13 min read

How to Tell Real AI Capability From a Pitch

If you run an engineering firm, a testing lab, or a technical consultancy, you have probably noticed the pitches changing. The agencies that disappointed you last time are back, with the same promises wearing a new coat. The word on the coat is "AI".

You were never the marketing person. That is not a failing. Keeping up with this stuff is a full-time job, and you already have one. So when a vendor tells you their system is smart, learns your business, and runs your marketing for you, you have no easy way to tell whether that is real or whether it is the same cookie-cutter work with a better sales deck.

This guide gives you five questions you can ask in an actual sales conversation. Each one is testable on the spot. Each one is hard to fake.

A fair warning first: we build the thing we are teaching you to evaluate. FF Tech runs a marketing module for technical businesses. So we have skin in the game. We are not pretending to be neutral. Instead, we are going to show you how to apply each question to us, in this article, so you can see whether we pass our own test. If a vendor will not let you do that, you have your answer.


Question 1: What does it produce on its own, and what does a person do?

The useful question is not "is there a human involved". Of course there is. The useful question is where the line sits. Ask: what does your system produce without anyone touching it, and what does a person actually do before it reaches my customers?

A real answer is specific. It names the work the software does and names the work the person does. A weak answer waves at "a blend of human and AI" and moves on.

For us: the software learns your business and writes from it, then a real person reviews everything before it is published. Nothing reaches your customers unreviewed. That person owns your account and is accountable for the result. The software does the marketing work you never get to. The person makes sure it is right.

Question 2: Can you show me output from a business like mine?

This is the question that ends most sales conversations early, which is exactly why you should ask it.

A generic demo proves nothing. Anyone can show you polished sample copy about a fictional plumber. What you want to see is writing about a business in your world, using your kind of language: your specifications, your accreditations, your methods, the sort of jobs you actually quote on. A testing lab does not sound like a marketing agency, and a vendor whose system genuinely learns a business will not produce copy that does.

For us: the proof is the site you are reading. FF Tech is pilot zero. We run the engine on our own marketing first, before selling it to anyone. This article was produced by the same engine we would run for you. You can judge the work directly, not a curated screenshot of it.

The test for any claim of writing in your voice: could only a business like yours have produced this sentence? If a competitor could paste their name over it unchanged, it has no substance.

Question 3: What compounds, and what resets every month?

Most marketing you have bought was a project. It launched, it ran, it stopped, and when it stopped you were back where you started. The next agency began again from zero, learning your business from scratch, often badly.

So ask: what does your system know about my business in month six that it did not know in month one? A real capability builds on what it learns. A repackaged service does the same thing every month and calls the invoice a relationship.

This is where we are careful to be honest with you. Our engine is designed to get more capable the longer it runs, knowing your business better over time. That is the reason we built it this way. It is the thesis behind the model, not a result we are claiming to have proven, because we are pilot zero and the long-run client record does not exist yet. A vendor who promises you the compounding outcome as a done deal is selling you the destination. Treat that the way you would treat any promise of a result no one can yet show you.

A comparison of a marketing project that resets each month against a system that builds on what it learns over time.
A project that restarts versus a system that builds

Question 4: Explain the mechanism without the category words

Here is a quiet tell. Ask a vendor how their system works, and listen for whether they explain a mechanism or just stack up labels.

Labels are easy. A real explanation is plain, and a person who actually built something can give you one without hiding behind the category. If every answer is the buzzword again in a slightly different order, there may not be much underneath it.

For us, plainly: most of the work runs on our own hardware rather than renting it from a cloud provider, which is how we hold the cost steady and keep your data under control. We use outside AI services selectively, where they genuinely help. That is the mechanism. You do not need to understand the internals to test the answer. You only need to notice whether the vendor can give you one.

Question 5: What happens to my data and my work if I leave?

The last question is about leverage, and it tells you how a vendor really sees the relationship.

Ask: if I stop paying you, what do I keep? A vendor confident in the value of the work does not need to trap you to retain you. One that locks your data, your content, and your accounts away is telling you they expect you to want to leave.

For us: month to month, no lock-in. There is a 30-day calibration period at the start, and if there is no working baseline at the end of it, you get your money back. On the way out you keep your data and the knowledge artifacts we built about your business. The engine and the infrastructure stay with us. We say results typically take three to six months, and we say it up front rather than promising you a number we cannot stand behind.


What this all comes down to

There is a common thread under these five questions. A real capability survives being asked exactly how it works. A pitch needs you not to ask.

You are good at separating a real specification from a hopeful one in your own field. You read a data sheet and you know what is measured versus what is implied. Apply the same instinct here. Make the vendor show the work, explain the mechanism, and tell you what you keep. We are bound by the same rules every Australian business is, including consumer law that prohibits misleading conduct, and a vendor who is comfortable with these questions is a vendor comfortable with that.

We wrote this guide to be used against us as much as anyone. That is the point. The way you tell a real capability from a marketing claim is to ask the questions a claim cannot answer, and then watch what the vendor does next.