What are the warning signs of a bad AI consultancy?

The warning signs of a bad AI consultancy are vague case studies without specific metrics, no production systems you can verify, guaranteed ROI promises, reluctance to provide client references, and an insistence on selling you a platform or product before they have understood your specific problem. The single biggest red flag is a consultancy that talks about what AI can do in general rather than asking what your business specifically needs.

Short answer: Vague case studies, no production systems, guaranteed ROI, no client references, and selling solutions before understanding your problem. They talk about AI, not your business.

Red flag 1: Vague or unverifiable case studies

Good AI consultancies have specific stories. “We built an intake system for a 25-person employment law firm that reduced unqualified calls by 70% and saved £78,000 per year. The system has been running for 14 months and has processed 8,000 enquiries.” That is a case study.

Bad consultancies have generalities. “We helped a leading professional services firm transform their operations with cutting-edge AI solutions, resulting in significant efficiency improvements.” That is marketing copy.

What to ask: “Can you name a specific system you built, what it does, and what measurable result it delivered?” If the answer involves words like “transformative,” “cutting-edge,” or “significant improvements” without numbers, you are talking to a marketing operation, not a delivery one.

What to verify: Ask for client references. Call them. Ask: “Is the system still running? Did it deliver the promised results? Would you hire them again?” A consultancy that will not provide references is hiding something.

We are transparent about our own track record: 30 production systems across 6 clients. The Calder and Reid system saved £78,000 per year. The Meridian system generated 1,000 times its cost in pipeline value. These are specific and verifiable.

Red flag 2: They sell before they listen

A good AI consultancy spends the first conversation asking questions about your business. What are your workflows? Where do you spend the most time? What frustrates your team? What would make the biggest difference to your bottom line?

A bad AI consultancy spends the first conversation presenting their platform, showing demos, and explaining what AI can do. They have a solution and they are looking for a problem to attach it to.

The test: After your first meeting, consider what percentage of the time they spent asking versus presenting. If they talked more than you did, they are not trying to understand your needs. They are trying to sell their product.

Why this matters: AI systems that work in production are built around specific business workflows, not generic AI capabilities. A consultancy that does not deeply understand your workflow will build a system that looks impressive in demo and fails in practice.

Red flag 3: No production track record

There is a vast difference between building a proof of concept and running a production system. A POC handles the happy path with clean data in a controlled environment. A production system handles edge cases, bad data, user errors, API outages, and the thousand things that go wrong in real-world operation.

Many AI consultancies have impressive POCs and zero production deployments. They can demo something that looks amazing. Whether it can handle your actual data, integrate with your actual systems, and work reliably for your actual staff is a different question.

What to ask: “How many systems have you built that are currently running in production? What is the oldest one? What has been your biggest production failure and how did you handle it?”

A consultancy that has never had a production failure has never run a production system. Things go wrong. How they handle problems matters more than whether they occur.

Red flag 4: Guaranteed ROI or unrealistic timelines

ROI depends on factors outside the consultancy’s control: your data quality, your staff’s willingness to adopt new tools, your client volume, and your business environment. Any honest consultancy knows this.

When a consultancy guarantees specific ROI figures, one of two things is happening. Either they are being dishonest and will blame external factors when the guarantee is not met, or they have built enough margin into their pricing to absorb the guarantee cost, meaning you are overpaying.

Similarly, unrealistic timelines are a red flag. If they promise to deploy a complex AI system in two weeks, they are either oversimplifying the work or planning to deliver a POC that they will call production-ready.

Realistic timelines: An AI audit takes 1 to 3 weeks. A single-system build takes 4 to 12 weeks. Firm-wide deployment takes 3 to 12 months. If a proposal is significantly faster than these benchmarks, ask detailed questions about what is being cut to achieve the speed.

Red flag 5: They cannot explain how AI works in plain language

AI is not magic, and a good consultancy should be able to explain exactly what their system does, how it works, and why it will deliver results without resorting to jargon or hand-waving.

If a consultancy’s explanation involves phrases like “proprietary algorithms,” “advanced neural networks,” or “our unique AI platform” without being able to explain what those things actually do with your data, they are either hiding technical weakness behind jargon or their team does not understand their own technology well enough.

What to ask: “In simple terms, what happens when a client enquiry enters this system? What does the AI do, step by step, and what happens if the AI gets it wrong?”

A good answer walks through the specific process with your specific data. A bad answer talks about AI in the abstract.

Red flag 6: No plan for when they leave

The best AI consultancies build systems your team can maintain. The worst create dependency by building proprietary systems that only they can update, using their own hosting infrastructure that you cannot migrate from, or retaining intellectual property rights that prevent you from hiring someone else to maintain the system.

