Generative AI Consulting: What Businesses Actually Need vs What Vendors Sell

In 2019, if you told a mid-sized business owner that within five years a machine would be able to draft their proposals, summarise their meetings, write their client communications, and generate financial reports from raw data — they would have asked you to leave the room. In 2026, they're asking why it hasn't happened yet.

The gap between the promise and the reality of generative AI consulting is now wide enough to cost businesses serious money. Not because the technology doesn't work — it does — but because the industry that has sprung up around it is optimised for the wrong outcome. It is optimised for selling consulting engagements. Not for delivering working systems.

Generative AI consulting has become one of those phrases that means everything and nothing simultaneously. Every consultancy in the country has added it to their website. Every conference has a panel on it. Every vendor has a slide deck explaining why their platform is the answer to a question nobody in the audience asked.

Meanwhile, the business owner who actually needs help is sitting in a meeting room in Leeds or Bristol or Edinburgh, listening to a pitch about "large language models" and "multimodal architectures," and thinking: I just want to know if this thing can write my proposals faster.

There is a significant gap between what generative AI consulting firms sell and what businesses actually need. This piece is about that gap — what it costs, why it exists, and how to avoid falling into it.

What Most Vendors Sell

The typical generative AI consulting engagement, as sold by most firms in 2026, follows a predictable structure. I have seen this structure in enough proposals to describe it from memory.

There is a discovery phase. This lasts 4–8 weeks. It is billed by the day. A team of consultants interviews your staff, maps your processes, and produces a document describing what they found. This document tells you things you already know, expressed in language you wouldn't use.

Then there is a strategy phase. Another 4–6 weeks. The same team, or a different one, takes the discovery findings and produces a strategy document. This document recommends 10–20 potential AI use cases across your business, ranked by impact and feasibility. It has a matrix. Every consulting strategy document has a matrix.

Then there is a proof-of-concept phase. One of the use cases — usually the easiest, not the most impactful — gets built as a prototype. The prototype works in a demo environment. Everyone is impressed. A review meeting is held.

The whole thing has taken 3–6 months and cost £50,000–£200,000. At the end of it, you have a document and a prototype that doesn't connect to your actual systems.

Building something that actually works inside your business? That's Phase 2. Phase 2 has a separate proposal. Phase 2 costs more.

This model exists because it de-risks the engagement for the consultancy. They get paid for thinking. The client bears the risk of whether the thinking leads to anything useful. For enterprise clients with dedicated innovation budgets and 18-month planning horizons, this is tolerable. For a business with 30–200 people trying to solve a real problem, it's an expensive way to learn what you already suspected.

What Businesses Actually Need

When a business owner says they need generative AI consulting, what they're actually asking for is one of five things. They may not use these words. They may not know which of the five applies to them. But the underlying need is always one of these.

1. "Where would AI save us the most time?"

This is an audit. Someone maps your operations, measures where time goes, and identifies the highest-value automation opportunities. The output is a prioritised list with real numbers — hours recoverable, cost savings, implementation cost, payback period. You walk away knowing exactly what to do first and exactly what it will cost.

Duration: 2 weeks. Cost: £2,500–£5,000. This should not take longer or cost more than this for a business your size.

2. "Can AI write our proposals, reports, or client communications faster?"

This is a content automation build. Generative AI drafts documents using your templates, your tone, your data. The system connects to your CRM or document management, pulls the relevant context, and produces a first draft the partner can review and send. The output is working software connected to your actual processes.

Duration: 4–6 weeks. Cost: £8,000–£15,000.

3. "Can AI handle our client intake and onboarding?"

This is a workflow automation build. AI processes incoming enquiries, collects the information you need, routes to the right person, and pre-populates your CRM. The client experience improves because response times drop from hours to minutes. Your team's experience improves because they stop doing admin and start doing advisory from the first interaction.

Duration: 4–8 weeks. Cost: £10,000–£20,000.

4. "How do we train our team to use AI properly?"

This is training. Not a lecture about what AI is. Hands-on workshops using your actual workflows and tools. At the end, your team uses AI daily as part of how they work — not as a novelty they tried once and forgot about. The output is a team that is measurably faster at the tasks that eat their week.

