Measuring marketing with a baseline

Leo divided 248,000 UAH of new-client revenue by 20,000 UAH of webinar cost and got 12.4. The arithmetic is right; the question is wrong, in two ways.

1. Revenue isn't what Klasna keeps. Of each 800 UAH lesson, the tutor gets 430 and video and AI cost 31, so Klasna keeps 339 UAH, its contribution per lesson. A campaign has to pay for itself out of that, not out of the full price.
2. It credits every lesson to the webinars. Some of those clients would have come anyway: through referrals, the founders' contacts, the season or plain luck.

So the honest question isn't "how much did we sell after the campaign?" but "how much more did we sell because of it, and what did that extra earn?" Both mistakes can be fixed, which means a campaign in a long sale can be measured, just not with one division.

The baseline is what the metric would have been without the campaign. You can't observe it, so you estimate it, and two tools do most of the work.

• The same months last year take care of the season. June–August 2025: new clients in Kyiv and Lviv bought 200 lessons. Both numbers include the summer dip.
• A control group takes care of the trend. The webinars ran only in Kyiv and Lviv; in the other cities, with no webinars, new-client lessons grew 15%. The market moved without any help.

Combine them: baseline = last year × (1 + the control's growth) = 200 × 1.15 = 230 lessons. In 2026 the two cities bought 310, so the webinars get credit for 310 − 230 = 80 lessons.

Using last year alone (310 − 200 = 110) would hand the campaign the market's own growth. Only what's above the baseline is marketing's.

Extra lessons are not yet money. To judge a campaign, turn them into what Klasna actually keeps: the incremental margin.

Incremental margin = extra lessons × contribution per lesson.

Contribution is the price minus the costs that come with each lesson: 800 − 430 (tutor) − 31 (video and AI) = 339 UAH. For the webinars: 80 × 339 = 27,120 UAH.

Why not revenue? Because 80 × 800 = 64,000 UAH would count the tutors' pay and delivery costs as if Klasna earned them. A campaign that "brings 64,000" might not even cover its own cost once those are paid.

Two cautions: use only the lessons above the baseline (80, not 310), and use the contribution of the product actually sold. If a campaign brings clients on a discounted price, their contribution per lesson is lower, and so is the margin.

ROMI answers one question: for each hryvnia spent on a campaign, how much did it earn back above its cost?

ROMI = (incremental margin − spend) ÷ spend.

For the webinars: incremental margin 27,120 UAH, spend 20,000 UAH. After paying for themselves they earned 27,120 − 20,000 = 7,120 UAH, so ROMI = 7,120 ÷ 20,000 = 35.6%.

Read it like this: above 0% the campaign earned more than it cost; below 0% it lost money; exactly 0% it broke even.

Compare the versions: on revenue Leo got "12.4 times"; without the control group (110 extra lessons) ROMI would be (110 × 339 − 20,000) ÷ 20,000 = 86.45%; with both corrections it's 35.6%. Every shortcut makes the number bigger and less true. A modest positive ROMI with its working is a result a board can trust.

A control group is only useful if it shows what would have happened to the campaign's audience without the campaign. Two conditions make that true:

1. Comparable. The control should behave like the treated group: similar cities, similar clients, the same months. Cities without the webinars worked for Kyiv and Lviv because new-client sales there follow the same market.
2. Untouched. Nobody in the control should get the campaign. If half the control cities' HR managers joined the webinars online, their growth includes the campaign's effect, and the baseline is too high.

That's why the control is chosen before the campaign. Picking it afterwards invites choosing the cities that grew least, which makes any campaign look good.

For a stronger answer, assign cities to campaign and control at random and plan the test in advance; the Product Analytics course shows how to set up and read such an experiment.

Two traps make campaign numbers in B2B look better or worse than they are.

Timing. A company that meets Klasna at a webinar in May may sign in September, after a pilot and procurement. If you judge the campaign by leads or deals in the month it ran, you see almost nothing, and you may stop a campaign that's working. Measure when the deals land, and say in advance how long you'll wait.

Noise. Small numbers swing a lot by chance. 80 extra lessons could easily have been 50 or 110 with a couple of clients signing a month earlier or later. One campaign on a small sample is a hint, not proof.

So report a range or a caveat with the result ("about 80 lessons, a small sample"), and treat a promising result as a reason to repeat it as a planned test, not as a law of nature.

Measurement is a plan made before the money is spent, not a calculation after it.

1. Before the campaign: choose the metric (new-client lessons) and a control group that won't get the campaign.
2. During: run it only where planned and record its full cost, including people's time and any discounts.
3. When deals land: build the baseline from the same months last year, adjusted by the control's change.
4. Then: extra lessons above the baseline × 339 UAH = incremental margin; subtract the spend; ROMI = result ÷ spend.
5. Report: the result with its working and its uncertainty.

The order matters because each step depends on the one before. A control chosen after the fact may not be untouched; a cost recorded later tends to shrink; a ROMI without its working can't be checked. Planned this way, even a small team gets an answer it can defend.

The last step is the sentence the board will hear. An honest report gives three things:

• the working: the baseline (last year adjusted by the control cities), the extra lessons, the margin after spend;
• the result: about 80 extra lessons, 7,120 UAH after the webinars' cost, a ROMI of 35.6%;
• the caveat and the next step: a small, noisy sample, so repeat the webinars as a planned test.

Avoid both extremes. "12.4 times their cost" or "growth of 55%, all thanks to the webinars" credits what would have happened anyway. "Marketing can't be measured in a long sale, so we'll keep spending" gives up on a question that has a reasonable answer. A board trusts a modest number with its working more than a big one without it.

How it fits together: baseline from last year and a control, incremental margin, ROMI, timing and noise, reported with its working. Next lesson, the module's case: a whole quarter, five channels and 325,000 UAH, and where next quarter's money should go.