Marketing measurement for small businesses: what you can and cannot measure on a small budget

The most common measurement request I get from a business is a return on investment for each marketing channel, on a budget of around AED 50,000 or £10,000 a month spread across five or six channels. It is a reasonable thing to ask for. At that size of budget, spread that widely, it cannot be measured reliably. This article explains why, using the clearest guide I know, and then sets out what a small business can measure if the spending is planned with measurement in mind.

The question most businesses ask

"Which channel is working?" sounds simple. To answer it, you have to be able to see the effect of each channel on sales, separately from everything else that affects sales. That is harder than it looks. Sales go up and down every week for reasons that have nothing to do with marketing: the weather, the end of the month when salaries are paid, a competitor's offer, school holidays, Ramadan, a news story. A marketing campaign has to change sales by more than those normal ups and downs, and for long enough, before anyone can be confident the campaign caused the change. If a small budget is spread across many channels in short bursts, the effect of each channel is smaller than the normal ups and downs of the business, and it cannot be separated from them.

What econometrics and marketing mix modelling are

Econometrics, in marketing, means using statistics to work out what caused past sales. The more common name today is marketing mix modelling. The method takes several years of weekly sales figures and lines them up against everything that might have affected them: advertising in each channel, price changes, promotions, weather, seasons, competitor activity and so on. It then works out how much of each week's sales would have happened anyway, and how much each of those factors added or took away. The result is a breakdown of sales by cause, and from that a return for each marketing channel.

The Institute of Practitioners in Advertising publishes a free guide to the method, Econometrics Explained 2, written by the econometrician Louise Cook. It was published in 2017, before the current generation of modelling software, but the principles have not changed and it is the clearest guide I know. Two of its lessons matter for any small business, and neither requires a model to apply.

Why a small campaign cannot be seen in the sales figures

Think of the normal week-to-week ups and downs in your sales as background noise. A model measures that noise and gives it a figure, called the standard error. Cook's rule is that a campaign which runs for one week has to lift sales by roughly twice that figure before anyone can say with confidence that the campaign did it, rather than the normal ups and downs.

Her example is a large grocery brand whose weekly sales vary by about 2 per cent. A one-week campaign would need to lift that brand's sales by 4 per cent to be seen. A campaign whose effect lasts ten weeks would only need to lift sales by about 1.2 per cent a week, because a small effect repeated over ten weeks adds up to something visible. In her words, "small, brief impacts are least likely to achieve statistical significance."

A small business's sales vary far more than a national grocery brand's. If your weekly sales normally move by 10 per cent, a one-week campaign would have to lift sales by about 20 per cent before it could be seen in the figures. Very few channels do that on a small budget. The ones that do are usually price promotions, and the lift comes from the price cut rather than the advertising. This is why a small budget spread across five channels in short bursts cannot be measured channel by channel. The effects are too small and too brief to be told apart from normal variation.

Why advertising keeps working after it stops

The guide also explains that advertising keeps having an effect after the week it runs. Modellers call this carry-over, or adstock. If a week of advertising lifts sales by 10 per cent and the carry-over rate is 80 per cent, the following week still has an 8 per cent lift from that advertising, the week after that 6.4 per cent, and so on. At 80 per cent carry-over, the effect has mostly gone after six months. At 90 per cent, it lasts about a year.

Two practical consequences. Cutting spend just before a busy season loses returns you have already paid for, because the effect of earlier advertising is still arriving. And a monthly report is too short a period to judge most marketing, because much of the effect of this month's spend appears in later months.

What a modelling project costs and needs

Cook is clear about what a proper modelling project requires. Projects "cost from tens to hundreds of thousands of pounds", which in dirhams means well into six figures. Most take several weeks at minimum, and large ones three to six months. The usual requirement is three years of weekly sales data, more for businesses with strong seasonal patterns. A model can be built on less, but "the risk of inaccurate quantifications rises as the data sample shortens". Her summary: "Good and cheap won't be fast. Fast and good won't be cheap. Fast and cheap won't be good."

Two more limits matter for a small business. A channel whose spend has stayed the same for years cannot be measured, because there is no change for the model to compare against. And channels that always run at the same time cannot be told apart, so if you switch on search advertising and social advertising in the same weeks every time, no model can say which one worked. Both problems can be avoided by planning the spend differently, and that is covered in the last two sections.

Businesses are spending more and proving less

The IPA Bellwether for the second quarter of 2026, published on 16 July, found that more UK companies were increasing their marketing budgets than cutting them, by a margin of 6.9 percentage points. Events led every category, at plus 11.0 per cent, for the second quarter running. Over the same period, companies became less confident about their own financial prospects, with a margin of minus 9.6 per cent compared with plus 0.6 per cent in the first quarter. Businesses were less confident about their outlook and were still spending more on marketing. That only makes sense if the spending can be shown to work.

The BoF-McKinsey State of Fashion 2026 shows what happens at a larger scale when it cannot. Across the 400 listed companies in the McKinsey Global Fashion Index, revenue grew 1 per cent in 2024 while operating costs rose 3 per cent, "largely due to rising marketing spend as brands competed for more value-conscious consumers". Economic profit fell 12 per cent. An entire industry spent more on marketing and made less money, and few of the companies involved would have been able to say which part of the extra spending was the mistake.

This is the pattern I described in five signs your spend is not driving growth. Reporting activity is easy. Showing what the activity added to the business needs the approach in the rest of this article.