Duration: 1–2 days. Cost: £2,000–£5,000.

5. "We want to build an AI-powered product."

This is product development. Design, build, test, launch. The output is a working product that you own. This is the engagement where generative AI's capabilities are most transformative — products that would have taken 6 months to build two years ago can now ship as MVPs in 4–12 weeks.

Duration: 4–12 weeks for MVP. Cost: £15,000–£50,000+.

Notice that none of these five require a three-month discovery phase. None of them require a strategy document you can't act on. All of them have clear outputs, fixed timelines, and measurable outcomes. The reason most generative AI consulting engagements don't look like this is not because the work is more complicated. It's because longer, vaguer engagements are more profitable for the consultancy.

How to Evaluate a Generative AI Consultant in 15 Minutes

The market is flooded with people who learned ChatGPT six months ago and now call themselves generative AI consultants. Here is how to separate the people who can deliver from the people who can present.

Ask them to name a product they've built. Not a client engagement. A product. Something they designed, shipped, and put in front of real users. If they've only ever advised, they don't have the operational understanding to build something that works inside your business. Building and advising are different skills. You need both. Most consultancies only have one.

Ask for case studies with specific numbers. Not testimonials. Not "we helped a client in the professional services sector." Specific numbers: hours saved per week, cost reduced per quarter, revenue gained per year. If they can't produce them, they either haven't done it or they didn't measure whether it worked. Both are disqualifying.

Ask what happens after the audit. Some firms do brilliant audits and then hand you a beautifully formatted document you can't act on without hiring developers separately. The best firms audit, build, implement, and measure — end to end, one team, one relationship, one point of accountability.

Ask for a fixed price. Generative AI consulting for a business your size should not be billed by the hour. Hourly billing incentivises the consultant to take longer. Fixed pricing means they've scoped the work, they understand what's involved, and they're confident enough in their ability to deliver that they'll put their margin on the line. If they can't give you a fixed price, they don't understand the work well enough to do it.

Ask what they'd tell you NOT to automate. Any consultant who recommends automating everything is either inexperienced or dishonest. Good generative AI consultants know where the technology falls short. They know which processes need human judgement, which data isn't clean enough to automate reliably, and which workflows will cost more to automate than they save. If they can't articulate what's not worth doing, they don't know the space well enough to advise you.

The Generative AI Landscape in 2026

Generative AI has matured remarkably in the last two years. The models are better — GPT-4o, Claude, and their successors produce output that is genuinely useful for business tasks, not just impressive demos. The costs are lower — API pricing has dropped by 90%+ since 2023. The tools for connecting AI to existing business systems are production-ready — you no longer need a six-person engineering team to integrate AI into your workflow.

This means the barrier to getting real value from generative AI has dropped significantly. But — and this is the critical point — only if you work with someone who knows how to connect the models to your actual workflows, not just demonstrate what ChatGPT can do in a browser tab.

The difference between a generative AI demo and a generative AI system is the difference between a car in a showroom and a car on the road. The showroom version looks great. The road version needs to handle potholes, traffic, weather, passengers, and the fact that the driver hasn't read the manual. Every generative AI consultancy can do the showroom version. Very few can do the road version.

The businesses getting the most from generative AI in 2026 are not the ones with the biggest budgets. They're the ones who started with a specific, measurable problem, found a consultant who could build the fix in weeks rather than months, and measured the result in pounds or hours rather than sentiment.

Why This Matters Now

The window is closing on "early adopter" advantage. In 2024, using generative AI for proposals, client communications, or document processing was a competitive edge. In 2026, it's baseline. The practices and businesses that haven't implemented it aren't ahead — they're behind. Their competitors are already producing proposals in an hour that used to take a day. Their competitors are already handling client intake at 11pm that used to wait until 9am.

The cost of inaction is no longer theoretical. It is measurable in lost clients, slower response times, and team members who leave because the work is still manual when everyone knows it doesn't have to be.

The businesses that are still "exploring" generative AI in mid-2026 are the ones that hired the wrong consultant — the one who sold them a strategy document instead of a working system. Don't be that business.

Not sure where to start? Read about what AI consulting actually costs for small businesses, or take the AI Readiness Assessment to find out where the biggest opportunities are in your business.