One successful customer-service AI product answers “can we build something useful?” The enterprise question is whether MonGo can scale AI across products and functions without duplicated platforms, inconsistent controls, uncontrolled cost or fragmented ownership.

This article is Part VIII of the Banking Customer-Service AI playbook. It follows Part I through Part VII.

---

1. Purpose of this phase

Customer-service AI results to date:

Measure	Baseline	Current
Knowledge-search time	95 sec	31 sec
Average handling time	8.4 min	7.4 min
First-contact resolution	68%	76%
Customer satisfaction	71%	80%
Agent adoption	0%	82%
Critical policy-error rate	2.8%	0.4%
Successful safe resolution	—	72%

Not: deploy AI everywhere.
Rather: create an enterprise capability that consistently selects, delivers, governs and operates the right AI solutions.

---

2. Pressure to scale

Retail, commercial, financial crime, risk/compliance, technology, HR and internal audit all propose assistants, summarisation, investigation support and drafting. Without portfolio management: multiple model contracts, duplicate retrieval, inconsistent security, repeated evaluation, competing pipelines, unclear ownership, uncontrolled cloud cost, divergent RAI interpretations and incompatible architectures.

---

3. AI Portfolio Management

Four levels: strategic themes (trusted digital service, employee productivity, risk/compliance transformation, intelligent ops, data-driven engagement) → value streams → AI products (e.g. Intelligent Customer Service, Financial Crime Investigator Assistant) → capabilities/use cases.

Portfolio register: name, theme, owner, users, benefit, risk class, regulatory significance, data, tech dependencies, reusable capabilities, stage, funding, performance, next decision date.

Example: Financial Crime Investigation Assistant — high risk, recommendation only; proceed to discovery on existing platform (case data, ingestion, ACL, evaluation, oversight, model gateway).

---

4. Portfolio segmentation

Segment	Examples
Business products	CS assistant, RM copilot, employee knowledge
Shared platforms	Gateway, retrieval, evaluation, observability, registries
Data and knowledge	Data products, ingestion, metadata, lineage, quality
Governance and assurance	Inventory, risk workflow, validation, red team, continuous assurance
Research and innovation	Multimodal, autonomous orchestration, PETs, on-device, synthetic data

Platform and governance investments are not judged only on short-term revenue.

---

5. Lean Portfolio Management

Investment funding (example): products 45% · shared platform 20% · data/knowledge 15% · governance 10% · research 10%. Evaluation platforms are enterprise assets—not funded entirely by the first product.

Portfolio Kanban: Funnel → Reviewing → Analysing → Ready for discovery → Discovery → Ready for experiment → Experiment → Pilot → Production → Scaling → Measuring → Retired.

WIP limits: 5 concurrent discoveries · 3 major experiments · 2 new customer-facing pilots · 1 high-risk agentic pilot.

Lean governance: evidence proportionate to risk, scale, autonomy, impact, investment and reversibility—but every production system still needs owner, inventory, purpose, evaluation, monitoring and incident management.

---

6. Portfolio prioritisation

Score strategy, value, feasibility, reuse and risk-adjusted position.

Use case	Risk-adjusted score	Decision
Employee knowledge assistant	4.6	Proceed
IT incident copilot	4.1	Proceed
Financial-crime case summary	3.9	Proceed
Marketing-content generator	3.0	Lower priority
Credit-decision AI	2.4	Defer
Autonomous account agent	2.0	Defer

High theoretical value does not win when risk is high and readiness is weak.

---

7–8. Three Horizons and core/adjacent/transformational

Horizon	Focus	Allocation
H1 (0–18m)	Lower autonomy, productivity, reusable foundations	60%
H2 (18–36m)	Value-stream transformation, controlled actions	30%
H3 (36m+)	Options, regulatory dependence, emerging tech	10%

Innovation mix: Core 70% · Adjacent 20% · Transformational 10%.

---

9–11. Enterprise operating model, CoE, Centre for Enablement

Central: strategy, platform, model contracts, AI security, RAI policy, evaluation standards, inventory, observability, FinOps, capability.
Federated: outcomes, discovery, domain knowledge, backlogs, adoption, benefits, process change.
Independent: risk policy, regulation, model-risk challenge, privacy, audit, control testing.

