Before you start
You will be able to
- Calculate reach, average frequency, gross rating points, CPM, CTR, conversion rate and CPA from campaign data.
- Distinguish ROAS from marketing ROI and show how a campaign with a high ROAS can lose money.
- Explain incrementality and design a holdout or A/B test that measures it.
- Identify the bias in last-click attribution and in other correlational evidence from marketing data.
- Choose a small set of dashboard metrics that match a plan’s objectives.
- Evaluate a marketing mix for consistency across product, price, place and promotion.
Review first
- The promotion mix and integrated marketing communications — You need the promotion tools and the objective-and-task budget whose results this lesson measures.
- Distribution channels and decisions — Lantern’s gross margin per can in each channel is the margin that turns attributed revenue into profit.
- Pricing approaches — Marketing ROI is built on contribution, not revenue.
Fictional case A fictional coffee business founded by Ana Lim. It starts as a catering and barista-training service, opens its own cafés and an online shop for beans, and plans a canned cold brew.
The idea
Fictional case
Six weeks after launch, three reports land on Ana Lim’s desk (all figures hypothetical). The agency’s says the social and search ads delivered 6.2 million impressions with a click-through rate of 0.9%. A supermarket chain’s report says Lantern’s ads in its app returned $4.19 of sales for every $1 spent. The sampling team reports 38,000 tastings, and supermarket sales of Lantern are rising. Ana’s question is simple: which of this money worked? Every report is accurate, and none of them answers it.
Marketing metrics answer four different questions, and most mistakes come from using a number from one question to answer another. Did people see it? Exposure metrics count impressions, people reached and how often. Did they respond? Response metrics count clicks, sign-ups and purchases. Did it pay? Return metrics compare what came back with what was spent. And the hardest: did the promotion cause it? That needs a comparison with what would have happened without it.
Did people see it? Impressions, reach, frequency and GRPs
An impression is one display of an ad: one person seeing one ad once, or the same person seeing it twice counts as two. Digital impressions need care. The Media Rating Council, the US body that accredits audience measurement, distinguishes a served impression, counted when the ad is delivered, from a viewable one; its guidelines count a display ad as viewable only if at least half of its pixels are on screen for at least one continuous second (two seconds of play for video). Many served impressions are never seen.
Reach is the number of different people exposed at least once, often given as a percentage of the target audience. Frequency (average frequency) is the average number of times each reached person was exposed. Gross rating points (GRPs) combine the two into one measure of weight:
Suppose a hypothetical campaign targets Harbour City’s 400,000 adults aged 18 to 44 and delivers 1,200,000 impressions to them, reaching 240,000 different people. Reach is , average frequency is , and the campaign delivered GRPs. The same figure comes from impressions as a percentage of the audience: . GRPs count exposures, not people: 300 GRPs could mean everyone seeing the ad three times or 30% of people seeing it ten times, which are very different campaigns.
The usual cost measure for exposure is CPM, the cost per thousand impressions (M is the Roman numeral for a thousand). If those 1,200,000 impressions cost $9,600, the CPM is dollars. CPM lets a planner compare media, but a cheap impression nobody notices is not a bargain.
Did they respond? Clicks, conversions and cost per acquisition
Digital media add response metrics. CTR, the click-through rate, is clicks divided by impressions. CPC, the cost per click, is spend divided by clicks. Conversion rate is the share of clicks (or of visits; definitions vary, so always check) that end in the action the campaign wants, such as an order or a sign-up. CPA, the cost per acquisition or cost per action, is spend divided by conversions. The four are linked:
The identity shows the three levers that lower the cost of a customer: cheaper impressions, more clicks per impression and more conversions per click. It also shows why chasing one of them can mislead: an ad that wins cheap clicks from people who never buy has a good CTR and a terrible CPA.
