Measuring Marketing When Attribution Can't Tell the Whole Story

Most leadership teams meet the attribution problem in the same meeting. Google Ads claims a number of conversions, Meta claims a number of its own, the website analytics claims a third, and the CRM shows fewer real leads than any of them. Then someone proposes cutting the campaign that “doesn’t convert”, usually the one that built the awareness the others are harvesting.

Attribution is not useless. It is a narrow instrument, measuring something real but smaller than the question a CEO is asking. The question is not “which ad was clicked last?” but “what would we lose if we stopped spending this money?” This guide sets out why click-based attribution cannot answer that on its own, and gives a layered set of measures a mid-sized company can run without a data science team: business outcomes against a break-even line, simple tests that switch something off, and indicators of demand that no click report can see.

Why attribution over-credits the final touch

Attribution assigns credit for a conversion to the interactions that came before it. What it cannot do is tell you whether the conversion would have happened anyway. That gap is the whole problem.

Consider how a B2B purchase usually starts. Professor John Dawes of the Ehrenberg-Bass Institute set out the 95:5 rule in 2021: up to 95% of business clients are not in the market for many goods and services at any one time. His illustration is that corporations change service providers such as their principal bank or law firm around once every five years on average. Dawes calls the 95% a heuristic, not a precise rule, and the share in your category depends on how often it is bought. The implication still matters: the buyer who searches your company name today probably formed their shortlist months earlier, at an exhibition, through a colleague, a LinkedIn post or a video they never clicked. None of that appears in a click path. The branded search ad that collects the click gets the credit.

The tools themselves are also narrower than many teams assume:

• The menu of models has shrunk. Google Analytics now offers three attribution models: data-driven, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position-based models were removed in November 2023. Google Ads offers two models, last click and data-driven, with data-driven as the default for most conversion actions.
• Data-driven is still click-based. Google Analytics describes it as using your account’s data to calculate “the actual contribution of each click interaction”. It can share credit more fairly among the clicks it can see; it does not see the conversations, events and impressions that never produced a click.
• Direct visits are excluded. Google Analytics’ attribution models give direct visits no credit unless the whole path consists of direct visits. A buyer who types your web address after hearing about you at a trade show is credited to whatever tracked touch came before, or to nothing.
• Each platform counts by its own rules. Google Ads, for example, reports view-through conversions (people who saw a Display or video ad without clicking and converted later) in a separate column from clicked conversions, and cannot report them from browsers that block cross-site cookies. Other platforms apply their own windows and definitions, which is why platform totals often add up to more leads than the CRM contains.

Les Binet, writing for Think with Google in March 2025, puts the core point plainly: “Not every click translated to a genuine, incremental conversion.” Click- and last-touch-based reporting, he notes, offers a “readily available and easily digestible view of performance” but can miss much of the customer journey behind a conversion.

Layer one: business outcomes against a break-even line

Start with numbers the platforms cannot inflate: CRM-qualified leads, opportunities, won revenue and gross profit by source, as recorded by the sales team rather than by the ad platform, with forms, calls and WhatsApp clicks tracked as GA4 key events so the source survives the hand-over. Our guide to the monthly CEO marketing dashboard covers how to lay these out on one page. To that, add one question to your inquiry form and sales script: “How did you first hear about us?” Self-reported answers are imperfect, but they regularly surface exhibitions, referrals and LinkedIn content that click tracking misses, and they cost nothing.

Then judge each channel against a ceiling. Neil Krikul, a marketing scientist at the Ehrenberg-Bass Institute, argued in B&T in August 2026 that short-term performance spend should be assessed against “the maximum cost per acquisition/conversion they can afford to break even”. For a B2B company that sells through a sales team, the practical version is a maximum affordable cost per qualified lead: average gross profit per won deal, multiplied by the share of it you are willing to spend on acquisition, multiplied by your lead-to-deal close rate from at least two quarters of CRM data. Our article on setting your own brand and performance split uses the same ceiling to size the activation budget; here it is a test each channel has to pass.

Hypothetical example, figures invented for illustration: a supplier whose average won deal yields 40,000 in gross profit (in any currency), who is willing to spend half of that on winning it and who closes one in twenty qualified leads, has a ceiling of 40,000 × 50% × 5% = 1,000 per qualified lead. Any channel delivering qualified leads well under that line is a candidate for more budget; any channel above it needs a reason to continue, and “the platform says it converts” is not one.

This layer will not tell you which channel caused a sale. It tells you whether the overall system is profitable and where spend has clearly crossed the line.

Layer two: tests that switch something off

The only direct way to measure what a channel adds is to compare people or places that saw it with people or places that did not. This is what incrementality testing means, and it is less exotic than it sounds.

Platform lift studies

Google Ads’ Conversion Lift splits an audience into a group that sees your ads and a control group that does not; the difference in conversions is the lift. Google offers two versions:

The help page still lists Discovery alongside Demand Gen, so check the current campaign list with Google before planning a study.

