No borrowed logos: how to read an AI agency's proof, and how to read ours

A case study is proof only when you can do three things with it: name the client, trace the headline figure back to how it was actually measured, and attribute the quotation to a real, identifiable person. If any one of those is missing, you are looking at a piece of design that has been styled to resemble evidence. That distinction matters most in a young field like AI services, where almost every provider is newer than its marketing implies, and where a confident page is cheap to produce.

The AI services market is barely a few years old. Almost no agency in it has the long, signed-off client record that a mature firm can point to, and the ones that imply otherwise are usually borrowing credibility they have not earned yet. So the pressure to manufacture proof is intense, and the tools for doing so are everywhere: a stock logo wall, a generated testimonial, a percentage with no working behind it. The result reads like confidence and contains almost nothing you can check.

I find it more useful to treat any unverifiable claim as if it were absent. Not as a lie, simply as data you cannot act on. A figure you cannot trace is worth the same to your decision as no figure at all, which is to say nothing. Once you read pages that way, the difference between a real record and a decorated one becomes obvious quickly.

Three patterns account for most of what passes as evidence on agency sites, and each has a quiet failure built in.

• Borrowed logos. A grid of brand marks suggests a client list, but a logo is not a case. It often means the agency used a tool the brand makes, attended an event, or worked with someone who once worked there. Ask which engagement each logo represents and the grid usually thins out fast.
• Numbers with no method. A figure such as a large percentage uplift means nothing without its working: measured how, over what period, against what starting point. A real result survives that question. A decorative one was never measured against a baseline at all, because there was no baseline.
• Anonymous praise. A glowing quote attributed to an initial and a job title, or to nobody, cannot be checked and therefore cannot be trusted. A testimonial earns its place only when a named person stands behind it and could, in principle, be asked whether they said it.

None of these are always dishonest. A firm may have a genuine relationship behind a logo and simply present it badly. But the burden sits with the page to let you verify it, and a page that makes verification impossible has told you something either way.

You do not need a checklist for every claim. One question covers almost all of them: who, measured how, over what period, against what baseline. Put it to any result you are shown. A firm with real evidence answers without flinching, because the answer is just how they ran the project. A firm without it changes the subject, widens the scope, or tells you the data is confidential. Confidentiality is a fair reason to anonymise a client. It is not a reason a figure cannot be explained.

The same test works on claims about our own field, which is worth practising on because the numbers are public. The lead-response research the whole sector quotes is a good example. When a figure is real, it names its source and its year, so you can read the study yourself: in one audit of 433 B2B companies, 55% took more than five working days to reply to an enquiry, or never replied at all (Drift lead response survey, 2017). When a figure is invented, it floats free of any source, because attaching one would let you check it. The presence or absence of a traceable citation is itself a tell.

Here is the honest position, because the rule applies to us too. 7 Minds Systems is early. We do not yet have a library of signed-off client results, and I would rather tell you that plainly than dress up something thinner than it looks. So we made a deliberate choice about what to publish, and it is the same choice I would want a supplier to make for me.

Where we show numbers in a worked example, they are labelled as modelled scenarios, with the inputs visible and described as projections rather than client history. The ROI calculator works the same way: it runs your own figures through deliberately conservative assumptions and shows its working, so the number you leave with is yours, not a borrowed one. The only outside figures we cite are from named, dated, public research you can read for yourself, like the Drift audit above. We would rather show you an honest model and a real study than an unverifiable triumph.

On clients, we name a small founding cohort of real businesses we work with, and we name them as a roster only. We do not attach invented revenue figures or put words in their mouths. When a client has results worth publishing and signs them off, those will appear with the client named and the method shown, measured the way this whole post argues a result should be. Until that day, a stand-in honestly labelled is better than a fabrication confidently presented.

It helps to know what you are waiting for, from us or from anyone. A case study worth the name will name the business, state the problem in that business's own terms, show the starting position before any work began, and report the change against it over a stated period. It will quote a person you could find. It will not round every number to a memorable figure, because real results are rarely that tidy. When you see one built that way, you are reading evidence. Everything short of it is a claim awaiting proof, and you are entitled to price it as such.

If you are weighing AI suppliers right now, take the question above into every sales call you have, ours included. And if you would rather start from your own numbers than anyone's testimonials, run them through the ROI calculator or bring them to a thirty-minute call, and we will model what an Autonomous Digital Branch would be worth to you, with the working shown.