Practitioner · Lesson 5 of 5

Worked example: cutting programmatic CPA by 15%

About 10 minutes · Yogender Kumar, Managing Partner

By the end of this lesson you'll be able to

  • Run a placement review that finds where the budget is wasted
  • Check tracking, device and attribution before cutting anything
  • Plan changes so you can judge the result fairly

Why this matters

This one is real, from a campaign we worked on. It shows how a campaign can look healthy on clicks while a quarter of its budget buys almost nothing.

The core idea

A placement review lines up spend against sales, one site or app at a time. Averages hide waste. The detail shows it.

Before acting on what you find, check three things that can make a good placement look bad:

  1. Is tracking firing on that placement?
  2. Is one device type dragging the numbers down?
  3. Is the attribution window, the period after a click in which a purchase is credited to the ad, hiding sales that come later?

Three checks before you cut a placement

Before excluding anything

Is tracking firing?
Is one device skewing it?
Is the window hiding later sales?
Then cut
Any of the three can make a good placement look bad. Cut only once all three are clear.

Worked example

This is a real campaign, not an illustration: a £10,000 display campaign for an ecommerce brand.

The campaign brought 50 purchases, so its cost per purchase was £200. Clicks were cheap, and for a while it looked fine.

The placement review told a different story. A quarter of the budget, £2,500, had gone on casual-gaming apps and websites nobody spent any time on. Those placements produced plenty of cheap clicks and 2 of the 50 purchases.

  • Weak placements: £2,500 ÷ 2 = £1,250 per purchase.
  • Everything else: £7,500 ÷ 48 = about £156 per purchase.

Before cutting anything, we ran the three checks. Tracking was firing on those placements. No device type was hiding good results. A longer attribution window didn't change the picture.

Then we made four changes together:

  • Excluded the weakest placements.
  • Switched to an approved list of relevant sites and apps, so new low-quality inventory couldn't creep back in.
  • Capped frequency at three impressions per user per day, where the inventory supported it.
  • Kept a device review running, to make sure no device type was dragging results down.

We kept the offer and the creative the same through the next review period, so the comparison was fair. We also kept an eye on seasonality.

Cost per purchase fell from £200 to £170, a 15% reduction. Because several changes went in at once, we credit the whole set, not any single one.

What we'd do differently. We gave cheap clicks too much weight at the start. A low cost per click made those placements look efficient while they used up a quarter of the budget.

Now we look at placement-level engagement and approved sales in the first weeks, before cheap traffic has had time to spend much money.

Cost per purchase in the real campaign

Weak placements: £2,500 for 2 purchases

£1,250

Everything else: £7,500 for 48

£156

Whole campaign: £10,000 for 50

£200
The £200 average hid two very different groups: a quarter of the budget at £1,250 a purchase, and the rest at about £156.

In practice

  • Run a placement review in the first few weeks, then monthly.
  • Line up spend against purchases for every placement, not clicks.
  • Check tracking, device and attribution before you exclude anything.
  • Keep the offer and creative steady while you judge the changes.
  • Move to an approved list once you know where sales come from.

Common mistakes

  1. Trusting cheap clicks. Casual-gaming apps produce a lot of accidental taps. Judge placements by purchases.
  2. Cutting before checking. A tracking fault can make a good placement look dead. Run the three checks first.
  3. Changing everything, then crediting one thing. If four changes go in together, report them as one result.

Try it yourself

A £12,000 campaign brings 60 purchases. A review finds £3,000 spent on placements that brought 3 of them. What's the cost per purchase on those placements, and on everything else?

Show answer

Weak placements: £3,000 ÷ 3 = £1,000 per purchase. Everything else: £9,000 ÷ 57 = £157.89 per purchase. The campaign as a whole is £12,000 ÷ 60 = £200, which hides both.

Quick check

Not scored. Pick an answer to see the explanation.

  1. In this case, why did the weak placements look fine at first?

  2. What should you check before excluding a placement that shows no sales?

Key takeaways

  • Averages hide waste. Review spend against sales placement by placement.
  • Check tracking, device and attribution before you cut.
  • Keep the offer and creative steady while you judge changes.
  • Credit a set of changes made together as one result.

Further reading: the full guide on our blog