What is pilot purgatory in banking and insurance?

It is the industry's descriptive name for the gap between pilots that pass demonstrations and workflows that reach production. It is rarely a failure of ambition; it usually reflects governance requirements being reconstructed after the build rather than agreed before it.

What is a single decision fabric in BFSI?

One governed data layer spanning retail, corporate, operations and compliance, with traceable lineage, so every decision draws on the same facts rather than per-pilot extracts.[^3]

Why should compliance be a design partner in AI pilots?

Model risk and AI governance frameworks expect controls across the lifecycle. Designing them in from day one produces evidence as work happens and shortens the path to production.[^2][^4]

Which skills do BFSI operations teams need for scaled AI?

Data literacy, model operations and model oversight in the managers and teams who run the workflows, so oversight lives where the work happens rather than only in specialist labs.

What should a 90-day BFSI integration test measure?

Cycle time, manual touchpoints, exception volume and systems consulted per decision across three named workflows, each with one accountable owner and one productivity metric, compared with a baseline.

Episode 08 — BFSI·8 min read·Updated 26 October 2026

The institutions turning AI pilots into production did not slow their ambitions. They stopped running compliance, data and operations as separate businesses and wired controls into how the work already happens.

How does AI pilot momentum become production?

BFSI, part two: the operating model that turns pilot momentum into production — compliance as a design partner, and every decision with one owner, one fabric and one measure.

Join discussion

The series

An executive briefing on Strategy, Operations, Compliance, Workforce and Logistics. From silos to synergy.

You do not have an operations problem. You have a five-pillar integration problem. Strategy picks it. Operations run it. Compliance governs it. Workforce delivers it. Logistics moves it. These five are bought separately and must be run together.

When the five pillars operate as one system, organisations cut cost, reduce risk, and scale with confidence. When they do not, growth stalls in the gaps between them — and every function can prove the problem sits somewhere else.

Each weekly episode takes one industry at a time — BFSI, Retail, Manufacturing, Energy, Telecom, Technology, Logistics, Public Sector, Real Estate, Healthcare, Life Sciences — using real-world challenges, identifiable company lessons, and ideas you can test inside a quarter.

Strategy

picks it

Operations

run it

Compliance

governs it

Workforce

delivers it

Logistics

moves it

The challenge

What does each new pilot cost while the path to production is unclear?

The first BFSI briefing asked why every audit still feels like an emergency. This second part looks at the growth side: how the strong appetite for digital and AI experimentation becomes workflows in production. The environment asks a lot. Supervisors expect demonstrable control of model risk and data lineage, AI-specific regulation is arriving with enforcement dates, and boards have committed to digital growth targets that depend on exactly these capabilities.¹²

Pilot momentum is an asset. Each new pilot carries learning, options and institutional confidence that the organisation can modernise. What the industry calls pilot purgatory is not a failure of ambition — it is a descriptive name for the gap between pilots that pass demonstrations and workflows that reach production. That gap costs something every time a new pilot starts: risk, compliance and IT reconstruct what the pilot team already knows, and the business buys a second tool to do what the first pilot promised. None of that is a reason to slow down. It is a reason to give the momentum one path.

The five gaps sit in the operating model, not the ambition. Strategy sets growth targets without an integrated operating model to deliver them. Operations run fragmented core systems and vendor stacks, with manual reconciliations and exception handling by email and chat. Compliance treats regulatory reporting, model risk and AI governance as review at the end, not design at the start. Workforce gaps sit in the middle: managers and operations teams are not enabled for new workflows, and skills in data, model operations and model oversight are scarce. Logistics, the technology and data flow, is the deepest opportunity: data held separately across retail, corporate, operations and compliance means no single decision fabric yet exists.

A useful self-diagnosis: of your last ten pilots, how many reached production with a named business owner and a control framework agreed before build? Whatever the number, the constraint is rarely the ambition — it is the operating model around it. That is the part an institution can redesign without pausing a single pilot.

Lessons from the field

What does good look like: one decision fabric, embedded compliance and clear roles?

Three disciplines turn pilot momentum into production.

The first is a single decision fabric. Data principles developed for risk aggregation after the financial crisis remain the benchmark: accuracy, completeness, timeliness and adaptability, with lineage traceable end to end.³ When retail, corporate, operations and compliance data share one governed layer, the expected outcome is faster approvals, because risk and business teams reason from the same facts instead of reconciling competing extracts.

The second is compliance as a design partner. Model risk guidance has expected independent validation and governance from inception for over a decade,⁴ and AI frameworks now ask for the same: map, measure, manage and govern built into the lifecycle rather than audited afterwards.² When control requirements shape the pilot on day one, the expected outcome is shorter time to production, because the evidence the gatekeepers need is produced as the work happens.

The third is role clarity. Every scaled workflow has one accountable owner, one risk approver and one productivity metric. Where ownership is shared across a committee, decisions wait for meetings; where it is named, decisions wait for facts.

