — running on hyperscaler compute (Google Cloud, Microsoft Azure) — that combines agent orchestration, multi-channel deployment, CRM integration, and continuous optimisation into a single operational service.

The Factory treats AI agents as operational roles — not software modules. Each agent swarm is trained with company knowledge, behavioral guardrails, workflow logic, and CRM access. Swarms scale elastically: we spin up more agents during peak volume, fewer during quiet periods, and specialised swarms for specific task types and intensity levels. The result is an AI workforce that can complete tasks end-to-end, escalate intelligently, and improve over time without requiring internal engineering resources from Cirion. For engagements requiring deep custom integration work — complex CRM configurations, multi-system data pipelines, or bespoke API orchestration — Norfolk AI partners with 10.123 Ventures, a specialist AI deep integration firm that works alongside our in-house team on the most technically demanding projects.

The Factory's design principle is guardrail-bounded autonomyAgents operate freely within defined policy, compliance, and behavioral guardrails. When an interaction exceeds those boundaries, the agent pauses and escalates with full context. The human decision is then codified back into the guardrail library., not human replacement. Agent swarms handle volume — the repeatable, high-frequency interactions that consume human capacity without requiring human judgment. When an interaction requires nuance, authority, or empathy beyond the agent's defined guardrails, it routes to a human with full context: the conversation history, the CRM record, the agent's reasoning, and a recommended next action. That human decision is then codified into the guardrail library, expanding the resolution envelope for next time. Cirion's human team shifts from answering the same question 200 times a day to managing the exceptions, building relationships, and making the decisions that actually require a person. The result is operating leverage with compounding intelligence: more interactions handled, higher quality on the ones that matter, and a system that gets smarter with every escalation.

Appendix

AI Agent Taxonomy

The Agent Factory deploys six distinct agent swarm roles, each designed for a specific operational function. Unlike scripted bots, these agent swarms are goal-directed — they pursue defined outcomes through multi-step reasoning, tool calling, and CRM write-back. Each swarm scales independently based on interaction volume and task intensity.

Agent Role	Primary Function	Key Capabilities	Engagement Channel
Inbound Customer Operations Agent	Handles inbound service calls and queries autonomously	Account lookup, FAQ resolution, ticket creation, escalation routing	Voice, Chat
Technical Support Agent (Tier-1)	Tier-1 troubleshooting for colocation, hyperscaler connectivity, and enterprise broadband services	Diagnostic scripts, knowledge base retrieval, ticket creation with full context	Voice, Chat
Proactive Notification Agent	Sends service alerts and maintenance notifications	Outage alerts, maintenance windows, service restoration updates, opt-in management	WhatsApp
Onboarding & Activation Agent	Guides customers through service activation	Milestone check-ins, document collection, activation confirmation, escalation to CSM	WhatsApp, Chat
Account Management Agent	Proactive account health and renewal management	Usage monitoring, contract renewal reminders, upsell identification, CSM alerts	WhatsApp
Inbound Sales Agent	Handles inbound product enquiries and qualification	Product guidance, pricing information, meeting booking, lead logging	Voice, Chat

Appendix

Agentic Capability Reference

The six capability layers that distinguish agentic AI from traditional conversational AI. Each layer is a production-grade component of the Agent Factory — not a roadmap item. Together, they enable guardrail-bounded autonomy at scale.

Context & Memory

Short-term + Long-term

Agents maintain full conversation context across sessions — customer history, prior interactions, CRM data, and open tickets are all available in real time via short-term memory (session) and long-term memory (vector database). Retrieval-Augmented Generation (RAG) grounds responses in verified enterprise knowledge.

Action Execution

Tool Calling + CRM Write-back

Agents don't just respond — they act. Book appointments, create tickets, update CRM records, send follow-up messages, and trigger workflows autonomously via function calling and API integration. Every action is logged for auditability.

Proactive Outreach

Event-Driven Architecture

Agents initiate conversations based on system triggers — service outages, maintenance windows, onboarding milestones, and renewal dates. This is proactivity at scale: the agent notifies the customer before they need to call in.

Human-in-the-Loop

HITL + Governance

Agent swarms work best when humans stay in the loop for the decisions that matter. Every agent operates within guardrail-bounded autonomy — escalation triggers, compliance boundaries, and audit logs. When an interaction falls outside the agent's defined envelope, it pauses and routes to a human with full context. That human decision is then codified into the guardrail library: the resolution envelope expands, and the same edge case is handled autonomously next time. Human expertise becomes compounding intelligence.