Did it pay? ROAS and marketing ROI
ROAS, return on ad spend, is the revenue attributed to a campaign divided by what it cost. It is quoted as a ratio (4.0, or 4:1) or a percentage (400%). Marketing ROI compares the profit the campaign added with its cost:
Incremental contribution is the extra revenue the campaign caused, minus the variable costs of delivering it (in Lantern’s case cost of goods, trade margins and fulfilment); it is the contribution margin idea from the pricing lessons applied to a campaign. Variants exist: some companies use gross profit, some include fixed costs created by the campaign, and some count expected repeat purchases over a customer’s lifetime. Whichever version you use, say which, and keep it the same when you compare campaigns.
In depth
Incrementality: what would have happened anyway?
Every return metric above depends on one assumption: that the sales credited to the campaign were caused by it. Incrementality is the part of an outcome that would not have happened without the marketing. The rest would have happened anyway: loyal customers who were going to buy, shoppers who already had the can in their basket, people searching for Lantern by name who would have found it without an ad.
This is not a theoretical worry. In large field experiments at eBay, Blake, Nosko and Tadelis (2015) found that paid search ads on the company’s own brand name had no measurable short-term benefit (people searching for a company by name can reach it through the free search results without the ad), and that the returns from paid search were a fraction of what the usual non-experimental estimates suggested.
The only reliable way to measure incrementality is to create a comparison group that didn’t get the marketing, chosen so that it is otherwise the same:
- Holdout (control) group: withhold the ad, email or offer from a randomly chosen holdout group and compare its behaviour with the exposed group. The difference is the incremental effect.
- A/B test: randomly split people (or stores, or weeks) between two versions, such as two messages or two discount levels, and compare outcomes.
- Geographic experiments and test markets: run the campaign in some areas or store groups and not in comparable others. A test market launches a whole plan in a limited area before rolling it out.
Random assignment is the key. If Lantern sampled only in its twenty busiest supermarkets and compared them with the rest, the difference would mix the effect of sampling with the effect of being a busy store. Randomising which stores are sampled makes the two groups alike on average, so the remaining difference can be credited to the sampling.
Experiments have limits too. Sales vary a great deal from person to person and week to week, while the effect of one campaign is usually small. Lewis and Rao (2015), analysing 25 large online advertising experiments with US retailers and brokerages, found that measuring returns precisely could need samples of more than ten million person-weeks, and that the median confidence interval on ROI was over 100 percentage points wide. Good tests need planning: enough stores or people, a long enough period, and a decision rule set before the results arrive.
Attribution and its biases
Attribution is the set of rules that divides credit for a sale among the marketing touchpoints that preceded it. Suppose a customer sees a Lantern social ad, later reads a food blog’s review after the launch event, then searches “Lantern cold brew”, clicks a search ad and orders a $41.88 12-pack online.
| Touchpoint | Last click | First click | Linear (equal shares) |
|---|---|---|---|
| Social ad | $0.00 | $41.88 | $13.96 |
| Food-blog review (PR) | $0.00 | $0.00 | $13.96 |
| Search ad | $41.88 | $0.00 | $13.96 |
PR coverage usually earns no clicks to track, so click-based rules rarely credit it at all.
Last-click attribution gives all the credit to the final click. It is simple, and it systematically favours the touchpoints closest to the purchase, such as search ads on the brand’s own name and retargeting ads shown to people who already visited the shop, which often capture demand that other activity created. Data-driven models, which share credit according to patterns in past conversion paths, are now common; Google Ads, for example, offers only last-click and data-driven attribution, having retired its first-click, linear, time-decay and position-based models. But every attribution model, however sophisticated, divides up credit for sales that happened. None of them can say which sales would have happened without any marketing. Gordon and colleagues (2019) compared 15 Facebook advertising experiments with the observational methods commonly used in the industry and found the observational methods often failed to reproduce the experimental results.