Give a range, and agree how it will be read

Michael Kaminsky, writing in Harvard Business Review in the January to February 2026 issue, argues that marketing directors have among the shortest tenures in senior management because marketing cannot answer the question finance asks: if we spend more next quarter, how much comes back, and how confident are you. His answer is to give a range instead of a single figure. Rather than "this channel returns 3.1 times what we spend", say "most likely between 2.2 and 3.8 times, on current data".

I agree with the method and I use it. There is a practical problem he does not mention. Finance departments tend to plan on the lower figure. Present 2.2 to 3.8 and you will usually be funded as if the answer were 2.2, while a colleague who presents a single figure of 3.1 is funded at 3.1. Ranges work when the finance director has agreed in advance how they will be read. Have that conversation privately before you present one.

A second idea in Kaminsky's article is rarely used. Ask which channels you are least sure about, then spend some budget specifically to find out, rather than to generate revenue. On a small budget this is the most useful thing you can do, because it turns the planning problem described above into a plan: change the spend on one channel, on its own, for long enough to see the effect.

Customer acquisition cost depends on how you define it

Daniel McCarthy and Eva Ascarza's Harvard Business School note Measuring Customer Acquisition Cost (June 2026) makes a point that every business with an investor presentation should read carefully. Customer acquisition cost, or CAC, is the amount you spend to win one new customer. It looks like a simple sum: marketing spend divided by new customers. In practice both halves of the sum involve judgement. What counts as a new customer depends on your business. Which costs count as acquisition costs depends on how your marketing is run. Spending this month wins customers in later months, so the timing has to be matched. And there are four different versions, blended, by channel, by cohort and marginal, which answer different questions and are often confused. The note lists five common mistakes: leaving out relevant costs, including costs that belong to retention or general overheads, counting people as customers before they have bought anything, ignoring timing, and assuming a low CAC is automatically good.

A worked example in dirhams. A business spends AED 60,000 a month on paid advertising and wins 40 new paying customers. CAC on advertising alone is AED 1,500. Add the AED 25,000 monthly cost of the person and the agency who manage the advertising and it is AED 2,125. Count the 200 people who signed up for a free trial, instead of the 40 who paid, and it is AED 300, which is the figure that tends to appear in the investor presentation. All three are called CAC. Only the second will stand up to questioning, and it is the one that looks worst.

The practical rule is to write your definition down once and keep to it. State what counts as a customer, which costs are included, which period the spend is matched to, and which of the four versions you are reporting. Then report it next to what a customer is worth over time and how long it takes to earn back the cost of winning them. A CAC figure on its own tells nobody whether growth is worth having. The same rule separates a marketing return on investment figure a board trusts from one it merely accepts.

A simple way to separate marketing sales from normal sales

You do not need a statistical model to start separating the sales that would have happened anyway from the sales that marketing produced. A simple version takes an afternoon in a spreadsheet.

• Put your weekly sales in a chart, going back as far as you have them. Mark any weeks with no paid marketing at all. Sales in those weeks are your first estimate of what happens without marketing.
• Mark the seasonal periods. Ramadan, Eid, summer, Christmas, back to school, whatever affects your business. Compare similar weeks with similar weeks.
• Add your spend by channel and by week. Where a channel was switched on and then off, look at what sales did in the following weeks, allowing for carry-over. Where a channel has never changed, accept that you cannot see its effect and plan a test.
• Run one clean test each quarter. One channel, switched on for six to eight weeks, then off for the same period, with nothing else changing. Or, if you sell in more than one city or region, on in one and off in the other. This creates the change a model would need, and it gives you a readable answer on its own.
• Report the split. Normal sales, additional sales from marketing, and what the additional sales cost. Even at low precision, this is a better conversation with a board than a dashboard of channel returns that add up to more than total revenue.

In my experience the normal level of sales is usually much higher than the business owner expects, and finding that out changes the budget conversation from "how much should we spend" to "what should we spend it on". If you want a structure for the report, the board-level KPI dashboard is built for it.

What to do instead

Use fewer channels. Run each one for longer, and at different times, so they do not overlap. Report three numbers with fixed definitions every month, and do not change the definitions: enquiries or pipeline created, cost per qualified lead, and return on marketing spend on whichever definition you have written down. Put one brand measure next to one commercial measure on the same page for four quarters. Keep a split of normal sales and marketing sales, however simple. Run one clean test each quarter, designed to find out about the channel you understand least.

Then, when you have two or three years of weekly data, a budget that justifies a six-figure project and a decision that needs that level of precision, commission a model. Until then, be clear about what you can and cannot measure, and plan the spending so that what you can measure is worth knowing. That is most of what budget optimisation means at this scale, and it is the work a growth and performance mandate is for. When I built the marketing function at IFZA, the first quarter was almost entirely definitions and baselines, and every measurable result afterwards depended on that work.

Sources. Louise Cook, Econometrics Explained 2, Institute of Practitioners in Advertising, October 2017. IPA Bellwether Report Q2 2026, 16 July 2026. Michael Kaminsky, "When CMOs and CFOs Align Their KPIs, They Deliver More Value", Harvard Business Review, January to February 2026. Daniel McCarthy and Eva Ascarza, Measuring Customer Acquisition Cost, Harvard Business School, June 2026. The Business of Fashion and McKinsey, The State of Fashion 2026, McKinsey Global Fashion Index chapter, November 2025.