AI advisory decides what your company should do, an implementation partner builds and integrates the systems, and AI training changes how your teams actually work. They are three different purchases, and buying the wrong one is the most expensive mistake in AI procurement. Most firms in our comparison of the best AI consulting firms in London and the UK sell one of the three and quietly upsell their own lane. Here's how to diagnose which purchase your problem actually is.
Key Takeaways
- AI advisory produces decisions: strategy, governance, investment cases, roadmaps. Implementation produces working systems: builds, integrations, deployments. Training produces changed behaviour: measured usage in real workflows.
- Firms diagnose in their own favour — a build firm hears every problem as a build, a strategy house as a strategy gap. Diagnose before you shortlist, or the shortlist diagnoses for you.
- The UK's stuck-adoption numbers are a category error at scale: 54% of SMEs have adopted AI but only 11% use it extensively. That gap is a training problem being answered with more advisory decks and more software.
- The standard sequence is advisory → implementation → training, but most mid-market companies in 2026 need advisory → training first, because the tools are already bought.
- One firm can legitimately cover two lanes; be sceptical of "all three, at scale, equally well". Ask which lane the firm would name as its centre of gravity — then verify with question nine of our vendor questions.
The three purchases, defined
AI advisory (also sold as AI strategy consulting) answers: where does AI create value and risk for us, in what order, under what rules? Deliverables are an AI charter, a governance framework, a prioritised use-case portfolio and a 30/60/90-day roadmap with owners. It's what our AI strategy sprint produces in two weeks.
An AI implementation partner answers: who makes the systems exist and run? That covers custom builds (models, agents, data platforms — Faculty-into-Accenture and QuantumBlack territory) and estate-wide integration (Copilot, ChatGPT Enterprise, Gemini rollouts — the large SIs' home game).
AI training — specifically AI adoption training — answers: who changes what people do on Tuesday morning? Deliverables are workflow-specific skills, champions, manager enablement and measured weekly usage.
Which one is your problem? A symptom table
| Your symptom | The purchase | Wrong purchase that gets sold instead |
|---|---|---|
| "We have no coherent AI position; every department is improvising" | Advisory | An enterprise platform build to "force alignment" |
| "Legal/security keeps blocking pilots; no one knows the rules" | Advisory (governance) | More pilots, launched quietly |
| "We need a capability our stack genuinely can't do" | Implementation (build) | A strategy phase re-discovering what you told them |
| "We bought 500 Copilot/ChatGPT seats; usage is stuck around 20%" | Training | Another integration, or a new tool |
| "The pilot worked; the org didn't change" | Training (+ light advisory) | Scaling the pilot's infrastructure |
| "We don't know where value would even come from" | Advisory (readiness audit) | A tool demo tour dressed as discovery |
The fourth row is the UK's default condition in 2026. Research puts roughly 90% of companies investing in AI while only about 20% of employees actively use the tools — and among UK SMEs, 54% report adoption but just 11% use AI extensively (British Chambers of Commerce, collected in our UK SME AI statistics). That is not a missing-system problem. It's a behaviour problem wearing a software budget.
"Ask a build firm why adoption is stuck and you'll get an architecture answer. Ask a strategy house and you'll get an operating-model answer. Neither is lying — firms genuinely see problems through their own delivery muscle. Which is exactly why the diagnosis has to happen on your side of the table."
— Meera Sanghvi, Co-Founder, Spicy Advisory
The sequence, and where it inverts
The textbook order is advisory → implementation → training: decide, build, enable. It holds when the capability genuinely doesn't exist yet. But for most UK mid-market companies in 2026 the tools are already licensed, which inverts the middle: advisory → training, with implementation entering later, once usage data shows which workflows deserve deeper automation. Starting with a light readiness audit rather than a heavy strategy phase keeps the diagnosis cheap — what that audit must contain is in our AI readiness audit guide.
"The sequence mistake we see most is buying implementation second when nothing proved the demand. Usage data from a trained team is the best requirements document ever written. Build after it exists, and the build pays back; build before it, and you've automated a guess."
— Toni Dos Santos, Co-Founder, Spicy Advisory
Can one firm do all three?
Two lanes, credibly, yes — the pairings are natural: strategy houses bolt builds on (BCG X), builders bolt strategy on, boutiques like us pair advisory with training. All three at equal depth is rare, because the economics differ: implementation scales with engineers, advisory with partners, training with practitioners. The test is simple: ask for the firm's revenue centre of gravity and a reference in each lane you're buying. Then apply the ten questions from how to choose an AI consultancy — question nine ("what work are you wrong for?") does most of the work. Where we land: advisory plus training is our centre of gravity, we don't build production systems, and we name build partners when the diagnosis says build. Budget ranges for each lane are in the UK AI consulting cost guide.
One-line diagnostic: finish the sentence "we'd consider this year a success if ___". A decision in the blank → advisory. A system in the blank → implementation. A behaviour in the blank → training. Two blanks → sequence them; don't buy them blended and unpriced.
Frequently asked questions
What is the difference between AI advisory and an AI implementation partner?
Advisory produces decisions — strategy, governance, roadmap — typically in weeks. An implementation partner produces working systems — builds and integrations — typically in months. Different deliverables, different pricing logic, different firms at the top of each market.
Do we need AI strategy consulting before training?
A light version, yes: training without a charter and priorities produces enthusiastic chaos. But a two-week sprint is usually enough to frame a quarter of training. A six-month strategy phase before anyone touches a tool is the classic over-purchase.
Can training replace implementation?
No — they solve different problems. Training makes licensed tools productive; it can't create a capability your stack lacks. The honest question is order: in 2026 most companies have unused capability already paid for, which makes training the higher-ROI first move.
What should each type of AI partner cost in the UK?
Advisory: £3,500–£40,000 for audits and sprints. Implementation: £50,000 to seven figures depending on scope. Adoption training: £15,000–£50,000 for a mid-market quarter. Full bands and day rates are in our UK AI consulting cost guide.
Sources
- British Chambers of Commerce / Atos, UK SME AI adoption research 2026; ONS Business Insights — primary sources gathered in our UK SME AI statistics roundup
Get the diagnosis before the shortlist
Twenty minutes with the free scorecard tells you which of the three purchases your situation actually is — before any vendor, us included, diagnoses it in their own favour. We're Spicy Advisory: advisory + adoption training, founder-led, 50+ companies, 1,500+ professionals trained, 4.98/5.
Run the Free AI Adoption Scorecard Book a Free 20-Minute Call