Red Flags When Hiring an AI Consultancy: What to Watch For Before You Sign

The short version: The biggest red flags when hiring an AI consultancy are vague pricing, no named senior person on the account, promises of guaranteed ROI before they've seen your data, and a demo that's dazzling but tells you nothing about how it fits your business. If a consultancy can't explain what they'd do differently for your business specifically, walk away before you sign anything.

Why this list exists

I've spent the last two years rebuilding my own consulting business in public, which means I talk to a lot of business owners about who they've hired, who they wish they hadn't, and what it cost them to find out. Some of these conversations are painful. One founder told me she'd paid £22,000 to an "AI transformation" firm over four months and the only thing that had changed in her business was her bank balance. That's not an isolated story. It's a pattern, and the pattern has specific shapes.

This isn't a generic list of things to look out for. These are the exact behaviours I've watched play out with real clients, real invoices, and real disappointment.

The demo that's designed to distract you

If a consultancy's pitch meeting is 80% "look what AI can do" and 20% "here's what we'd build for you," that ratio is telling you something. A slick demo of a chatbot answering questions, or an AI summarising a document in three seconds, is impressive and irrelevant. Any of us can pull up a demo like that in ten minutes using tools that are freely available. The question that matters is whether they can describe, in plain language, what your specific bottleneck is and how their approach solves it.

I sat in on a pitch call for a client (a 40-person recruitment firm) where the consultancy spent 35 minutes showing off a generic AI video generator with no connection to recruitment, hiring, or the client's actual problem, which was slow candidate screening. Nobody in the room asked why. That's on us as buyers too. If you leave a sales call impressed but unable to say what they'll do for your business on Monday morning, that's a flag.

They quote a fixed price before they understand your business

A proper scoping conversation for AI work takes time, because the answer to "what will this cost" depends on your data, your systems, your team's technical maturity, and what "done" looks like. If someone gives you a fixed number like £8,000 or £15,000 in the first call, before they've asked about your CRM, your data quality, or who on your team will use the thing, they're pricing a template, not your business.

Real AI consultancy work in the UK tends to land somewhere between £3,000 for a focused audit and process map, up to £25,000 or more for a multi-month implementation with training and ongoing support. Those numbers vary enormously depending on scope, and that's exactly the point: if a consultancy can name a firm figure without asking a single question about your operations, they haven't done the thinking yet, they've just done the selling. I've written a longer breakdown of what fair pricing looks like on what an AI consultant should reasonably cost, and it's worth reading before any call where numbers come up.

Nobody senior is named until after you've signed

This one catches people out constantly. You have your discovery call with an experienced, articulate person, maybe even the founder of the firm. You sign. The actual work gets delivered by a junior team member or a subcontracted freelancer you've never spoken to, based somewhere the founder has never mentioned, working from a brief that's lost half its detail by the time it reaches them.

Ask directly, before you sign: "Who will be doing the hands-on work, and can I meet them?" A consultancy with nothing to hide will introduce you. One that gets vague, or says "our team will allocate the best fit once we start," is telling you the person you're impressed by isn't the person doing the work.

They promise ROI numbers before seeing your data

This is the one that should worry you most, and it's the one that gets waved through most often because it sounds like confidence rather than a warning sign. If someone tells you "clients typically see a 30% reduction in admin time" or "you'll save £50,000 a year" before they've looked at your actual processes, your actual data, or your actual team's workflow, they are guessing, or worse, repeating a number from a case study that has nothing to do with you.

Here's the uncomfortable bit that most people selling AI consulting won't say out loud: nobody can honestly forecast your AI ROI in a first meeting, because the biggest variable isn't the technology, it's whether your team will adopt and use whatever gets built. I've seen a brilliant, well-built AI tool sit unused for six months because nobody trained the team or built it into their daily routine. The consultancy that hedges and says "we'll know more once we've mapped your processes, but here's what similar businesses typically achieve" is being straight with you. The one that leads with a big confident number is selling you the number, not the work.

They can't point to a client who says no, or who left

Every consultancy that's been operating for more than a year has had a project that didn't work, a client relationship that ended badly, or a scope that had to change halfway through. That's normal. What isn't normal is a firm that presents a spotless record with zero friction anywhere. Ask them directly: "Tell me about a project that didn't go the way you expected." A confident, specific answer ("we misjudged how long data cleaning would take on a project last year, and now we build in a two-week buffer for that") tells you they've learned something. A deflection, or a vague "we always deliver," tells you they either haven't had a real setback or won't admit to one, and both are worrying.

The chatbot that wasn't

I want to give you the full story of one client, because it's the clearest example I've seen of how this plays out in practice. A small ecommerce business, doing around £1.2 million a year in revenue, hired a consultancy that pitched a "fully custom AI chatbot trained on your product catalogue and customer history" for £18,000, delivered over six weeks. The pitch deck had the word "custom" on nearly every slide.

What arrived, five weeks late, was a generic chatbot builder with a prompt pasted in that referenced the client's product categories, no integration with their actual order system, and no training on real customer conversation data at all. The founder only found out it wasn't custom when she asked why it couldn't answer a question about a specific order, and the answer she got was that the tool "wasn't built to connect to that yet." There was never a plan to connect it to that. The word "custom" had been doing a lot of work in that sales deck that the delivery never backed up.

