TV doesn't stop working when the ad stops

Stop a TV campaign on Friday and sales don't fall off a cliff on Monday. They ease down over the following weeks. The same is true, to different degrees, of almost every channel.

That delay has a name in mix modelling: carryover, or adstock. Ignoring it is one of the most common reasons brand channels look worse than they are.

How carryover works

When someone sees an ad, the effect doesn't all arrive at once. Some people act straight away. Others remember the brand next week, or next month, when they're ready to buy. The effect of each week's spend is spread across the weeks that follow, fading as it goes.

Channels differ a lot. Paid search mostly works in the moment: someone searches, clicks and buys. TV, video and outdoor build familiarity that pays out over weeks. Mix models estimate how quickly each channel's effect fades.

Why short-window reporting gets it wrong

Most marketing reporting counts results in the same week, or within a short attribution window. That flatters channels that work immediately and penalises channels that work slowly.

A brand campaign judged on its first week can look like a poor investment, while a large share of its effect is still to come. Cut it on that basis, and the drop in sales arrives weeks later, when it's harder to trace back to the decision.

What to do about it

• Judge channels over the right period. For slow-burn channels, look at contribution over the weeks the effect lasts, not the week of spend.
• Don't switch channels on and off on short-term results. The data that would justify it hasn't arrived yet.
• Include carryover in plans. When you forecast next quarter, the spend you've already made is still working. A plan that ignores it will misjudge both the baseline and the effect of new spend.
• Ask for the half-life. "About half the effect has gone after three weeks" is a useful, plain number for each channel.

The bigger point

Carryover isn't a technicality. It changes which channels look good, how fast you can expect results, and what happens after a cut. If you judge a channel only in the week you paid for it, you're measuring the invoice, not the effect.