The intelligence shift — how AI is redrawing the BPM and analytics map.

The BPO industry spent two decades selling one proposition: labour cost arbitrage. Move the work somewhere cheaper, keep the quality acceptable, pass on the savings. AI is replacing that proposition with something more demanding — intelligence arbitrage. The ability to embed codified domain knowledge into systems that operate at machine speed, rather than simply moving human effort around the map.

That's not a cosmetic shift. It changes who wins.

I've spent the last several months pulling apart this market shift across four questions that come up constantly in conversations with clients, providers, and people building ventures into this space: how is AI actually changing demand for BPM and analytics services, which end markets are generating real spend versus interesting pilots, what enterprise buyers genuinely prioritise when selecting a transformation partner, and where the whitespace sits for providers willing to go deep rather than broad. What follows is my own read, built from market data, practitioner conversations, and the pattern-matching that comes from sitting on both sides of these mandates for over two decades.

Three waves, one uncomfortable truth

The market is moving through three distinct waves of AI maturity, and most providers are still operating in the first one while pitching the third in their sales decks.

Here's the uncomfortable part. McKinsey's most recent research on agentic AI adoption puts the number of organisations actually scaling an agentic system in at least one business function at roughly 23 percent — meaning the large majority haven't begun enterprise-wide scaling at all.¹ The gap between the ambition in the boardroom and what's actually running in production is enormous, and it is not, in my view, primarily a model quality problem. It's a systems design problem — permissions, observability, data quality, process ownership, governance. The providers who are actually winning the scaling race are the ones who understand how the work flows through an organisation, not just how the underlying models technically function.

Where the spend is actually going

Not every vertical is moving at the same speed, and conflating "AI interest" with "AI budget" is one of the more common — and more expensive — mistakes I see organisations make. Here's my honest sectoral read, with realistic time horizons attached rather than vendor-deck optimism.

Spending now, and likely to sustain for years

Insurance is the standout market by a wide margin. Deployment tracking firm data shows insurance AI deployments growing 87 percent year-over-year, with agentic AI accounting for roughly one in five publicly tracked deployments by the fourth quarter of 2025.³ Commercial P&C insurers running agentic underwriting are reporting quote-to-bind time reductions in the 60 to 99 percent range.⁴ The reason this sector is moving faster than most isn't hype — it's structural. Claims, underwriting, policy administration, and reinsurance reconciliation are all process-dense, data-rich, and directly translatable into combined ratio terms. A CFO can see the return without commissioning a twelve-month attribution study, and that makes budget release considerably faster. Specialty lines — the Lloyd's market, E&S, MGAs — remain particularly under-automated relative to personal lines, and are actively buying.

Banking and capital markets are leading in AML/KYC document review, fraud detection, loan processing, and regulatory reporting. The underserved segment here is mid-market banking — institutions carrying the same compliance burden as the Tier 1 banks but without the in-house AI teams to build their own solutions. A right-sized managed service for that segment is, as far as I can see, still largely unoccupied territory.

A strong three-to-five-year case, with slower near-term conversion

Healthcare and life sciences carry the most compelling long-term thesis — payer-provider friction, prior authorisation, pharmacovigilance, clinical trial operations are all genuinely well-suited to agentic intervention. But HIPAA, FDA data governance requirements, and the realities of health system procurement cycles create meaningful lag. The fact that domain-tuned models are now outperforming general frontier models on healthcare-specific benchmarks is a signal that deep vertical specialisation is buildable here. The total addressable market is enormous. So is the patience required to capture it.

Horizontal business process automation — finance and accounting, procurement, supply chain analytics — is where decades of process intelligence built up in traditional BPO is being redeployed as AI-native capability. Genpact's own Q1 2026 results are the clearest proof point I've seen that this is genuine revenue rather than vendor narrative: their Advanced Technology Solutions segment grew 24 percent year-over-year, now representing 27 percent of total company revenue.⁵ That's not a pilot budget line. That's a business model shift showing up in quarterly earnings.

