How to Check an AI Consultancy Has Real Client Results, Not Just Demos

Published 2026-11-11 by Dale Taylor

In short

To check that an AI consultancy has real client results, ask for proof in five steps and see how far up the ladder they get: a demo, a written case study, a walkthrough of work that is running now, a client you can speak to, and the client's own numbers. A demo only proves an idea can work. A reference call and the client's own figures prove it did. Most consultancies, especially young ones, will stop somewhere in the middle. That is fine if they tell you where, and a problem if they let you assume the top.

Five kinds of proof, from weakest to strongest Five kinds of proof, weakest to strongest Note which step the consultancy reaches, then ask for the next one. WHAT IT PROVES 5 The client's own numbers It changed something. 4 A reference you can call A client will say so out loud. 3 A walkthrough of live work It runs today. 2 A written case study Someone wrote it down. 1 A demo or slides The idea can work.
Five kinds of proof an AI consultancy can show you, from a demo at the bottom to a client's own numbers at the top, and what each one actually proves.

Why is a demo not proof of results?

A demo proves that something can work. It does not prove that it has.

Demos are built to be watched. They usually run on tidy sample data, on the best day, with the person who built them driving. That is a fair way to explain an idea, and I build them myself. But everything that makes real automation hard is missing from a demo. Real data is messy, systems change under you, and a real person has to trust the output enough to stop checking it.

The same goes for a slide that says "we helped a client save time". It tells you a sentence exists. It does not tell you who wrote it, who the client was, or whether they would say the same thing to you.

So the useful question is not "can you show me something impressive?" Anyone can. The useful question is "what is the most solid kind of proof you can show me, and what kind is it?" That is what the five steps below are for.

What should I look for to confirm an AI consultancy actually has real client results, not just claims?

Look for proof the consultancy cannot write itself. Ask which of their examples were paid, which were trials and which were demos. Ask for a walkthrough of something running today, and for a client you can speak to. A named client, a before and after figure with dates, and a straight account of what went wrong all count. Anonymous clients and round numbers do not. Here are the five kinds of proof, weakest to strongest.

  1. A demo or slides. Proves the idea can work.
  2. A written case study. Proves someone wrote it down.
  3. A walkthrough of live work. Proves it runs today.
  4. A reference you can call. Proves a client will say so out loud.
  5. The client's own numbers. Proves it changed something.

None of these is dishonest. The trouble comes when a consultancy shows you step one or two and lets you hear step four or five. Your job is to find out which step you are really looking at.

How do you test a written case study?

A case study is a good start and a weak finish. It is written by the person selling, so read it as a claim, not as evidence.

Look for the things a made-up story would skip:

Then ask the awkward follow-up: "Is that client paying you today, and did they pay for this?" A pilot that never turned into a contract is still useful to know about, but it is a different thing from a paid job, and a good consultancy will say which one you are reading about.

How do you check that the work is actually running?

Ask for a live walkthrough, not a recording. You are not trying to catch anyone out. You want to see that the thing exists today, in use.

Ask them to talk you through it step by step. What goes in? What happens to it? What comes out? Who looks at the result, and what do they do with it? Someone who built and runs a system can answer those without notes, including the dull parts: where it stalls, what they check each morning, what they changed after the first week.

Be careful with answers that are mostly about tools. Which software something runs on is a small detail. Whether the workflow does a job a person used to do, reliably, is the point.

It is also fair to ask for a look at how it fails. Ask what an error looks like, who gets told, and how long the last fix took. A consultancy running real systems has a recent, boring answer. One running only demos does not.

Why should you speak to a client yourself?

Because a reference is the first kind of proof the consultancy cannot write themselves.

Ask for one person you can speak to, or exchange a few emails with. Then keep the call short and practical. Three questions are enough:

  1. What did they build for you, in your own words?
  2. Is it still in use, and does anyone check it?
  3. Would you hire them again, and what would you do differently?