The fix wasn't complicated in the end, but it cost her another £4,000 and six weeks with a different provider to get something that worked. The lesson isn't "don't buy chatbots." It's this: ask exactly what "custom" means, in writing, before you pay a deposit. Does it mean trained on your data? Integrated with your systems? Or does it mean your logo on a template? Get the answer in the contract, not the pitch.

Nobody asks about your data before quoting a solution

This is a quieter red flag but just as reliable. Good AI work starts with a proper look at your data quality, where it lives, who owns it, and how clean or messy it is. If a consultancy proposes a solution before asking a single question about your data setup, they're building on an assumption, not an audit. I've watched projects stall for months because nobody checked, at the start, that the client's customer records were split across three systems that didn't talk to each other. That's a two-hour conversation that should happen in week one, not a discovery made in week eight.

They can't explain how they measure success beyond "engagement"

If you ask "how will we know this worked," and the answer is soft, things like "engagement" or "adoption" without a number attached, push back. A consultancy worth hiring will agree on something specific and trackable before the work starts: hours saved per week, response time reduced from X to Y, cost per lead down by a set percentage, whatever is relevant to your business. If they resist setting a measurable target, ask yourself why they'd rather not be held to one.

They talk about AI in general terms, never your industry specifically

I've read a lot of AI consultancy pitches over the past two years, and the ones that hold up under questioning are always the ones that get specific fast. A consultancy that's good at, say, marketing automation for professional services firms will talk your language within the first ten minutes: your sales cycle, your typical lead volume, the tools you're probably already using. If everything stays at the level of "AI can transform your business" with no reference to your sector's actual workflow, that's a sign they're running the same pitch for every industry, which usually means the same solution for every industry too. I go into this in more detail in the piece on what to check before hiring an AI consultancy for marketing work specifically, because marketing is one of the areas where generic pitches are most common right now.

The uncomfortable truth about who's selling AI consulting right now

Here's the bit that will annoy some people in this industry, and I'll say it anyway because it's true and it's useful to know before you buy: a large share of firms calling themselves AI consultancies today were doing something else entirely eighteen months ago. Social media management agencies, generic "digital transformation" consultants, web design shops. Nothing wrong with pivoting, businesses adapt, I've reinvented parts of my own career more than once. But if a firm can't tell you plainly what they were focused on before AI became the thing everyone sells, and can't show you evidence of hands-on technical work rather than just marketing about AI, that gap is worth noticing. Rebranding isn't automatically disqualifying. Pretending it didn't happen is.

Ask directly: "How long have you been doing AI implementation specifically, not just talking about it?" A firm with a real track record will answer with dates, projects, and names. A firm that's newer to it can still be worth hiring if they're upfront and their team has relevant technical background, but you deserve to make that decision with the real information, not a polished "we've always been AI-first" line that isn't true.

A five-minute checklist before you sign anything

• Ask who specifically will do the hands-on work, and ask to meet them before signing.
• Ask for a fixed quote only after they've asked at least five specific questions about your data, systems, and team.
• Ask for one honest example of a project that didn't go as planned.
• Ask how success will be measured, in a number, before the work starts.
• Ask how long they've been doing AI implementation specifically, and ask for a reference client you can call.

If a consultancy answers all five without flinching, you've likely found someone worth working with. If two or more answers are vague, that's your answer too. For a more detailed walkthrough of this whole process, including questions to put in the contract itself, I'd point you to how to evaluate an AI consultancy before you sign a contract, which covers the paperwork side of this in more depth than I have room for here.

What good looks like, briefly

Not every consultancy is a red flag waiting to happen. I've worked alongside firms that ask hard questions in the first call, quote a range rather than a fixed number until they've scoped, name the person doing the work, and are upfront when something in your business isn't ready for AI yet. That last one matters more than people think. A consultancy willing to tell you "your data isn't clean enough for this yet, let's fix that first" is doing you a bigger favour than one that says yes to everything. I've written a full breakdown of the twelve most common warning signs, with the fast version you can run through in about sixty seconds, in how to evaluate an AI consultant and spot the red flags quickly, and it's worth bookmarking before your next pitch call. If you're based outside the UK, the same principles hold, and I've written region-specific versions for founders comparing quotes, including the honest hiring guide for American founders.

When you are ready to talk to someone, hire an AI consultant explains what a first conversation covers; the how to find an AI consultant worth hiring page covers what gets built.

The sector-by-sector version of this is on AI consultant by industry.

Related: measure roi ai consultant engagement.

Frequently asked questions

What is the single biggest red flag when hiring an AI consultancy?

A fixed price or a promised ROI figure given before the consultancy has asked detailed questions about your data, systems, and team. Real scoping takes time, and a number given too fast usually means a template solution, not one built for your business.

How much should an AI consultancy cost?

In the UK, a focused audit or process review typically runs from £3,000, while a full implementation with training and ongoing support can run £15,000 to £25,000 or more, depending on scope. If a firm quotes a firm number before understanding your business, treat that as a warning sign rather than good value.

Should I worry if an AI consultancy used to do something else, like marketing or web design?

Not automatically. Many good AI consultants have pivoted from adjacent fields and bring useful skills with them. The concern is when a firm won't be upfront about that history or can't show hands-on AI project experience beyond marketing themselves as AI experts.

How do I check if an AI consultancy's case studies are real?

Ask for a reference client you can contact directly, not just a written testimonial or logo on a slide. A consultancy confident in its work will happily connect you, and hesitation there tells you as much as an answer would.

Sources worth reading

• Anthropic's own documentation
• the NIST AI Risk Management Framework