What has actually escaped pilot purgatory

The honest answer to "what's actually in production" is fewer use cases than the marketing suggests, but more than the sceptics will allow. Here's where I draw the line, based on what I'm seeing in client conversations and what's verifiable in deployment data.

Scaled and in production

Scaling now — 12 to 18 months to broad deployment

Still predominantly piloting — a three-to-five-year horizon

Autonomous underwriting decisions in life insurance. Clinical trial operations in life sciences. Fully agentic source-to-pay — intake through PO through reconciliation, without human initiation at any step. Strategic sourcing negotiation. None of these are impossible. They're delayed by accountability gaps, regulatory liability, and data readiness — not by what the models are technically capable of.

I've written elsewhere about why closing that gap is a data and architecture problem before it's an agent deployment problem — adding more autonomous capability on top of an unprepared foundation doesn't accelerate the transition, it just makes the eventual reckoning more expensive.

What enterprise buyers actually care about

I've sat on both sides of transformation mandates over the years — advising the buyer, and at other points being evaluated as part of the solution. The decision criteria that show up in the RFP document are not always the ones that determine the actual outcome. Here's my honest ranking of what actually moves these decisions.

#1 — Domain and process depth

This is the non-negotiable filter. Enterprise buyers, particularly in regulated sectors, will not hand autonomous agents to a generalist integrator who doesn't understand how the work actually flows through their organisation. What they're looking for is proprietary process maps, pre-built vertical accelerators, reference clients in the same sector, and a partner capable of having a credible conversation with the COO and the CFO — not only the CTO. You don't win a mandate because of this criterion. You lose one without it.

#2 — Integration capability and data readiness

Every serious analyst study on failed AI deployments cites the same root causes — fragmented legacy systems, poor data quality, agents hitting exactly the same walls that humans have always hit. Buyers have been burned before by partners who delivered an impressive prototype that simply couldn't talk to their ERP. Demonstrated integration depth, proprietary connectors, and a credible, honest answer to "what happens when our data turns out to be a mess" — because it always is — have become execution differentiators rather than nice-to-haves.

#3 — Outcomes-based engagement model

Buyers are increasingly reluctant to sign FTE-based managed services contracts for AI transformation. The question that actually gets asked in the room now is: will you be commercially on the hook for results? Outcome-based pricing — per transaction processed, per error reduced, per cycle time saved — signals genuine confidence in the solution and aligns incentives correctly between provider and buyer. In a market saturated with "agentic AI" rebadging of existing RPA, a provider's willingness to be measured on business metrics is the clearest trust signal currently available.

What falls just outside the top three, in my experience: cost (important, but rarely decisive at the mandate-approval level), AI asset quality (rising fast as a differentiator, but not yet a baseline qualifier), compliance posture (table stakes in regulated sectors — you lose without it, but you don't win on it alone), and delivery geography (increasingly secondary as agentic AI compresses the headcount that made onshore-offshore ratios matter in the first place).

Where the whitespace is — and what it takes to win it

The providers who will own the next five years of this market are not the ones with the broadest capability portfolio. They're the ones who have identified a specific intersection of sector, process, and technology depth — and gone deep rather than wide. Here is where I see the genuine gaps.

By sector

By service line

By capability

The common thread

Every whitespace opportunity above shares the same unlock condition. It isn't better models. It isn't broader capability breadth. It's the ability to answer three questions that enterprise buyers are asking — sometimes explicitly in the RFP, more often implicitly in how they evaluate the room.

Do you know my problem well enough to have designed a solution before we started talking? Can you actually make that solution work inside my systems, on my data, with my constraints? And will you be commercially on the hook if it doesn't deliver?

The providers who win the next five years of BPM and analytics spend won't be the ones who can demonstrate the most impressive AI in a demo. They'll be the ones who have codified the deepest process intelligence, built genuine integration capability, and had the confidence to price on outcomes rather than inputs. That combination is rarer in practice than it looks on a capability slide.