Contract terms to scrutinise:

IP ownership. You should own the code and data from any custom build. Some consultancies retain IP and licence it back to you, creating permanent dependency.

Hosting lock-in. If the system runs on the consultancy’s infrastructure, what happens if you want to move it? Are there migration tools and documentation? Or are you locked in?

Knowledge transfer. Is documentation and training included in the engagement? Will your team be able to make routine changes without calling the consultancy?

Exit provisions. What happens if you want to end the relationship? Can you take the system with you? Is there a transition period?

A consultancy that profits from your dependency is incentivised to make systems hard to maintain without them. A consultancy that profits from your success is incentivised to make systems easy to maintain.

Red flag 7: They overpromise on AI capabilities

AI in 2026 is powerful but not omniscient. It cannot reliably handle novel legal reasoning, complex strategic decisions, or tasks requiring deep domain expertise that it has not been trained on. Any consultancy that suggests otherwise is overselling.

Common overpromises:

“AI will replace your junior staff.” It will not. It will change what they do, automating data entry and routine tasks while requiring them to focus on review, quality assurance, and exception handling.

“Our AI has 99% accuracy.” On what task, with what data, measured how? Overall accuracy figures are meaningless without context. Ask for accuracy on the specific task with data comparable to yours.

“This will transform your business.” Maybe, over time. But AI adoption is incremental. The first system improves one workflow. Transformation happens when multiple systems compound over months and years.

“You do not need to change anything.” AI systems require workflow changes, staff training, and process adjustment. Any consultancy claiming zero disruption is underestimating the implementation challenge.

What good looks like

For contrast, here is what a good AI consultancy does:

Asks detailed questions about your business before proposing anything
Shows specific case studies with measurable results in comparable businesses
Provides client references willingly
Gives honest assessments of what AI can and cannot do for your situation
Provides fixed-price proposals for defined scope
Transfers IP ownership to you
Includes documentation and training as standard
Offers ongoing support without creating dependency
Admits limitations and recommends alternatives when appropriate

The AI consultancy market is growing fast, and with it the number of firms selling capability they do not have. The warning signs are consistent and recognisable. Trust consultancies that show you what they have built, not what they can imagine. Trust specific metrics over general claims. And trust your instinct when something feels like a sales pitch rather than a professional consultation.

FAQ — RELATED QUESTIONS

What is the biggest red flag when evaluating an AI consultancy? +

They cannot name a specific system they built that is still running in production. Any consultancy can build a demo. The test is whether their systems survive contact with real users, real data, and real business requirements over months and years.

Should I be suspicious of guaranteed ROI figures? +

Yes. ROI depends on your data quality, staff adoption, process complexity, and business factors outside the consultancy's control. Honest consultancies provide ROI estimates based on comparable engagements. Guaranteed ROI is either dishonest or the consultancy is pricing in enough margin to cover the guarantee, which means you are overpaying.

Is it a red flag if a consultancy does not specialise in my industry? +

Not necessarily, but it means you are paying for their learning curve. Ask how they will acquire domain knowledge. If the answer is 'we learn from you,' you are subsidising their education. If the answer involves specific research, comparable sector experience, or domain experts on their team, it may be acceptable.

How many case studies should a good AI consultancy have? +

Quality matters more than quantity. Three detailed case studies with specific metrics, named practice areas, and referenceable clients are worth more than twenty vague claims about transformation. Ask for case studies relevant to your sector and firm size.

Is it bad if a consultancy recommends their own product? +

It depends. If they have a product that genuinely fits your needs, recommending it is efficient. If they push their product regardless of your actual requirements, it is a sales tactic. The test: do they assess your needs first and recommend their product second, or do they lead with the product pitch?

Should I worry if a consultancy is very new? +

New consultancies can be excellent if their team has relevant experience from previous roles. Ask about the team's background, not just the company's age. A two-year-old consultancy founded by engineers who spent a decade building AI systems may be stronger than a ten-year-old firm that pivoted to AI last year.

What contract terms should I watch out for? +

IP retention (they own the code they build for you), lock-in clauses (you must use their hosting), no exit provisions (no way to leave without rebuilding from scratch), and open-ended billing (time and materials with no cap). Good contracts have fixed prices, client IP ownership, and clear handover provisions.

Is it a red flag if a consultancy will not give a fixed price? +

For well-defined projects, yes. A competent consultancy that has built similar systems before can price them with reasonable accuracy. Time-and-materials billing for a defined scope usually means the consultancy is uncertain about their ability to deliver, and the uncertainty costs fall on you.