The most useful thing we did for a client was tell them no

In July 2026 a client asked us to add an outreach channel to their stack. It was a reasonable ask: the technology was not difficult, the integration was well documented, and we have built similar things before.

On any normal reckoning it was a straightforward build and an invoice at the end of it. What we did instead was spend that time on a question nobody had asked: whether they were actually allowed to do it.

What we found

They were not, on two separate grounds. The plan needed prior opt-in consent from the people it would contact, and they did not have it in the form required.

Separately, and independently of the law, the proposal breached the platform’s own terms of service. That would have put the account at risk whether or not a regulator ever looked at it.

We delivered that finding on 26 July 2026. The channel was never built.

Why this is worth publishing

Because it is the piece of work from that month we most want a prospective client to know about. There is no screenshot and no dashboard. There is a dated document explaining why something should not exist.

Every supplier in this market has a portfolio. Almost none of them publish the month they talked a client out of spending money.

The reason is structural rather than dishonest. A supplier who bills for what gets built has an obvious incentive not to reach the conclusion that nothing should be.

That incentive does not announce itself. It works quietly, as a slight reluctance to ask the awkward question early, and a slight preference for finding out later, by which point the thing exists and somebody has to decide whether to switch it off.

What it would have cost to find out afterwards

Consider the version where we built it first. The channel goes live. It works, in the narrow sense that messages send and some of them get replies, which is exactly the signal that makes a thing hard to switch off.

Then one of three things happens. Somebody complains and a regulator takes an interest. Or the platform notices, and the account goes, taking with it whatever else the business was doing on it.

Or nothing happens for eighteen months and the practice quietly embeds itself into how the company operates. That is the worst of the three, because it is the one that scales.

In all three the remediation costs more than the build. In two of them it costs more than the build several times over, and in one of them it costs something that is not money.

The general version

Compliance questions get asked at the wrong end of a project almost universally, and not out of negligence. At the start there is a keen client and a clear brief, and a legal question feels like an obstacle to work everybody wants to do.

So it goes on the list for later, and later is after the build.

The cheapest moment to discover that something is not allowed is before anybody has spent anything. That means the question belongs in the scoping conversation, in front of the person paying, at the point where the answer can still change the plan for free.

It is also the moment at which a supplier is least incentivised to raise it, which is why we raise it there, and why it is worth asking any supplier rather than waiting to be told.

What we would ask a supplier

Two questions, and the second matters more than the first.

What are the rules that apply to this, and how do you know? A good answer names the source and admits where it is unsettled. A poor one is confident and general.

When did you last tell a client not to build something? If the answer is never, either they have had an unusually clean run of briefs, or the question is not being asked at the point where it would be inconvenient to answer.