A dashboard: a few numbers that matter
A marketing dashboard is a short, regular report of the measures that show whether the plan is working. Its value comes from what it leaves out. Each number should be tied to an objective in the plan, defined the same way every period, and paired with a target. Leading indicators (awareness, distribution, trial) warn early; lagging ones (repeat purchase, contribution) say whether the plan is paying.
| Measure | Question it answers | Source and frequency |
|---|---|---|
| Awareness among category buyers | Is the brand known? (objective: 12% to 30%) | Survey, quarterly |
| Stores stocking the can | Can shoppers find it? | Distributor data, monthly |
| Trial (tastings, first purchases) | Are people trying it? | Sampling logs, online orders, monthly |
| Repeat purchase rate | Do triers come back? | Online-shop and loyalty data, quarterly |
| Sales per store per week, by channel | Is it earning its shelf? | Retailer sales data, weekly |
| Share of units sold on promotion and average price paid | Are deals eroding the price position? | Retailer data, monthly |
| Contribution per can and in total | Is the plan paying? | Lantern’s accounts, monthly |
| Incremental ROI of each vehicle | Which spending works? | Holdout tests, when run |
What isn’t on the list matters too. Impressions, likes and followers are easy to grow and easy to report, but they belong on the dashboard only when they are tied to an objective, for example impressions as a check that the awareness campaign is running at the planned weight.
Bringing the mix together
The course began with the idea that the four Ps must fit each other and the target market. Measurement is where a misfit shows: a premium product that sells only on discount, ads that send people to stores that don’t stock the can, a range whose listing fees eat the margin. A consistency check asks one question for each pair of decisions.
| Fit | Question | Lantern’s first plan |
|---|---|---|
| Product and price | Does the price match the quality relative to rivals? | Single-origin (taste score 7.5) at $3.49, between Arcadia (6, $2.89) and Summit Brew (8, $3.79). |
| Price and place | Does every channel leave enough margin at this price? | Gross margin of $1.42 to $2.30 a can across the chosen channels. |
| Product and place | Do the range and the channels’ fees fit? | One flavour keeps supermarket and natural-grocer fees at $11,000 a year; three flavours would cost $33,000 plus $12,000 of production cost. |
| Place and promotion | Does promotion send people where the product is sold? | Sampling and geo-targeted ads only around stocking stores. |
| Promotion and price | Do deals undercut the price position? | Trade promotions limited to four windows of about 20% off. |
| Promotion and product | Does the message match what the product delivers? | “Single-origin, steeped 20 hours”: a claim the product supports. |
The simulator and the real world
The Marketing Mix Simulator turns a plan into simulated sales, contribution and share using a model with stated assumptions and invented coefficients. It is good for what models are good for: comparing plans on the same basis, seeing which decisions matter most, and finding trade-offs you hadn’t noticed, such as a trade-promotion budget that adds units but lowers contribution. It is not a forecast. Real shoppers, competitors and retailers respond in ways no teaching model captures: a rival cuts its price, a buyer refuses a listing, a heatwave doubles summer sales. A real launch would test its key assumptions before committing the whole budget, with a test market in part of the city, store-level experiments for sampling and A/B tests for messages, and would keep measuring after launch.
Formulas
Average frequency
- Average frequency: times each reached person was exposed
- Impressions delivered to the target audienceimpressions
- Reach: different people exposed at least oncepeople
Gross rating points
- Reach as a percentage of the target audience%
- Average frequency
- Impressions delivered to the target audienceimpressions
- Size of the target audiencepeople
GRPs measure the weight of a schedule, not how many people saw it.
Cost per thousand impressions
- Cost of the media$
- Impressions boughtimpressions
Click-through rate
- Clicks on the adclicks
- Impressions of the adimpressions
Conversion rate
- Completed target actions, such as ordersconversions
Some reports divide by visits or sessions instead of clicks; check the definition.
Cost per acquisition
- Cost of each conversion$
Return on ad spend
- Revenue credited to the campaign by the attribution rule in use$
Assumes. Says nothing about costs or about whether the revenue was caused by the ads.
Marketing ROI
- Incremental contribution: extra revenue caused by the marketing minus its variable costs$
- Marketing cost$
Variants use gross profit or include lifetime value; state which you use.
Break-even ROAS
- Contribution margin ratio on the revenue counted in ROAS
Assumes. All attributed revenue is incremental; if it isn’t, the true return is lower.