Knowledge Retrieval

RAG + Grounded Knowledge Base

Agents retrieve answers from a grounded knowledge base — FAQs, product documentation, pricing, and policy documents — using semantic search over embeddings. Hallucinations are mitigated by grounding every response in verified sources.

Observability

Telemetry + Audit Logs

Full observability into agent behavior, decisions, and outcomes. Every conversation is logged with telemetry data — intent classification, tool calls made, escalation triggers, and resolution status. Explainability reports are available for compliance review.

Appendix

The Five-Stage Agent Orchestration Pipeline

The Agent Factory operates as a structured production pipeline — from business discovery to continuous optimisation. Each stage builds on the last, ensuring agent swarms are not just deployed but continuously improved.

Business Process Discovery

Before building an agent, the system captures conversation goals, business rules, knowledge base content, and integration requirements. Inputs include FAQs, CRM schema, escalation rules, and sales qualification criteria. This step defines the agent's mission and behavioral guardrails — without it, you're deploying a very expensive autocomplete.

Agent Design & Knowledge Base Configuration

The agent is configured with conversation logic (how it speaks and interacts), a grounded knowledge base (documents and operational instructions), and tools — the APIs it can call, including CRM systems, scheduling platforms, ticketing systems, and databases. Prompt engineering and system prompts define the agent's role, tone, and policy boundaries.

Multi-Channel Deployment

The same agent operates across inbound phone calls, voice widgets on websites, WhatsApp Business API, SMS, email, chat widgets, and internal Slack channels. Architecturally, this is possible because the agent orchestration framework routes messages across communication platforms through a central gateway — one agent, many engagement channels, zero duplication of effort.

Runtime Orchestration & State Management

Once deployed, the agent runs continuously. A runtime engine manages conversation memory, reasoning, tool calling, and escalation paths. The agentic loop — Observe → Reason → Act → Evaluate → Update Memory — allows agents to operate autonomously while maintaining control and accountability. Session management and state management ensure continuity across multi-turn conversations.

Monitoring, Telemetry & Continuous Optimisation

The factory collects call transcripts, conversion metrics, conversation analytics, and operational KPIs. Observability dashboards surface autonomous resolution rates, escalation rates, and AHT. These insights drive prompt refinement, knowledge base updates, and workflow expansion. The system becomes self-improving over time — each interaction makes the next one better.

Appendix

How Agent Swarms Work

Every Norfolk AI agent swarm operates on a continuous agentic loopThe continuous cycle every agent follows: Observe → Reason → Act → Evaluate → Update Memory. This closed-loop system enables goal-directed behavior and continuous improvement.: Observe → Reason → Act → Evaluate → Update Memory. This is not a linear pipeline — it is a closed-loop system where each interaction makes the next one better. Swarms scale elastically: we spin up more agents during peak volume, fewer during quiet periods, and specialised swarms for specific task types and intensity levels.

The Agentic Loop

Observe

→

Reason

→

Act

→

Evaluate

→

Update
Memory

↺Continuous loop — every interaction expands the resolution envelope

Elastic Swarm Scaling

↕1–5×

Off-Peak Hours

Baseline swarm

↕5–20×

Normal Volume

Standard operations

↕20–50×

Peak Demand

Outage / campaign surge

↑Specialised

Task-Specific Swarms

Sales, support, onboarding

Observe: The agent ingests the customer's input — voice, text, or event trigger — alongside full context: CRM record, conversation history, open tickets, and account status. RAG retrieval pulls relevant knowledge base articles. The agent sees everything a human agent would see, in real time.

Reason: Multi-step reasoning determines intent, identifies the optimal resolution path, and selects which tools to call. The agent evaluates multiple hypotheses, checks guardrail boundaries, and decides whether to handle the interaction directly or prepare a context-rich handoff to a human specialist. This is not pattern matching — it is goal-directed planning.

Act: The agent executes: creates a ticket, updates a CRM field, books an appointment, sends a WhatsApp notification, or resolves the query with a grounded response. Every action is logged with full audit trail. Tool calling is bidirectional — the agent both reads from and writes to enterprise systems.

Evaluate: Post-interaction analysis scores the outcome: was the customer's intent resolved? Did the agent stay within guardrails? Was escalation appropriate? Telemetry data feeds back into the optimisation layer — prompt refinement, knowledge base updates, and guardrail adjustments.