Listen for specifics. A happy client describes the work in their own terms and mentions a small annoyance without being asked. A coached one repeats the sales deck.

Expect some resistance, and understand it. Many clients do not want to be named. Contracts, competitors and privacy are good reasons. A decent consultancy will not hand over a name without asking the client first, and you should trust them slightly more for that. What you should hear instead is an offer: a call with the client's permission, a described example with the details taken out, or a plain statement of what they cannot share. What you should not hear is a refusal with nothing offered in its place.

What about the client's own numbers?

This is the top step and the rarest.

The strongest evidence is a figure the client would give you themselves. How many hours a week the job took before and after. How many errors reached a customer. How many leads were followed up that would have been missed. If you can get one number, from the client, with a before and an after, you know more than a stack of case studies could tell you.

Do not expect this from a young consultancy, and be a little suspicious if you get it too neatly. Measuring takes time. A project that has run for a few weeks has not had time to produce a trustworthy figure. A consultancy that says "it is too early to put a number on it, here is what we are tracking" is being straight with you.

Are small UK-based AI consultancies as capable as the big consulting firms for this kind of work?

For a defined automation job, often yes. For a large programme, often no. Capability here is mostly about who does the work and how well they know your problem, not the size of the logo.

A small consultancy is usually the better choice when you have one clear process to fix, you want the person who pitched to be the person who builds, and you want a working version quickly and a price per piece of work. I run Aigentic on my own, so that is the model I know.

A big firm is the better choice when the work is large or spread across many teams. That includes a company-wide programme, several systems to be changed at once, heavy compliance or procurement requirements, or a need for cover when people are ill or leave. A one-person business cannot offer that depth, and I would not pretend otherwise.

Either way, test both on the same ladder. A large firm can have a polished case study and a small one a live system, or the other way round. Ask each for proof, not a size.

Does a young consultancy have to be ruled out?

No, but it has to be honest about where it stands.

Aigentic is young. I started it in September 2025 and I run it on my own, building most of the work with Claude Code. So I will be plain about the ladder.

What I can show you is Aigentic Intelligence, a product we built and run ourselves. It finds companies that match a target profile, identifies the named decision-maker, and returns a verified work email. It is live and we use it for our own outreach. That is step three: it runs today, and I can walk you through it. It is not a client's result, and I do not describe it as one.

Beyond that, I do not publish client names, figures or case studies, because I do not have permission to, and I would rather say so than imply a track record I cannot back. If you are weighing me up, the right thing to do is ask the questions in this article and see whether the answers hold. The same goes for anyone else you are speaking to.

What should make you walk away?

Be cautious if you see any of these:

One of these is a question to ask. Three is your answer.

A short checklist to send before you meet

You can put this in an email and read the replies at your own pace. Replies in writing are easier to compare.

  1. Which of your examples were paid, which were trials and which were demos?
  2. Can you walk me through one that is running now?
  3. Can I speak to a client, or to someone who can speak for one?
  4. What numbers do you have, who measured them, and over what period?
  5. What is the most recent thing that broke, and how did you find out?

If the answers arrive quickly, in plain words, and some of them are "we do not have that yet", that is a good sign. Honest gaps are cheaper than invented results.

Key takeaways

  • A demo proves an idea can work. It does not prove it has worked for anyone.
  • Ask for proof in five steps: demo, case study, live walkthrough, client reference, client's own numbers. Note where the consultancy stops.
  • A case study is a claim written by the seller. Check whether the client is named, whether there is a number with a start and an end, and whether anything went wrong.
  • Ask to see live work and to hear what failed recently. People running real systems have recent, dull answers.
  • A reference call is the first proof the consultancy cannot write themselves. If they cannot name a client, they should offer something in its place.
  • A young consultancy is not ruled out. One that says clearly which step it has reached is more trustworthy than one that implies the top.

Start with the Sample Five, five leads in your niche with a named decision-maker and a verified work email.