I’d just taken over marketing at a business turning over $300m a year. Positive EBITDA, targets being hit, nothing obviously wrong with it.

What bothered me was one line in the budget I’d inherited.

Three million dollars a year on paid search. Thirty per cent of everything I had.

I’d never seen that much go into a single channel. That was the whole of my objection. Not a theory, just a number that looked out of proportion, and a new person’s licence to ask what it was doing.

The reporting on it was good.

That’s the part people find hardest to believe. It was run by a good digital team and supported by an expensive agency who also knew what they were doing. Cost per click, cost per lead, conversion rates, all of it detailed and all of it real.

Everyone was busy and the dashboard was green.

Nobody was pleased to be asked about it, and I understand why. A new marketing director questioning the biggest line in the digital budget looks like a new marketing director who doesn’t understand digital. Which was fair. I’m not a digital specialist. I’m a marketer.

So rather than argue about it, I asked the team and the agency to walk me through how the optimisation actually worked.

It went like this. Somebody clicks an ad, lands on a page, converts. That becomes a lead in the CRM. The lead goes to a sales rep. The rep works it. Eventually it becomes an opportunity, and eventually that opportunity closes, won or lost.

Every week, the agency took an extract from the CRM, found the deals that had closed, traced them back to the ad clicks that started them, and optimised the campaigns towards whatever had produced those wins.

Which sounds entirely sensible until you sit with it for a second.

So I asked how you optimise today’s advertising using buying signals that might be three months old.

The room went quiet.

Not because anyone had been hiding anything, and not because anyone had got it wrong on their own terms. Because nobody had said it out loud in that order before, and once it was said, everyone in the room could see it at the same time.

The method the agency were using is a good method. In e-commerce it’s the right one, because the click and the purchase are minutes apart and the feedback loop is tight enough to steer by. Optimise towards the clicks that bought, get more clicks that buy.

Here, the signal left the internet at the point of conversion and disappeared into a CRM and a sales process for weeks or months. By the time it came back, the campaigns that produced it had already been adjusted a dozen times on the basis of the ones before.

We were steering by a wake that had nothing to do with where the boat was pointing.

And once the optimisation is untethered, what it optimises for is volume.

Cheap clicks, cheap leads, lots of them. Which is exactly what it had been delivering, faithfully, for years, at three million dollars a year.

When we worked out what those customers were actually worth, the answer was about $40 each. Not $40 a month. Not $40,000 across a contract. Forty dollars, in total, for as long as they stayed.

The last question is why nobody had noticed.

We had around 13,500 customers. Roughly 1,100 of them were large, long-tenured and extremely profitable. Those eleven hundred paid for everything. They paid for the other twelve thousand four hundred, they paid for the three million, and they kept every number on every dashboard the right colour.

Average the whole book and it looked healthy, because on average it was.

Two completely different businesses were running under one roof on one budget, and the good one was quietly funding the bad one. Nothing in the reporting separated them, because nothing had ever been cut that way.

That’s what makes this failure so durable. It doesn’t arrive as a loss. It arrives years later as growth that stopped paying for itself, long after the decisions that caused it.

The spend moved. Not away from marketing, and not down. Away from the segment that couldn’t repay it, towards the one that could.

The team, the agency and I worked on it very differently from that point on, and it was a great deal better than what we’d all been doing separately.

One last thing, which is why I tell this story rather than a tidier one.

I didn’t know I was doing ratio work. I’d never calculated a customer acquisition cost in my life and I couldn’t have told you what customer lifetime value meant. Nobody I worked with used either term.

I worked out years later what I’d actually done, which was compare what a customer cost against what a customer was worth, because the alternative was signing off a number I couldn’t explain.

You don’t need to be qualified to ask that. I wasn’t.

What you need is to be suspicious of an average, and willing to look stupid in a room full of specialists.