Consider a worked scenario, not a client case. An insurer pilots an underwriting triage model. It waits seven months because model risk needs lineage the data team did not capture, and operations discover the exception queue still runs on email. Re-run with a governed data layer, validation criteria agreed at kick-off and a named underwriting owner, and the same model reaches production in one quarter, with the audit trail already written. The first pilot was not wasted: it taught the institution exactly which agreements were missing.

Seen through several lenses: an engineer sees missing interfaces; a strategist sees growth targets funded without the plumbing to meet them; a psychologist sees teams who keep change in the lab because that is the safest option their operating model offers them. Founders and executives fear the headline; middle managers fear owning a number. The defensible move from today's position is one workflow where a pilot crosses into production with its controls already in place — no pilot paused, nothing slowed.

The five-pillar integration play

One move per pillar

Strategy picks it: fund the fabric, not just the pilots

Tie each digital and AI growth target to the operating capability that delivers it: governed data, agreed controls, named owners. If investment cases show the path to production before funding, the expected outcome is more pilots completing their journey, because ambition and delivery mechanism are approved together.

Operations run it: retire reconciliation by hand

Map the manual reconciliations and email-based exception handling that sit under your priority workflows, and replace them with system-enforced handoffs. If exceptions surface in the workflow, the expected outcome is lower operational risk and freed capacity, because people stop being the integration layer.

Compliance governs it: design controls in from day one

Move regulatory reporting, model risk and AI governance colleagues into pilot design from day one, with validation criteria agreed before build. If controls are designed in, the expected outcome is faster, calmer approvals, because the evidence exists as the work happens rather than being assembled afterwards.²⁴

Workforce delivers it: enable the middle, not just the lab

Build data, model operations and model oversight skills in the managers and operations teams who will run scaled workflows, not only in the specialist teams who build them. If the people running the process understand the model's limits, the expected outcome is fewer silent failures, because oversight lives where the work happens.

Logistics moves it: one decision fabric across the clouds

Connect retail, corporate, operations and compliance data into a single governed layer with traceable lineage, instead of extracting per pilot. If every decision draws on the same fabric, the expected outcome is consistent answers across the institution, because there is one version of the facts.³

Who benefits

The gain, by seat at the table

Chief Executive / Board: Digital growth targets with a credible delivery mechanism and a defensible risk story.
Chief Operating Officer: Fewer manual reconciliations and exceptions, with capacity returned to the business.
Chief Risk / Compliance Officer: Controls designed into pilots from day one, with evidence produced as work happens.
Chief Data Officer: One governed decision fabric replacing per-pilot extracts and competing versions of the truth.
Business Line Leaders: Pilots that reach production with a named owner and a metric they control.

Idea worth testing

A 90-day test: three workflows, one owner each, one productivity metric

Name three workflows that matter to one growth target, for example customer onboarding, credit decisioning and claims triage. Existing regulatory obligations and control frameworks remain fully in force throughout.

Days 1–30: baseline. For each workflow, record cycle time, manual touchpoints, exception volume and the number of systems consulted per decision. Note how many pilots are active and how many reached production in the last year. Assign one accountable owner to each workflow.

Days 31–60: agree the control framework before changing anything: what model risk, compliance and data governance need as evidence, captured inside the workflow rather than afterwards. Connect the three workflows to one governed data view, even if narrow.

Days 61–90: run all three workflows end to end with one productivity metric each, such as decisions per person per day, and compare with the baseline. These are test measures, not promised gains. If cycle time improves but exceptions do not, the next constraint is upstream data quality, which becomes the next decision.

The executive question is: which of our current pilots would move faster if the control framework were agreed before the build started?

How Strategy Labs installs this

Phased, anchored, and handed back

Strategy Labs is the advisory artefact: diagnostics, benchmarks and recommendations you can see before you buy. Here the diagnostic is the pilot-to-production question; the benchmark is your three-workflow baseline; the recommendation is the smallest change that proves a pilot can scale with its controls in place. This digital advisory artefact gives the visitor capability, not just an answer: a practical way to embed best practice into how you make decisions, choose partners and track growth.

Use the Consulting Advisory Engine (CAE), provided by Context Digital Limited, to articulate the problem before funding the next pilot. Separate the symptom (the gap between pilots and production), the decision that lacks an owner and the operating gap behind it. Where consulting is appropriate, the five-pillar map connects decision rights, controls, data and execution so growth, risk and delivery align around one decision fabric.

Pilot momentum is worth protecting: each pilot carries learning. The operating model decides whether that learning compounds into production — and that is a design choice, available today.

Frequently asked

Discussion

(…)

Comments are moderated before appearing. Your email is only used for moderation and is never shown publicly.

Loading discussion…

Continue the series

More episodes

Episode 01 — BFSI