A real pitch, a real forecast, and a real £300k mistake.

The pitch

An agency pitching for a new account built a full-year forecast off the account's recent performance. The numbers looked optimistic — but not unreasonably so. The head of marketing ran some checks and came away believing the agency's promised improvements could genuinely deliver them. The pitch was won.

What the recent data was actually showing

The outgoing agency, in its final months on the account, had quietly increased brand search bidding — a common way to make departing numbers look better than the underlying account health actually was. Brand search converts at a very high rate almost by definition: people searching a brand name by name were already coming. Bidding on it doesn't create that demand, it just re-buys it, often at a real cost, for traffic that would have converted for free.

The incoming agency built its entire year-one forecast on this recent data, without checking it against a longer historical baseline. The recent numbers looked like the new normal. They weren't — they were a temporary, artificial lift.

What happened when it was corrected

The new agency, doing the account properly, identified the brand bidding for what it was — spend on traffic that should have been free — and removed it. That was the right call. But because the forecast had been built assuming the inflated numbers were the baseline, removing the brand spend caused blended ROAS to fall sharply, and the year's forecast collapsed with it.

The agency had to reforecast the full year. The revision: £300,000 lower than what had been promised at pitch stage.

The cost was real, not abstract

This wasn't a number that stayed on a spreadsheet. It cost jobs, and it cost the agency the contract. The business moved to a different agency — one whose reporting was built around profit, not ROAS, specifically so a blended metric couldn't quietly mask what was actually happening underneath it again.

The lesson

Nobody involved was being careless. The head of marketing did check the numbers — the checks just weren't the right ones. A forecast built on the last few months looks credible precisely because it's recent, which is exactly why it needs to be tested against a longer baseline before it gets believed. The question worth asking every time: is this the normal pattern, or is this a number that's been temporarily inflated by something that's about to change?