Worked examples
Fictional figures
Lantern spends $6,000 on social ads for a 12-can pack in its online shop, priced at $3.49 a can ($41.88 a pack). The ads deliver 750,000 impressions and 7,500 clicks, and 240 people order a pack. Each can costs $0.74 to make and $0.45 to fulfil. Calculate CPM, CTR, CPC, conversion rate, CPA, ROAS and marketing ROI, assuming for now that every order was caused by the ads.
- Exposure cost: dollars.
- Response: ; dollars; .
- Cost per order: dollars. Check with the identity: .
- Revenue and ROAS: , so .
- Contribution per pack: ; total . ROI .
Answer. CPM $8.00, CTR 1.0%, CPC $0.80, conversion rate 3.2%, CPA $25.00, ROAS 1.68 and marketing ROI 10.4%, if every order was incremental.
What it means. The direct channel keeps about 66% of each order as contribution, so the break-even ROAS is only , and a ROAS of 1.68 is just profitable. The result hangs on the assumption in the last step. If a holdout showed that 40% of these orders came from café regulars who would have ordered anyway, incremental contribution would be and ROI would be about −33.8%.
Fictional figures
Lantern spends $10,000 on ads in a supermarket chain’s app. The chain reports 12,000 cans of Lantern bought by shoppers who saw the ads, at the $3.49 shelf price, and a ROAS of 4.19. A holdout test (a random group of the chain’s app users who were not shown the ads) later shows that only 40% of those cans were incremental. Lantern nets $2.37 a can from supermarkets and earns $1.63 of gross margin (single-origin, $0.74 cost of goods). What was the campaign’s ROI?
- Reproduce the reported figure: of shelf sales, and . The report counts sales at the shelf price, which is what the chain’s tills record.
- Lantern’s own revenue is the net price: , a ROAS of about 2.85 on what Lantern actually received.
- Contribution if every can were incremental: , an ROI of . On shelf sales the break-even ROAS is , and 4.19 is well above it.
- Apply incrementality: only cans were caused by the ads. Incremental contribution is .
- True ROI: .
Answer. The campaign’s marketing ROI was about −21.6%: it returned $7,839.36 of incremental contribution for $10,000 of spending, despite a reported ROAS of 4.19.
What it means. Three things separate the healthy-looking 4.19 from the loss: the report counted revenue at the shelf price rather than Lantern’s net price, it ignored the cost of the cans, and most of the attributed sales would have happened anyway, probably because the chain’s app users who saw Lantern’s ads were already likely buyers. The fix is not to ignore ROAS but to translate it: into Lantern’s revenue, then into contribution, then into incremental contribution.
Fictional figures
To test sampling, Lantern takes 40 supermarkets that stock the can and randomly assigns 20 to get two demonstration days each in a month and 20 to get none. Each demonstration day costs $225. Over the four weeks, sampled stores sell an average of 300 cans each and the others 220 (hypothetical results). For the following eight weeks, with no more sampling, the sampled stores keep selling 30 more cans a week each than the others. Gross margin is $1.6332 a can. Did the sampling pay?
- Cost: dollars.
- Incremental cans in the sampling month: , worth of contribution. On the month alone, ROI .
- Carry-over: more cans in the next eight weeks, worth .
- Total incremental contribution: ; ROI over twelve weeks .
- Check the design: because the stores were assigned at random, the two groups are alike on average in size and shoppers, so the difference can be credited to sampling. Had Lantern sampled its busiest stores, part of the gap would have been store size.
Answer. Judged on the sampling month alone, ROI was about −71%; counting the eight weeks of repeat purchases it was about +16.1%.
What it means. Sampling works by creating triers who come back, so the measurement window decides the verdict. A manager who judged sampling on the month’s sales would have cut the tool that was working. The test also gives Lantern a number to plan with: about $9,000 of sampling bought about $10,450 of contribution in twelve weeks, a basis for deciding whether to extend sampling to more stores and for checking that the effect holds as it does.
Explore
Common mistakes
“A ROAS above 1 means the campaign made money.”