Update Memory: The interaction is codified into the agent’s memory: short-term (session context for multi-turn conversations) and long-term (vector database for knowledge retrieval). Edge cases that triggered escalation are reviewed and, where appropriate, codified into the guardrail library — expanding the resolution envelope. The swarm gets smarter with every interaction. This is compounding intelligence.

Appendix

Technical Architecture

A Norfolk AI deployment is a seven-layer stack running on hyperscaler infrastructure (Google Cloud, Microsoft Azure). Each layer has a distinct responsibility — from receiving the customer's first word to logging the last data point. The architecture is designed for enterprise-grade reliability, with low-latency audio pipelines, bidirectional CRM integration, and full audit log coverage. Agent swarms scale elastically across the stack based on real-time demand.

CUSTOMER ENGAGEMENT CHANNELS

Inbound Phone

Website Voice Widget

WhatsApp Business API

SMS

Chat / Email

↓

CHANNEL GATEWAY LAYER

Telephony / Messaging APIs

WhatsApp Business API

WebSocket Streaming

Event Routing

↓

VOICE + CONVERSATION INFRASTRUCTURE

Speech-to-Text (STT)

Text-to-Speech (TTS)

Streaming Conversation Engine

Low-Latency Audio Pipeline

↓

AGENT RUNTIME LAYER

Agent Reasoning Engine

Tool Calling Framework

Memory & Context Manager

Observe → Reason → Act → Learn

↓

AI INTELLIGENCE LAYER

LLM Reasoning (GPT / Claude / Gemini)

Retrieval-Augmented Generation (RAG)

Grounded Knowledge Base

Behavioral Guardrails + Policy

↓

ENTERPRISE TOOL LAYER

Salesforce / HubSpot CRM

Calendar & Scheduling

Billing / Ticketing / Provisioning APIs

Docs / FAQs / Knowledge Bases

↓

AGENT FACTORY CONTROL PLANE + OBSERVABILITY

Agent Design Studio

Prompt & Workflow Management

Deployment Orchestration

Conversation Transcripts

Agent Performance Dashboards

Audit Logs + Compliance Reports

Appendix

SaaS Platform vs. Agent Factory Model

The fundamental question is not which AI vendor has the best technology. It is which operating model delivers compounding intelligenceThe system-level effect where every human escalation decision is codified into the agent's guardrail library, expanding the resolution envelope. Each interaction makes the next one better — intelligence compounds over time.. SaaS platforms sell you access to hyperscaler compute and expect you to build, staff, and maintain the AI yourself. The Agent Factory is a managed service: we own the agents, operate them, and continuously expand the resolution envelopeThe boundary of what an agent swarm can resolve autonomously. Every escalation that gets codified into the guardrail library expands this envelope — the swarm handles more, humans handle less.. The table below maps the structural differences.

Dimension	SaaS AI Platform	Norfolk AI Agent Factory
What you buy	Software licence + access	Managed AI agent workforce — we own and operate the agents
Who builds the agents	Your team (with vendor tools)	Norfolk AI team
Who operates day-to-day	Your team	Norfolk AI team
Who optimises over time	Your team (or not at all)	Norfolk AI — continuous
Knowledge base maintenance	Your responsibility	Norfolk AI — monthly updates
CRM integration	Your team configures	Norfolk AI builds and maintains
Time-to-first-conversation	Months (build + QA + train)	6–12 weeks
What happens when it breaks	You debug it	Norfolk AI fixes it (operationally accountable)
What happens at renewal	Licence fee + your ops cost	Performance review + roadmap
Human-AI collaboration model	Customer builds and staffs oversight	Norfolk AI manages AI + human handoff layer
Investment trajectory	Depreciates without internal effort	Compounding intelligence — improves with every interaction

Appendix

Five Use Cases Across Cirion's Service Portfolio

Cirion Technologies operates hyperscaler-grade infrastructure across Latin America — 105,000+ km of fiber, 840+ backbone PoPs, 110+ metro access zones, and 5,500+ enterprise and carrier customers across 15+ countries in Latin America.^[1] That footprint generates a predictable set of high-frequency, high-value customer interactions: service incidents, maintenance notifications, onboarding milestones, account renewals, and inbound product enquiries. The five use cases below map the Agent Factory directly to those workflows — each grounded in Cirion's actual service portfolio, each operating within guardrail-bounded autonomy with full CRM write-back and compounding intelligence.