CoE: advisory, engineering reference, governance templates, capability—avoid becoming a bureaucracy, demo-only team, permanent BU owner or platform without customers.

Centre for Enablement: templates, golden paths, docs, coaching, office hours, design reviews, patterns, communities. Commercial Banking builds an RM assistant using RAG template, gateway, eval guidance, System Card and threat-model templates—while remaining accountable for outcomes, knowledge, adoption, benefits and service ownership.

---

12–14. AI Factory and platform product

Factory: intake → discovery → data/knowledge → architecture → engineering → evaluation → governance → pilot → production → operations → benefits → retirement.

Measures: idea→discovery and discovery→pilot time; % experiments stopped early (can be healthy); % pilots to production; reuse; quality; benefits; incidents.

Platform product: model access, knowledge/retrieval, agent infrastructure, evaluation, operations, governance integrations. Users = internal product teams. Success: production products on platform, time to prototype/production, reuse, developer satisfaction, control automation, cost, availability.

Before/after: model access 6 weeks → 2 days; standard tracing and eval templates by default.

---

15–16. Reusable patterns

Pattern	Use	Key components
Knowledge assistant	Policy/tech support	Hybrid retrieval, citations, escalation
Document processing	Contracts, apps, regs	Ingestion, OCR, extract, validate, human review
Agent-assist	CS, crime, audit, commercial	Context, recommendations, sources, override
Controlled-action agent	Reversible low-risk	Auth, scoped tools, confirmation, trajectory, rollback
Decision-support	Risk/investigation	Evidence, structured recs, uncertainty, human decision

Governance templates: purpose, prohibitions, cards, impact, DPIA inputs, threat model, oversight, PRR, incident playbook, decommissioning checklist.

---

17–20. Capability roadmap and workforce

Capability	Current → Target
AI PM / engineering / evaluation / ops / RAI / KM / benefits	3 → 5
Agent engineering / FinOps	2 → 4
AI security	3 → 5

Year 1 foundations · Year 2 scale across BUs · Year 3 optimise and transform.

Workforce: reduce search/drafting/summarisation; increase complex resolution, vulnerability support, AI review; create feedback, curation, incident reporting, evaluation, workflow optimisation. Role: AI-Enabled Customer Resolution Specialist.

Skills taxonomy: executive · product · engineering · risk · business. Build / buy / partner / borrow for ownership, platform talent, acceleration/assurance and secondments.

---

21–24. Vendors, concentration, ecosystem

Strategic / important / tactical vendors. Scorecard: reliability, security, privacy, model quality, cost predictability, regulatory support, portability, support, innovation.

If four major products share one foundation-model provider: secondary provider, gateway, task fallbacks, deterministic alternatives, service prioritisation, contractual resilience, failover tests. Outage priority: (1) safety/fraud (2) CS assist (3) employee knowledge (4) developer tools.

Ecosystem: cloud, universities, fintech, regtech, assurance, consortia, OSS—with data protection, IP clarity, risk ownership, portability and internal accountability.

---

25–26. Enterprise FinOps

Inform → Optimise → Operate. Visibility by product, BU, model, use case, environment, interaction type.

Product	Monthly cost	Outcomes	Cost / outcome
Customer Service AI	$310k	165k	$1.88
Employee Knowledge AI	$140k	220k	$0.64
IT Operations Copilot	$95k	18k	$5.28
Financial Crime Assistant	$180k	12k	$15.00

Outcomes differ—do not rank blindly. Employee knowledge routing (60% retrieval+summary, 25% small model, 15% large): $180k→$112k/month ($816k/year). Anomaly $9k→$31k/day: stop loop, step limits, circuit breakers. Start with showback before chargeback.

---

27–30. Benefits portfolio and performance

Avoid double-counting the same released capacity as both “productivity” and “avoided hiring.” Categories: revenue, cash, avoidance, capacity, risk reduction, CX, EX, strategic capability.

Confidence needs evidence, causality, owner commitment, measurement, timing, dependency completion—do not mark capacity as realised until converted.

Balanced scorecard: financial · customer · internal ops · learning/capability · risk/trust.

