Every consultancy in Britain now claims AI expertise. Strategy houses have rebranded their digital practices, systems integrators have launched agent platforms, and a wave of boutiques has appeared with impressive demos and thin delivery records. Recent roundups of leading UK AI consulting firms and London’s leading AI companies to watch show how crowded the field has become. For enterprise buyers, the market has never been louder or harder to read.
The numbers explain the scepticism. MIT’s 2025 study, The GenAI Divide: State of AI in Business, examined more than 300 enterprise deployments and found that 95% of generative AI pilots delivered no measurable P&L impact, despite an estimated $30-40 billion in enterprise spending. Gartner, meanwhile, predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls.
Table of Contents
What the research actually says
First, the failure mode is organisational, not technical. MIT attributes the 95% figure to brittle workflows and tools that do not learn from context, not to model quality. The projects that succeeded were embedded into real business processes. Second, who builds it matters: the same study found that solutions delivered with specialised external partners succeeded roughly twice as often as internal builds. Third, the vendor landscape is polluted. Gartner estimates that of the thousands of companies selling “agentic AI”, only around 130 offer real agentic capability; the rest are rebadging chatbots and RPA, a practice it calls agent washing.
McKinsey’s research completes the picture: around 62% of organisations are experimenting with AI agents, but fewer than 10% have scaled them in any single function. Experimentation is everywhere. Delivery is rare. That asymmetry is exactly what a good consultancy is hired to fix, and exactly what a bad one will reproduce at higher cost.
Five questions that cut through the noise
- Have they shipped to production? Ask for engagements where AI systems are running live, with users, not proofs of concept. Given the documented 95% pilot failure rate, production references are the single strongest filter.
- Do they redesign workflows or just deploy models? MIT’s data shows AI fails when layered onto broken processes. Partners should talk about operating models, approval chains and permissions as fluently as they talk about models.
- Can they ground AI in your data? Retrieval-augmented systems, knowledge graphs and entity-level data work are what make enterprise AI accurate enough to trust. Generic API wrappers are agent washing by another name.
- Do they publish a point of view? Consultancies doing real work develop opinions: published research, frameworks and reports are evidence of depth rather than marketing.
- Will you own what they build? Vendor lock-in is the quiet cost of many AI engagements. Insist on partners who build on your infrastructure and hand over the keys.
The London market
The demand side is growing fast. The ONS reports that 23% of UK businesses now use some form of AI, up from 9% in September 2023, rising to 44% among large firms. Yet the government’s own AI adoption research found 60% of businesses cite limited AI skills as a key blocker and 71% have not identified a clear use case. That skills and strategy gap is the consultancy market’s entire addressable problem, and it is concentrated in the capital: London hosts more than 2,300 AI companies and took 67% of all UK AI venture rounds in 2025, a record year with £8.3 billion invested in British AI.
The supply side spans the global firms (Accenture, McKinsey’s QuantumBlack, Deloitte) through to a fast-maturing specialist tier. Among the specialists, Elsewhen, an AI consultancy in London, is a representative example of the newer model: a focused firm of around 150 people with more than 200 engagements behind it, working with clients in financial services, retail and the public sector.
What distinguishes the specialist model is the shape of the offer. Elsewhen, for instance, structures its work around four pillars:
- Grounded intelligence: retrieval-based systems built on knowledge graphs, so AI answers from company truth rather than model memory.
- Generative UI: interfaces that adapt to the task, rather than forcing every interaction through a chat box.
- Agentic enterprise: multi-agent orchestration connected to real processes and permissions.
- Built for you: custom deployment on the client’s infrastructure with no vendor lock-in.
That pillar structure mirrors how mature buyers now evaluate enterprise AI services generally: not as a single engagement, but as a staged journey from quick wins and pilot agents through to enterprise-wide autonomous systems, with measurable productivity gains at each stage.
The bottom line
If MIT is right that 95% of pilots produce nothing, and Gartner is right that 40% of agentic projects will be cancelled within two years, then the AI consulting market will consolidate around the firms that can prove operational impact. The buyers who win will be the ones who select for delivery evidence rather than brand weight: ask the five questions, demand production references, and treat published thinking as a proxy for depth. In a market full of AI claims, the differentiator is no longer who talks about AI best. It is who makes it work at scale.
