A performance review compares what happened with what you meant to happen, and with what you would have expected anyway. It has three parts.

Results against objectives. Take each objective from your plan and put the result beside it. Use the same measure and the same dates. A target of "net new customers" is not met by a result of "gross new customers". A simple table does the job:

Results against baselines. You set baselines in the KPIs topic: where the numbers stood before you started. A result only means something beside its baseline. Sales of 500 are good news if the baseline was 300 and bad news if it was 700.

What changed besides the marketing. Season, a price change, a competitor's offer, a stock shortage, even the weather can move results. List them. You can't always remove their effect, but you can say they were there. A review that credits marketing for everything is as unreliable as one that blames it for everything.

Say what missed, and why. If you missed a target, say so in the first line, with the size of the miss. Then give the reasons you can support with evidence, and the ones you only suspect, labelled as suspicions. People trust a review that opens with the bad news far more than one that buries it.

Keep the sums checkable. Every figure should be traceable to a source, and every total should add up. Check your own arithmetic before anyone else does.

Attribution means giving credit for a sale or sign-up to the marketing touchpoints that came before it. Most people meet more than one before they act: a social ad, then a search, then an email. Attribution is the rule for sharing out the credit.

The common models.

• Last click gives all the credit to the final touchpoint before the conversion.
• First click gives it all to the first.
• Position-based gives most to the first and last and a little to those between.
• Data-driven uses a platform's statistical model of customer paths to share out the credit. It is only as good as the data it can see.

The choice of model changes the answer. Google removed first click, linear, time decay and position-based attribution from Google Analytics 4 and Google Ads in 2023, leaving last click and data-driven. Check the current help pages.

Why every platform over-claims. Each platform counts the conversions it can see and takes credit for those it touched. Add up the claims and the total is often more than the sales you made. Nobody is lying. Each is measuring its own view of the path.

What it cannot show. Attribution follows people who clicked. It can't tell you:

• who would have bought anyway, which the next lesson tackles
• the effect of things people saw but did not click, such as a poster, a video or a friend's recommendation
• offline influence, such as a conversation in a shop
• how changing ways of finding answers affect clicks. When search tools answer questions directly, fewer people may click through, and the data can understate what that activity did

UK example: published effectiveness research. In April 2024 Thinkbox, the UK television industry's marketing body, published Profit Ability 2. Five agencies combined their econometric databases of client data, covering 141 brands, to measure profit payback over several time periods. Read the source for its findings, and note that it relies on modelling, not experiments, and an industry body commissioned it. Les Binet and Peter Field's The Long and the Short of It (IPA, 2013) reports that, on average across the cases studied, a mix of roughly 60% brand-building and 40% activation performed well. That ratio is an average from their data and varies by category, so treat it as a starting point for discussion, not a rule. Both matter here because effects that take time to build don't show up in a last-click report.

AI activity. Ask an AI assistant to explain why last-click and incremental results can differ for your campaign. Compare its answer with this lesson and your data, and note anything it states as fact that you cannot check.

Incrementality is the part of a result that would not have happened without the marketing. The rest would have happened anyway. Attribution tells you who was touched. Incrementality asks whether the touch made a difference. If you take one question from this topic, take that one: would it have happened anyway?

The calculation is simple. Incremental results = results with the marketing − what you expected without it. The hard part is the second number.

Three ways to find it.

1. A holdout group. Choose a group at random and deliberately don't show them the marketing. Compare their results with the group that saw it. This is the strongest method, because the two groups differ only in the marketing. The cost is that you forgo some sales from the held-out group, which you can often afford if the group is small.
2. A geographic or matched comparison. Run the marketing in some areas or shops and not in similar ones. It works when you cannot hold out individuals, but areas never match perfectly.
3. Before and after against a baseline. Compare the campaign period with your baseline from the KPIs topic. It needs no special set-up, so most organisations start here. Its weakness is that it assumes nothing else changed. If the season, price or a competitor shifted, the baseline no longer shows what would have happened.

State the weakness whichever you use. A figure that comes with its limitation is more useful than one that pretends to certainty.

Marketing mix modelling (MMM) is a statistical method that uses months or years of sales and spend data to estimate how much each activity, and factors such as price and season, contributed to sales. It suits organisations with several channels and a good run of data, and it does not rely on tracking individual people. It needs a lot of clean data and skilled interpretation. Open-source tools exist, such as Meta's Robyn and Google's Meridian, but both come from advertising platforms, so ask who built a model and what they gain from the answer. For a small organisation, a simple holdout or comparison is often the practical choice.

Designing a test. Write down before you start:

• the question: what are you trying to learn?
• the groups: who sees the marketing, who does not, and how you chose them
• the measure and the period
• what result would change your decision
• how you will tell if something else interfered

Marketing ROI is the return on marketing investment. This course measures it on contribution, not revenue, because revenue ignores the cost of delivering what you sold. Some organisations use revenue. When someone quotes an ROI, ask which they mean.

The steps.

1. Incremental customers = customers gained − customers expected anyway.
2. Value created = incremental customers × lifetime value per customer.
3. ROI = (value created − cost) ÷ cost.
4. Cost per incremental customer = cost ÷ incremental customers.

Pull back what you know. Before reading on, try to recall how lifetime value is worked out. It is contribution per month multiplied by expected lifetime, and expected lifetime is 1 ÷ monthly churn. You met it in Module 12. Lifetime value is a forecast. It is earned over time, not paid into the bank this quarter. So an ROI built on it is an expected return, not cash already received.

Compare with a target. The cost per incremental customer is the figure to hold against the target you set when you planned the budget. It is also worth comparing with lifetime value: the gap between the two is what each new customer is expected to earn after paying for their own acquisition.

Test the assumptions. An ROI depends on inputs you estimated. Change the important ones and see what happens. Lifetime value is usually the largest. Run the sum at the figure you expect and at a more cautious and a more optimistic one. If the ROI stays positive across them, you can say so with confidence. If it turns negative at a plausible lower value, say that, too.

Read the worked example first. It models every step for Larkfield. Then use the Campaign ROI and Incrementality Calculator with your own figures.

AI activity. Give an AI assistant your spend, results, expected results and lifetime value, and ask it to check your ROI arithmetic and explain any difference between last-click and incremental results. Then redo every step yourself. Note any step it got wrong or any number it introduced that you did not give it.