ROAS compares revenue with spend, but the business keeps only the contribution from that revenue. With a 20% contribution margin a campaign needs a ROAS of 5 just to break even. Convert ROAS to contribution, then to incremental contribution, before judging it.
“The channel that got the last click caused the sale.”
Last-click attribution credits the final touchpoint, often a brand-name search ad or a retargeting ad that caught someone already on the way to buying. Experiments such as eBay’s brand-search test found such ads had no measurable short-term benefit. Credit is not causation; test with holdouts.
“Sales rose after the campaign, so the campaign worked.”
Something else may explain the rise: the season (the category sells about 19% more in the third quarter than in the second), a competitor’s stock-out, new stores stocking the product. A before-and-after comparison can’t separate these; a comparison with a randomly chosen group that didn’t get the campaign can.
“Impressions tell you how many people saw the ad.”
Impressions count displays, including repeats to the same person and served ads that never appeared on screen. Reach counts people; GRPs count weight. Report the measure that answers your question, and ask whether impressions are served or viewable.
“A higher click-through rate means a better ad.”
CTR measures clicks, not customers. An ad that draws curious clicks from people who never buy can have a high CTR and a high CPA. Judge ads by the outcome the campaign exists for, ideally the incremental one.
“The simulator tells us what will happen.”
A simulator applies stated assumptions consistently; that is why it is useful for comparing plans. Its coefficients are invented for teaching, and real markets react in ways no model captures. Use it to explore and compare, then test the key assumptions in the real world before committing the budget.
Guided practice
Work through these with hints when you need them. Every hint and every worked solution is there to be used.
Loading four questions…
On your own
No hints this time. Answer, check, and read the feedback; the solution appears once you have answered or if you ask for it. Your first answers count toward mastery.
Loading six questions…
In practice
At Lantern’s quarterly review the marketing manager doesn’t present every number the agencies sent. She presents the dashboard, the tests that ran and what they changed. Suppose (hypothetically) the brand-name search ads show the highest last-click ROAS of any spending, but a two-week holdout in which half of searchers saw no ad found almost no difference in orders. The manager moves most of that money into sampling in more stocking stores, where the randomised test above showed a positive return over twelve weeks, and keeps a small, continuing holdout to check the decision. The budget hasn’t grown; it has moved from where credit was claimed to where effect was shown.
The same habits pay off in other industries. An airline that emails a fare sale to its whole loyalty list should hold out a random slice of members to learn how many of the bookings would have come anyway. An online retailer whose retargeting ads follow people who left items in their basket should test how many of those people would have come back on their own. A university that runs open-day ads can compare applications from regions where the ads ran with similar regions where they didn’t. In each case the professional skill is the same: name the question, pick the metric that answers it, and build in a comparison that separates cause from coincidence.
Finally, the mix is judged as a whole. A plan with an excellent ROI on its ads can still fail if the product is in the wrong channels or priced against its positioning. That is why Lantern’s review ends with the consistency checklist and the contribution line, and why the course’s project asks you to defend a complete plan, not a single campaign.
Knowledge check
Six questions, one attempt each, no hints. Score 80% or more to complete the lesson; you can take it again at any time.
Loading six questions…
Summary
- Separate four questions: did people see it (impressions, reach, frequency, GRPs, CPM), did they respond (CTR, conversion rate, CPA), did it pay (ROAS, ROI) and did the marketing cause it (incrementality).
- GRPs = reach (%) × average frequency = impressions ÷ target audience × 100; CPM = spend ÷ impressions × 1,000; CPA = spend ÷ conversions.
- ROAS divides attributed revenue by ad spend; marketing ROI divides incremental contribution minus cost by cost. Break-even ROAS is 1 ÷ contribution margin ratio, so a high ROAS can still mean a loss.
- Incrementality is what would not have happened without the marketing; only comparisons with randomly chosen holdout or control groups, A/B tests and test markets measure it reliably.
- Attribution models divide credit among touchpoints; last-click favours the touchpoints nearest the sale, and no attribution model measures causation.