Use Case 1 · Voice + Chat

Inbound Tier-1 Support — Colocation & Hyperscaler Connectivity

Voice + Chat

Scenario: An enterprise customer calls Cirion's support line about degraded throughput on a cross-connect between their colocation rack and a hyperscaler on-ramp — Azure ExpressRoute or Google Cloud Interconnect. Without an AI agent, this call joins a queue, waits for a human to manually look up the account, and takes 8–12 minutes to triage. With the Agent Factory, the same interaction resolves in under 3 minutes — or escalates to Tier-2 with zero repeat questioning.

Instant answer, no hold queue

The agent answers within one ring. Caller ID or account number retrieves the customer's colocation contract, active services, and open NOC tickets in real time. The customer is greeted by name. The agent already knows which facility, which rack, and which circuit before the customer finishes their first sentence.

Network monitoring API query

The agent queries Cirion's network monitoring API for active incidents on the customer's circuit. It determines whether the degradation is a known infrastructure event (NOC-tracked) or an isolated fault — two very different resolution paths that a scripted bot cannot distinguish.

Resolution or context-rich escalation

Known event: agent confirms the NOC's ETA, creates a ticket with the incident reference, and sends a WhatsApp update to the customer's primary contact. Isolated fault: agent escalates to Tier-2 with the full handoff packet — conversation history, CRM record, circuit details, and the agent's diagnostic reasoning. No repeat questioning.

Post-resolution CSAT loop

24 hours after resolution, the WhatsApp agent sends a satisfaction check. A negative response flags the CSM for proactive outreach. CSAT score is logged to CRM automatically. The interaction is complete — and the next similar incident is handled faster because this one was codified.

Edge case codification

If the agent encountered a scenario outside its guardrails — an unusual circuit configuration, a non-standard escalation path — that edge case is reviewed and fed back into the guardrail library. The resolution envelope expands. The same situation is handled autonomously next time.

Result

Tier-1 containment rate of 40–60% for inbound service calls. AHT for escalated calls reduced ~25% from full context transfer on handoff — no repeat questioning, no manual lookup. Every escalation is an edge case codification event: the same situation is handled autonomously next time. [Source: BCG, 2025]

Use Case 2 · WhatsApp

Proactive Maintenance Notifications — Eliminating the Inbound Spike

Scenario: Cirion's NOC schedules a planned maintenance window affecting enterprise broadband circuits across a metro region. Without proactive notification, the same 200 customers call the support line in the same 30-minute window — a predictable inbound spike that overwhelms the human operations team and generates CSAT scores that have nothing to do with the quality of the maintenance work. The problem is not the maintenance. The problem is the silence before it.

NOC event triggers agent swarm

Maintenance window confirmed in the NOC system. The event triggers the Proactive Notification Agent with the affected circuit list, customer account IDs, maintenance window, and expected restoration time. The agent does not wait for customers to call — it reaches them first.

Personalised WhatsApp dispatch

Agent sends personalised WhatsApp messages to each affected account's primary contact — in Spanish or Portuguese, using their name, their circuit reference, and the exact maintenance window. WhatsApp penetration exceeds 90% in key Latin American markets including Brazil and Chile.[2] This is not a generic broadcast — it is a one-to-one conversation at scale.

Two-way conversation handling

Customers reply with questions. The agent handles common responses autonomously: 'Will this affect our backup circuit?' (account lookup + circuit check), 'Can we reschedule?' (escalate to NOC with context), 'Thanks, noted' (log acknowledgement to CRM). Customers who need a human get one — with full context already transferred.

Restoration confirmation

When the maintenance window closes and circuits are restored, the agent sends a confirmation message. Customers who did not acknowledge the initial notification receive a follow-up. Unresolved issues are flagged to the support queue with the full conversation thread.

Inbound call deflection measured

Customers who already know what is happening do not call. The support queue does not spike. Human agents handle only genuinely novel issues — not the same question from 200 customers who were never told. The deflection rate is measurable: compare inbound call volume during maintenance windows before and after deployment.

