Last-touch attribution is not a finding about your customers; it is an artifact of what your analytics tool could see, and every budget allocated on it inherits the tool’s blindness.
This matters acutely in India right now because Indian digital budgets are unusually performance-weighted. Direct-to-consumer brands, fintechs, and increasingly the e-commerce arms of legacy companies run monthly allocation reviews where the deciding table is a last-click report from a web analytics platform or an advertising platform’s own dashboard. Meanwhile the actual Indian purchase journey has never been messier or longer: a consumer sees a creator video, gets a WhatsApp nudge from a retailer, searches the brand name a week later, compares on a marketplace, and converts through a Unified Payments Interface (UPI) checkout — possibly offline. The measurement system credits the branded search click. The video, the nudge, and the marketplace visit vanish from the record, and next quarter’s budget follows the record, not the journey.
The mechanics of the artifact are worth stating precisely, because they have been documented since the earliest enterprise web analytics literature and they have not changed. Traditional web analytics is built largely on single-session clickstream data: when a conversion occurs, the last marketing touchpoint of that session — an email click, a search ad — receives full credit, with no consideration of the behaviours and touchpoints that reached the customer before it. But most conversions occur only after multiple days and numerous website visits or other interactions. Those two facts together mean last-touch is not a simplification of the truth; it is a systematic transfer of credit from the channels that create demand to the channels that harvest it. Branded search, remarketing, and transactional email sit closest to the conversion and therefore look heroic. Display, online video, social, and content sit early and look worthless. The report is internally consistent, precise to two decimals, and wrong in one direction, every month, forever.
Across a decade of deployments, the pattern repeats in three predictable failures. First, the closer channels saturate: the brand keeps feeding budget into branded search and retargeting until it is largely paying to intercept people who had already decided, a fact that only an incrementality test ever reveals. Second, the opener channels get cut, and six to nine months later the closer channels quietly get more expensive, because there is less demand arriving to harvest — and nobody connects the two events, because the attribution report cannot express the connection. Third, double counting: each platform’s own dashboard claims full credit for the same conversion, so the sum of channel-reported conversions exceeds the business’s actual sales, sometimes embarrassingly. Consistent attribution logic applied across all initiatives in one place is the unglamorous fix, and it is still rare.
The channels most damaged by last-touch are the ones that work without a click at all. Display and social advertising do a large share of their work at the impression level — a consumer sees the ad, does not click, and arrives days later by typing the brand name. Measuring these channels on click-through is measuring a swimming coach by how often she personally touches the water. The established alternative is impression-level attribution: connecting ad exposure to subsequent site behaviour and conversion, so that view-through influence becomes visible and comparable. Where user-level exposure data is no longer available — an accelerating reality under cookie deprecation and India’s Digital Personal Data Protection (DPDP) Act — the same question is answered in aggregate by mix modelling and geo-experiments. Either way, the principle stands: judge impression media on attributed outcomes, never on clicks.
The dual lens sharpens the point. The brand voice argues that upper-funnel work builds memory and preference that no click trail will ever contain — and the brand voice is right, which is exactly why it should refuse to be judged by a click trail. The evidence voice replies that “trust me, it builds the brand” is not a budget argument either; the honest middle ground has existed for years in the form of multiple attribution methods run side by side, statistically inferred response attribution that apportions credit across touches more rationally than any single-touch rule, and — the gold standard — holdout experiments that measure what happens when a channel is simply switched off for a matched population. When a channel’s inferred contribution and its experimental lift both round to zero, cut it without sentiment. When last-click says zero but the experiment says otherwise, you have found money the dashboard was hiding.
There is also a governance observation from practice: the attribution rule is a budget-allocation policy wearing the costume of a neutral measurement. Whoever selects the rule selects the winners. Leaving that selection to the default setting of an analytics tool — or worse, to each media platform’s self-graded homework — is a delegation of capital-allocation authority that no board would tolerate in any other function. It deserves a named owner, a written rationale, and an annual review, like any other financial policy.
On Monday, do three things. First, ask for last quarter’s conversions restated under two additional rules — first-touch and any statistically inferred multi-touch method your stack supports — in one table beside last-click, and look only at the budget deltas; the size of the swing is the size of your current decision risk. Second, mandate that every impression-led channel is reported on attributed conversions and experimental lift, with click-through rate demoted to a diagnostic. Third, schedule one genuine holdout this quarter on your most confident channel — the one everybody “knows” works. If the lift is there, you have defended the spend with evidence for the first time. If it is not, you have just funded the openers the last-click report has been starving for years.