Measurement
Part of Before you report on product photography measurement, fix your brief
Why product photography attribution methods shape your studio budget
A comparison of product photography attribution methods for English ecommerce teams, from last click and holdout tests to geo splits and blended models.
What to take away
- There is no single correct attribution method. The choice depends on whether you need speed, defensibility or both.
- Last click is cheap and fast but credits whatever the customer touched last, which is rarely the image.
- Holdout and geo tests cost more to run but produce evidence you can defend to a finance director.
- The ONS digital economy statistics show how much UK commerce now runs through online channels. That is the context for treating imagery as a measurable asset rather than a sunk cost.
- Whichever method you pick, fix the brief before you fix the numbers, as set out in our product photography measurement and reporting guide.
Why attribution is harder for images than for ads
A paid ad carries a click identifier. A product image does not. The customer sees it on a category page, a search result, a marketplace listing or a social post, then buys days later.
That gap is why teams argue about attribution. Marketing claims the image lifted conversion. Finance sees a studio invoice and no matching line of revenue. Both are looking at the same shoot.
The legal backdrop matters too. Under the Consumer Protection from Unfair Trading Regulations 2008, the rules on misleading commercial practices cover how products are presented, including imagery. Accurate attribution supports honest reporting about what that presentation achieves.
Version control sits alongside measurement. If nobody tracks which image files went live, a test can credit a lift to a file that never appeared on site. The business innovation, protection and cyber security guidance sets out the wider governance context for digital assets.
The four methods compared
Last click and first click
Last click assigns the sale to the final touchpoint. For product imagery this usually means the product page view, so every image refresh looks effective. It is quick to set up in any analytics package.
First click has the mirror problem. It credits the first page seen, often a category listing with dozens of images. Neither method isolates the image itself.
Holdout tests
You show the new images to a random share of visitors and keep the old ones for the rest. The conversion difference between the two groups is your effect. This is the cleanest method for a single product range.
It needs traffic volume. A page with 300 visits a week will not produce a readable result inside a quarter.
Geo and time splits
Run the new imagery in one region or period, the old in another, then compare. It suits multi-market sellers and seasonal ranges.
Seasonality is the trap. December imagery tested against January imagery measures the calendar, not the shoot.
Blended modelling
Blended models combine click data, sales data and media spend into a single estimate. They are useful for portfolio-level planning, not for judging one shoot. They also hide their assumptions unless you document them.
Before and after comparison
| Method | Setup effort | Evidence strength | Best use |
|---|---|---|---|
| Last click | Low | Weak for imagery | Quick sanity checks |
| First click | Low | Weak for imagery | Funnel entry analysis |
| Holdout test | Medium to high | Strong | Single range or page test |
| Geo or time split | Medium | Moderate | Multi-market rollouts |
| Blended model | High | Moderate at portfolio level | Annual budget planning |
Matching the method to the decision
If the decision is whether to reshoot one hero product, a holdout test answers it. If the decision is how much to spend on imagery next year, a blended model is more appropriate.
The mistake is using one method for both. Last click will always flatter a reshoot because the product page is where the sale completes.
For teams building ongoing reporting, the practical route is a dashboard that separates test results from routine metrics. Our guide to a product photography reporting dashboard covers the fields worth tracking.
Whichever route you take, document the assumptions before the test starts. A holdout needs a fixed window, a defined control group and a rule for what counts as a sale. Written down first, those choices stop a later argument about whether the result was cherry-picked.
Common questions
Can I use last click if I have no test budget?
Yes, provided you label it as directional. Last click shows whether a page converts, not whether the image caused the lift. Pair it with a holdout before you commit to a reshoot.
How long should a holdout test run?
Long enough to cover a full purchase cycle for the product. For considered purchases that can be four to six weeks.
Do I need consent to run image tests?
You are testing presentation, not collecting new personal data, so existing privacy notices usually cover it. Check with your data protection lead before changing tracking.
What if finance rejects modelled numbers?
Offer a holdout on one range as a calibration exercise. A single defensible result often settles the argument faster than a portfolio model.