Result

Inbound call volume during maintenance windows reduced by 40–60%. Customer satisfaction improves because customers are informed, not surprised. Support agents handle novel issues, not a queue of identical calls. The agent's proactive outreach is the product — silence is the bug it fixes. [Source: BCG, 2025]

Use Case 3 · WhatsApp + Chat

New Customer Onboarding — Guided Activation for Colocation & Cloud Services

WhatsApp + Chat

Scenario: A new enterprise customer has signed a colocation contract with Cirion. The onboarding process involves document collection, facility access provisioning, cross-connect ordering, and milestone confirmations across a 4–6 week timeline. Without an agent, this process lives in email threads and CSM calendars — easy to drop, hard to track, and invisible to the customer until something goes wrong.

Contract signed triggers onboarding agent

CRM event fires when the contract is marked closed-won. The Onboarding & Activation Agent initiates a WhatsApp conversation with the customer's primary contact — introducing itself, confirming the account details, and outlining the activation timeline. The customer knows what to expect before they have to ask.

Document collection

Agent requests required documents via WhatsApp — facility access forms, technical specifications, authorised contact lists. Customers upload directly in the conversation. Documents are logged to the CRM record and flagged for CSM review. No email attachments, no lost threads.

Milestone check-ins

At each activation milestone — facility access granted, cross-connect ordered, test traffic confirmed — the agent sends a status update and asks if the customer has questions. The CSM is notified only when the agent encounters a question outside its guardrails or a milestone is delayed.

Activation confirmation

When the service is live, the agent sends a confirmation message with the customer's circuit references, support contact details, and a link to the customer portal. First-use guidance is delivered in the same conversation — no separate welcome email that gets buried.

30-day health check

30 days after activation, the agent initiates a check-in: Is the service performing as expected? Any open questions? The response is logged to CRM. A negative signal flags the CSM for proactive outreach before the customer considers escalating or churning.

Result

Onboarding completion rate improves because every milestone is tracked and every delay is visible. CSM capacity is freed from status-check conversations — they focus on relationship management and upsell identification. The customer's first 30 days are a product experience, not an administrative process.

Use Case 4 · Voice + WhatsApp

Account Renewal & Upsell Identification — Proactive Account Management

Voice + WhatsApp

Scenario: Cirion manages contracts with 5,500+ enterprise and carrier customers across Latin America.[1] Contract renewals are predictable events — but without a systematic process, they surface only when a CSM happens to check a spreadsheet or when the customer calls to cancel. S&P Global downgraded Cirion to 'B-' in February 2026, citing weaker-than-expected results and a revised 2025 forecast.[3] Churn is rarely unexpected — it is usually unnoticed.

Renewal trigger — 90 days out

The Account Management Agent monitors contract end dates in the CRM. 90 days before expiry, it initiates a WhatsApp conversation with the account's primary contact — acknowledging the upcoming renewal, confirming service satisfaction, and asking whether their requirements have changed.

Usage analysis

The agent queries billing and usage data to identify whether the customer is at capacity on their current tier, underutilising services, or showing usage patterns consistent with expansion. This analysis informs the conversation — the agent does not pitch blindly.

Renewal conversation

For straightforward renewals, the agent handles the conversation autonomously: confirms terms, answers standard questions, and logs the customer's intent to the CRM. For complex renewals — pricing negotiations, service changes, multi-year commitments — the agent prepares a context-rich handoff to the CSM.

Upsell identification

If usage analysis identifies expansion signals — bandwidth approaching capacity, new locations added, increased cloud traffic — the agent flags the account for CSM review with a recommended upsell action. The CSM receives the recommendation with the supporting data already assembled.

Churn signal detection

If the customer expresses dissatisfaction, requests a contract review, or goes silent on renewal outreach, the agent escalates immediately to the CSM with the full conversation history and account context. Early detection is the entire value: the CSM intervenes before the customer has already decided to leave.

Result

Renewal conversations start 90 days earlier. Churn signals are detected before they become decisions. CSMs focus on the accounts that need human judgment — not on chasing renewals that an agent could have handled. The system compounds: every renewal conversation improves the agent's ability to identify the signals that matter.

Use Case 5 · Voice + Chat

Inbound Product Enquiries — Qualification & Meeting Booking

Voice + Chat

Scenario: A prospective customer calls or chats to enquire about Cirion's colocation or SD-WAN services. Without an AI agent, this call either goes to a general support queue (wrong destination), waits for a sales rep to become available (lost momentum), or gets routed to voicemail (lost lead). The Agent Factory handles the qualification conversation, answers product questions from a grounded knowledge base, and books a discovery call with the right sales contact — all without human involvement.

