At a webinar in this category, someone in the audience asked the best question of the whole session: it seems like a miracle machine, what's the downside? The answer that came was: I understand why you'd think that.

That's not an answer. And that's a shame, because the question deserves one, especially from the party selling the system. Below is the answer we would give, and it's useful with any provider in this market, including us.

One thing upfront: this isn't a doom story. AI systems work. They just do less than the brochures suggest, and the places where they do nothing are predictable enough to write down.

01 / the answerWhat is the disadvantage of an AI system?

That it's an amplifier, not a solution. A system makes existing work faster and more consistent, but it doesn't invent a way of working that isn't there. If something already runs well, it will soon run better, and without you. If something isn't right, it will now go wrong faster and more often.

That's the core of it, and everything below is an elaboration of it. Anyone who realises this beforehand buys a system at the right moment. Anyone who doesn't is buying acceleration of something that had no direction yet.

02 / the boundaryDoes it solve a shortage of clients?

No. This is the most important boundary, and also the most concealed one. A system takes over repeat work: the work that exists because there are clients. If your inflow is too low, your bottleneck is the front end, not the execution, and a system then amplifies precisely the part that isn't the bottleneck.

We see this more often than we'd like in first conversations. Someone comes in saying "I'm drowning in work", and on closer questioning it turns out not to be a volume problem but a foundation problem: an unclear offering, a shifting target audience, no fixed way of working. Building a system is then the most expensive way to let that problem persist.

A system amplifies a way of working that already runs. It doesn't create one.

If that's your situation, we say so in the conversation, and that's where it ends. That's not modesty, it's self-interest: a system built on top of a shaky foundation becomes a client who's dissatisfied about something we couldn't fix.

03 / the judgementDoes it take decisions off your hands?

No, and you shouldn't want it to. A system proposes what you'd probably do based on what you did before. Whether you give that discount, take on that job, have that conversation or let that client go remains your work. The model has no stake in the outcome and doesn't bear the consequences.

In practice this means something is asked of you every day. Drafts wait for review. That costs minutes, not hours, but it comes back every day. Anyone who puts it off builds up a backlog and then experiences the system as pressure.

There's also a fixed maintenance moment attached to it: once a week you tell it what's changed about your offering, your target audience or your way of working. Skip that and the knowledge layer quietly goes out of date, and you only notice once the output starts to jar. That's work you gain, not work that disappears.

04 / the dipWhy does it cost more time in the first weeks?

Because the system doesn't know you yet. The first drafts are too formal or too loose, miss context that only you have, or neatly answer the wrong question. You correct a lot, and sometimes correcting takes just as long as doing it yourself. That's not a fault: it's the period in which your judgement goes into the system.

This is the point where people drop out, and that can be prevented by knowing it in advance. Anyone who thinks in week two that they've bought something broken, stops. Anyone who knows week two is the dip carries on, and by week six has something that works.

We don't attach a fixed number of weeks to that. It depends on your volume and on how consistently you correct, and anyone who gives you a precise figure in advance can't actually know it.

05 / the wordsWhich promises don't hold up?

Four promises you hear a lot in this market that suggest more than they deliver. They're not lies, they're just too big.

06 / with usWhat does DataDrift explicitly not do?

We're not a campaign agency and we don't work by the hour. We don't sell standalone campaigns, advertising budgets or strategy days. What we deliver is a running system with maintenance on top, and that's a narrower offering than "we help with AI".

Furthermore, the build layer is ours. The connections, the instructions and the flows that do the work aren't handed over. Your accounts, your data and your documented way of working are and remain yours, but the engine running it is our methodology. If you want a party that hands over the complete engine so you never need anyone again, that's not us. That trade-off is set out in who owns it once it works.

And we don't build a button that lets the system go public on its own. That's a real limitation, and you're entitled to weigh it as a downside. The reasoning behind it is set out in why we don't publish anything automatically.

07 / usableWhat should you ask any provider, including us?

Five questions. They cost nothing, take two minutes, and the answer to them says more than a demo.

• What can your system explicitly not do? A provider without a sharp answer to that has either not thought about it or doesn't want to say.
• What will disappoint in the first few weeks, and how will I be supported through that?
• For which type of entrepreneur is this specifically not a good purchase? Anyone who says "everyone" is selling.
• What do I keep if we stop: accounts, data, the built work? And is that on paper?
• What's expected of me once it's running, in time per week?

If a provider answers all five calmly and concretely, you're dealing with someone who knows the work. If they talk around it, you already have your answer without it being spoken.

This text was produced with AI assistance and checked and approved by a human before publication.