Google states that Conversion Lift is not available for all accounts and that access is through a Google account representative; the required budget is set during setup from the campaigns’ historical conversion data. Its Experiment Center groups these studies with brand lift (ad recall, awareness, purchase intent) and search lift (whether searches for your brand rise). Meta’s lift study documentation describes a comparable design, with randomised test and control groups; it says Conversion Lift access is currently limited and comes through a Meta representative, and that breakdown results need at least 100 conversions across test and control combined. Many mid-sized B2B advertisers will not meet these thresholds, which is why the next option matters.

A do-it-yourself geographic holdout

If you sell across several comparable regions (say, a distributor active in Riyadh, Jeddah and the Eastern Province, or across several Emirates), you can run a simple test without any platform tool:

1. Pick one channel and one question: “Does our prospecting campaign on this platform add qualified leads?”
2. Choose matched regions with similar history. Keep spending in some; pause the channel in one or two. Record which is which before you start.
3. Run it long enough to cover your normal lead cycle (as a rule of thumb, at least six to eight weeks for most B2B companies) and hold everything else steady.
4. Compare the change in CRM-qualified leads (not platform conversions) in test and holdout regions against the same weeks before the test.

The result will be directional rather than precise: regions differ, and a single large tender can distort a month. But a channel whose removal changes nothing in the CRM has told you something no attribution report could. The IPA’s Making effectiveness work makes the same case: “There is no silver bullet to measure advertising effectiveness.” It notes that marketing mix modelling, experiments and attribution each have a role and limitations, that using them in isolation fragments what a company learns, and recommends formal learning agendas.

Layer three: demand you cannot click on

The longest-lasting effects of marketing show up as more people thinking of you when they enter the market. These need indicators of their own.

Search demand for your brand

Branded search volume is a useful proxy for whether more people know and look for you. Binet describes share of search – your brand’s share of searches across the brands in your category, tracked via Google Trends – as “a powerful indicator of brand health and future market performance”, noting that its movements often come before movements in market share. Read it with care. According to Google’s Trends FAQ, Trends uses a sample of searches, scales results from 0 to 100 rather than showing absolute volumes, adds statistical noise for privacy and shows low-volume terms as zero. For a niche B2B brand in one country, Trends may show nothing useful; in that case use the branded query data in Google Search Console and the branded keyword impressions in Google Ads, tracked month by month. Check both Arabic and English spellings of your name.

Brand health among the buyers who matter

Dawes recommends measuring mental availability: how strongly buyers link your brand with category entry points, the situations in which they need what you sell (a site handover, an inspection, a supplier failing to deliver). He notes that many well-established brands reach no higher than 20–30% of respondents linking them to a category entry point, and that even market leaders often get only 50%. A mid-sized company does not need a research agency to start. Twice a year, ask a consistent sample of prospects and clients, through your sales team or a short survey, three questions:

• When you think of [your category] in [your market], which companies come to mind? (unprompted)
• Which of these companies would you consider for [a specific buying situation]?
• Which of these logos or names do you recognise? (prompted)

Keep the wording, the sample type and the timing the same each round, so the trend means something even if the absolute numbers are rough.

Putting the layers together

No single layer settles an argument. Read together, they usually do.

A few patterns are worth recognising. If platform conversions rise but CRM-qualified leads do not, the platform is probably claiming conversions that were happening already, or counting low-quality ones. If a prospecting channel looks weak on attribution but a holdout shows qualified leads falling when it pauses, keep it and stop judging it on last click. If branded search and prompted recognition climb while cost per qualified lead stays under the ceiling, the long-term work is paying for itself even if no report can draw a straight line to a deal.

Building this kind of measurement plan, and the budget logic behind it, is part of RX’s marketing strategy work; the campaign-level tracking that feeds layer one sits within performance marketing.

Mistakes that undo the work

• Testing too briefly. A two-week pause in a market with a three-month sales cycle measures nothing except noise.
• Changing several things at once. If you pause a channel, change the offer and launch a new website in the same month, no layer can tell you what happened.
• Judging brand activity on last click. Awareness campaigns, exhibitions and executive LinkedIn content will always look poor in click reports (for shows, see how to measure exhibition ROI). Judge them on layers two and three, and set those expectations before the money is spent, not after.
• Letting each platform mark its own homework. Platform reports are for optimising within the platform. Budget decisions between channels should rest on the CRM and on tests.
• Treating the ceiling as fixed. Recalculate the maximum affordable cost per lead whenever margins, deal sizes or close rates change.

Start with the layer that costs nothing. This month, have sales record a source for every qualified lead in the CRM, add the “How did you first hear about us?” question, and calculate your ceiling per qualified lead. Bring those three to the next budget review before anyone opens a platform report.