Inbound enquiry captured

The Inbound Sales Agent answers the call or chat. It identifies the enquiry type — new service, pricing, technical specifications — and confirms whether the caller is an existing customer or a new prospect. Existing customers are routed to the appropriate support workflow. New prospects enter the qualification flow.

Qualification conversation

The agent asks structured qualification questions: company size, location, current infrastructure, primary use case (colocation, cloud connectivity, SD-WAN, enterprise broadband). Responses are logged to the CRM as a new lead record in real time — no manual data entry after the call.

Product guidance from grounded knowledge base

The agent answers product questions — pricing tiers, data center locations, connectivity options, SLA frameworks — by retrieving answers from Cirion's grounded knowledge base via RAG. Responses are accurate and consistent. The agent does not improvise outside verified content.

Meeting booking

When the prospect is qualified and interested, the agent books a discovery call directly in the sales rep's calendar — checking availability, confirming the time, and sending a calendar invitation. The sales rep receives the meeting with the full qualification summary already attached.

Lead handoff packet

The sales rep inherits the complete handoff packet: the conversation transcript, the qualification data, the product questions asked, and the agent's assessment of the prospect's primary use case. The discovery call starts at a higher level of context than a cold introduction.

Result

Inbound leads are captured and qualified 24/7 — not just during business hours. Sales reps receive qualified meetings, not cold introductions. The agent handles the volume; the sales team handles the judgment. First contact resolution for enquiries improves by ~14% with AI-assisted qualification. [Source: McKinsey, 2024]

Appendix

Integration Architecture

Norfolk AI agents integrate bidirectionally with Cirion's systems of record — reading context, writing outcomes, and triggering workflows in real time. The integration scope below covers the standard Enterprise tier configuration.

System	Integration Type	Data Flow	Use Case
HubSpot CRM	Native API	Bidirectional	Lead management, deal pipeline, contact records
Salesforce / Agentforce	REST API + Webhooks	Bidirectional	Opportunity management, case creation, account data
Calendar (Google / Outlook)	OAuth + API	Read/Write	Meeting booking, availability check, confirmation
Billing / ERP	REST API	Read	Account status, payment history, service tier
Ticketing (Freshdesk / ServiceNow)	REST API	Write	Ticket creation, status updates, escalation routing
Knowledge Base	RAG / Vector Database	Read	FAQ resolution, product info, policy lookup
WhatsApp Business API	WhatsApp Cloud API	Bidirectional	Template messaging, session management, opt-in flows

Appendix

A Day in the Life

24 hours with Gaspar — two tiers, two different operating realities

FoundationGrowth

Gaspar

Norfolk AI Agent

Background

B.Eng. Electrical Engineering, USP

M.A. Philosophy, USP

Ph.D. Data Analytics, Unicamp

Languages

Portuguese · Spanish · English

Availability

24/7/365 · No holidays

Tier

Foundation — 1 workflow

Channels

Voice · WhatsApp

Meet the Agent

Gaspar isn't a chatbot.
He's an operational role.

Gaspar's academic profile is not decorative. A background in electrical engineering from USP gives him a systems-level understanding of the infrastructure Cirion's customers rely on — circuits, connectivity, fault propagation. His master's in philosophy sharpened something rarer: the ability to reason under ambiguity, to know when a situation requires a judgment call versus a lookup, and to communicate that distinction clearly to a frustrated customer at 2 AM.

His doctorate in data analytics from Unicamp is what makes him improve. Every interaction Gaspar handles — resolved or escalated — is reviewed by Norfolk AI's in-house team and fed back into his operating model. He doesn't just answer questions. He learns which answers worked, which phrasings landed, and which edge cases need to be codified so he handles them autonomously next time.

Gaspar is deployed on Cirion's inbound customer operations workflow under the Foundation tier: one agent, one workflow, standard knowledge base and CRM integration. The scenarios below reflect that scope — no autonomous billing resolution, no complex technical decisions. Just consistent, documented, always-available first-line support.

Foundation Tier · Inbound Customer Operations

Gaspar's Day — All 24 Hours of It

Gaspar is deployed on Cirion's inbound customer operations workflow. He handles Voice and WhatsApp inbound queries, opens and routes tickets, and provides 24/7 coverage — including nights, weekends, and public holidays. He doesn't replace the support team. He makes sure nothing falls through the cracks when they're not there, and handles the routine volume so they can focus on what actually requires a person.