- Correlations in marketing data are often driven by store size, seasonality or targeting; ask what else differs before crediting the marketing.
- A good dashboard has a few measures, each tied to an objective with a target, mixing leading and lagging indicators.
- Check the whole mix pair by pair for consistency, use the simulator to compare plans, and test key assumptions in the real world before committing.
Key formulas
- Average frequency
- Gross rating points
- Cost per thousand impressions
- Click-through rate
- Conversion rate
- Cost per acquisition
- Return on ad spend
- Marketing ROI
- Break-even ROAS
Terms introduced
- Impressions
- The number of times an ad is displayed. Repeat views by the same person count separately; a served impression is counted on delivery, a viewable one only if it could actually be seen.
- Reach
- The number or percentage of different people in the target audience exposed to a campaign at least once.
- Frequency
- The average number of times each person reached was exposed to the campaign: impressions ÷ reach.
- Gross rating points
- A measure of the weight of a campaign: reach (as a percentage of the target audience) × average frequency, or impressions as a percentage of the audience. Abbreviated GRPs.
- CPM
- Cost per thousand impressions: spend ÷ impressions × 1,000.
- CTR
- Click-through rate: clicks ÷ impressions.
- CPC
- Cost per click: spend ÷ clicks.
- Conversion rate
- The percentage of visits, or of ad clicks, that end in a chosen action such as an order or a sign-up. Which denominator is used varies between tools and reports, so always state it.
- CPA
- Cost per acquisition (or per action): spend ÷ conversions.
- ROAS
- Return on ad spend: revenue attributed to a campaign ÷ the campaign’s ad spend. It ignores costs and whether the revenue was caused by the ads.
- Marketing ROI
- Return on marketing investment: (incremental contribution − marketing cost) ÷ marketing cost. Variants use gross profit or lifetime value.
- Break-even ROAS
- The ROAS at which a campaign’s contribution just covers its cost: 1 ÷ the contribution margin ratio.
- Incrementality
- The part of an outcome, such as sales, that was caused by the marketing and would not have happened without it.
- Holdout group
- A randomly chosen group deliberately not exposed to a campaign, used as the comparison for measuring its incremental effect.
- Attribution
- Rules for dividing credit for a conversion among the marketing touchpoints that preceded it, such as last click, first click, linear or data-driven.
- Test market
- Launching a product or marketing plan in a limited area to measure results before a full roll-out.
Sources
- Bendle, N. T., Farris, P. W., Pfeifer, P. E. & Reibstein, D. J. (2020). Marketing Metrics (4th ed.), chapters 10, “Advertising and Sponsorship Metrics”, 11, “Online, Email, and Mobile Metrics”, and 12, “Marketing and Finance”. Pearson FT Press. pearson.com/en-gb/subject-catalog/p/marketing-metrics/P20000…
- Media Rating Council (2015). Viewable Ad Impression Measurement Guidelines, version 2.0, 18 August 2015, with the IAB. mediaratingcouncil.org/sites/default/files/Standards/081815%…
- Blake, T., Nosko, C. & Tadelis, S. (2015). Consumer heterogeneity and paid search effectiveness: a large-scale field experiment. Econometrica, 83(1), 155–174. Working-paper version: NBER Working Paper 20171. nber.org/papers/w20171
- Gordon, B. R., Zettelmeyer, F., Bhargava, N. & Chapsky, D. (2019). A comparison of approaches to advertising measurement: evidence from big field experiments at Facebook. Marketing Science, 38(2), 193–225. ideas.repec.org/a/inm/ormksc/v38y2019i2p193-225.html
- Lewis, R. A. & Rao, J. M. (2015). The unfavorable economics of measuring the returns to advertising. Quarterly Journal of Economics, 130(4), 1941–1973. ideas.repec.org/a/oup/qjecon/v130y2015i4p1941-1973.html
- Google Ads Help. “About attribution models” (models currently offered and those retired). Accessed 22 October 2026. support.google.com/google-ads/answer/6259715