OKRs: scale trusted products (3 more production; 80% platform reuse; error thresholds; $15M validated benefit) · improve delivery (discovery→pilot 9→4 months; 70% eval automation; 500 trained; 85% platform CSAT) · strengthen governance (100% inventory; quarterly medium/high reviews; kill-switch tests; critical findings on time).

---

31–33. Scaling readiness, patterns, reuse

Customer-service AI scores well on value/ops/governance; geographic readiness 2.8—scale UK retail first; not internationally until language eval, product rules, local law and regional knowledge ownership are ready.

Horizontal (more users) · vertical (depth/autonomy) · domain · geographic—each needs different evidence.

Reuse shared gateway; replicate only for regulation, performance, residency or justified isolation—not convenience.

---

34–37. Continuous assurance and portfolio risk

Continuous / weekly / monthly / quarterly / annual assurance cadence. Automate evidence chains linking release → model → prompt → index → eval → security → approval → deployment.

Four products sharing one gateway: individually medium availability risk; portfolio-critical → multi-region, secondary gateway, failover tests, critical-service classification.

Dashboard: production systems, high-risk, overdue reviews, shared dependencies, incidents, supplier concentration, assurance findings, cost anomalies, fairness, retirement candidates.

---

38–40. Rationalisation, retirement, debt

Two overlapping policy assistants → consolidate common capability, retain domain permissions; $1.2M annual saving.

Retire Marketing Copy Generator v1: low adoption, duplicate function, high review, weak value—notify, export, disable, archive, delete, revoke, terminate, inventory, redirect, lessons.

Debt types: technical · data · governance · adoption. Allocate 15% of platform capacity to debt reduction.

---

41–42. Enterprise value and transformation roadmap

Product	Annual validated benefit
Intelligent Customer Service	$8.5M
Employee Knowledge Assistant	$4.2M
Technology Incident Copilot	$2.8M
Total	$15.5M

Platform $5.2M + product opex $4.1M → net ~$6.2M/year, plus tracked strategic benefits (faster delivery, reusable controls, skills, less supplier duplication, governance, data quality).

Phases: Establish (0–12m) → Expand (12–24m) → Transform (24–36m) → Optimise (36m+).

---

43–46. Forums, scorecard, portfolio decisions

Forums: Investment Committee · Design Authority · Risk Committee · Product Council · Operations Review.

Scorecard: strategic value · delivery capability · trust · operations · workforce.

Product	Stage	Value	Decision
Customer-Service AI	Scaling	$8.5M	Continue scale
Employee Knowledge AI	Production	$4.2M	Expand
IT Incident Copilot	Pilot	$2.8M	Proceed
Financial-Crime Assistant	Experiment	$6.0M pot.	Controlled pilot
Credit Decision AI	Discovery	$12.0M pot.	Decision support only
Marketing Generator	Production	$0.4M	Review consolidation
Autonomous Account Agent	Research	Unknown	Do not deploy

Stopping is a valid portfolio outcome when alignment, value, risk, ownership, data, adoption, reuse, cost or controls fail.

---

47. Enterprise cycle

Set strategy → manage portfolio → enable delivery → deliver products → govern and assure → operate → realise value → improve or retire.

---

48. Phase outputs

Enterprise portfolio and LPM; CoE/C4E; AI Factory; platform product; patterns; capability and workforce plans; vendor/ecosystem controls; FinOps; benefits portfolio; OKRs; scaling readiness; continuous assurance; rationalisation/retirement; debt registers; transformation roadmap; decision forums; stop criteria.

---

49. Core conclusion

Not: replicate chatbots in every department.

Rather: coordinated strategy, portfolio governance, shared platforms, reusable patterns, foundations, skills, suppliers, FinOps, continuous assurance and measurable value.

Not: “How many AI use cases have we launched?”

Rather:

Does the organisation have a repeatable, governed and economically sustainable ability to identify, deliver, operate and retire AI systems that create measurable value?

Next (final) phase: The Integrated 8D AI Solution Engineering Framework—map all frameworks into Define, Discover, Diagnose, Design, De-risk, Demonstrate, Decide, Deliver, plus selection guides, artefacts, gates, RACI, timeline, consulting workplan, executive deliverables and ConsultAI Lab canvas catalogue.