Marketing attribution explained: models, myths and what to actually use

The comfortable lie of last-click

Last-click attribution is the default in most analytics setups, and it is comfortable precisely because it is simple - whichever touchpoint immediately preceded the conversion gets all the credit. It is also deeply misleading. By definition it rewards the bottom of the funnel: branded search, retargeting, the email someone clicked because they were already convinced. It gives zero credit to the content that introduced your brand, the social campaign that built awareness, the webinar that earned trust weeks earlier. The predictable consequence is that teams over-invest in capturing existing demand and under-invest in creating it, then wonder why growth stalls. Last-click makes branded search look like a genius channel when often it is just intercepting people who were going to find you anyway. The danger is not that last-click is wrong - every model is wrong in some way - but that it is wrong in a consistent, seductive direction that flatters harvesting and starves demand generation. If your reporting runs on last-click alone, you are almost certainly mispricing your channels and slowly cannibalising the top of your funnel.

The model zoo and why each one bends the truth

Rule-based attribution models all try to spread credit across the customer journey, and each makes a different assumption that is sometimes right and often arbitrary. First-touch credits the channel that started the relationship, which flatters awareness and ignores everything that closed the deal. Last-touch does the reverse. Linear splits credit evenly across every touchpoint, which is fair-sounding but pretends a passing impression mattered as much as a sales demo. Time-decay weights touches nearer the conversion more heavily, reasonable for short cycles but punishing for long B2B journeys. Position-based, often 40-20-40, rewards the first and last touch and thins out the middle. None is correct, because none knows the true causal weight of each interaction - they are accounting conventions, not measurements of influence. Their value is comparative: viewing the same campaign through several models reveals how dependent your conclusions are on the model you picked. If a channel looks great under last-click and terrible under first-touch, that disagreement is the actual insight. Pick a model deliberately for the decision at hand, and never mistake its tidy output for ground truth.

• First-touch - credits discovery, ignores what closed the deal
• Last-touch - credits the finish, ignores what created demand
• Linear - fair-looking, but treats every touch as equal
• Time-decay - favours recent touches; weak for long cycles
• Position-based - rewards first and last, thins the middle

Three lenses: multi-touch, MMM and incrementality

Beyond the simple models sit three serious approaches, and mature teams use them together rather than picking one. Multi-touch attribution stitches together individual user journeys across touchpoints - granular and tactical, but increasingly crippled by privacy loss and cookie deprecation, since you can no longer reliably follow one person across devices and platforms. Marketing mix modelling, or MMM, takes the opposite, top-down view: it statistically correlates aggregate spend and external factors against aggregate outcomes over time, needing no user-level tracking, which makes it privacy-resilient and ideal for understanding channels you cannot click-track, like offline or brand. Its weakness is that it is coarse and needs substantial historical data. Incrementality testing is the closest thing to truth: you deliberately withhold or vary spend for a test group and measure the causal lift versus a control - geo holdouts, conversion lift studies, switch-back tests. It answers the only question that matters - what would have happened anyway? - but it costs time and forgone spend to run. MTA tells you the how, MMM tells you the how much across channels, and incrementality tells you whether any of it was real.

Just ask the customer

The most underrated attribution method costs almost nothing: ask people how they heard about you. A single "How did you find us?" field at signup or checkout captures signal no tracking pixel ever will - the podcast mention, the friend's recommendation, the conference talk, the half-remembered article. These dark-social and word-of-mouth touches are invisible to your analytics yet often drive a meaningful share of your best customers. Self-reported attribution is messy: people misremember, they name the last thing rather than the first, and the data is qualitative. But it is precisely strongest where digital tracking is weakest, which makes it a perfect complement rather than a competitor. The trick is to treat it as a directional signal, not a precise number - if a channel your pixels barely register keeps showing up in customers' own words, that is a strong hint you are under-crediting it. Many B2B companies discover their highest-LTV customers came through routes no dashboard tracked at all. Add the question, read the answers in aggregate, and you recover a layer of truth that the post-cookie web threw away.

• Captures dark social and word-of-mouth pixels miss entirely
• Strongest exactly where digital tracking is weakest
• Treat as directional signal, not a precise number
• Often reveals where your highest-LTV customers really come from

Triangulate with blended measurement

Because every method is flawed in a different direction, the sane response is to stop searching for the one true model and start triangulating. Blended, or holistic, measurement means looking at several imperfect signals together and trusting the conclusion where they agree. Start with the blunt, honest top-line metric: blended CAC - total marketing and sales spend divided by total new customers - which no platform can inflate because it ignores their self-serving attribution entirely. Layer in platform-reported numbers for tactical optimisation, knowing each over-claims. Add MMM for the strategic channel-mix picture, incrementality tests to validate the channels you are betting big on, and self-reported data to catch what the rest miss. When these independent lenses point the same way, act with confidence; when they conflict, you have found exactly where to investigate or run a test. This is less elegant than a single dashboard number, and some stakeholders will resist the ambiguity. But it is honest, and honest beats precise-and-wrong every time. The leaders who measure well are comfortable holding several partial truths at once and deciding anyway.

Get the data foundations right first

No attribution method works on broken data, and most attribution problems are really data-hygiene problems wearing a costume. The foundations are unglamorous but decisive. Disciplined, consistent UTM tagging on every campaign link is non-negotiable - inconsistent naming turns your reports into noise, so standardise the conventions and enforce them. Server-side tracking is increasingly essential as browsers block client-side scripts and cookies expire fast; moving tracking to your own server recovers signal you would otherwise lose to ad blockers and privacy defaults. And your CRM must be connected to your marketing data, because attribution that stops at the lead is useless - you need to follow each lead through to closed revenue to know which channels produce customers, not just clicks. A lead source captured at signup and carried through to the deal record is worth more than any clever model applied to bad data. Invest here before buying another attribution tool. The brands with trustworthy measurement are rarely the ones with the fanciest software; they are the ones whose tagging, tracking and CRM actually agree with each other.

• Standardise and enforce UTM conventions across every channel
• Move to server-side tracking to survive cookie and ad-block loss
• Connect marketing data to the CRM, lead through to revenue
• Capture lead source at signup and carry it to the deal record

What mix to actually use, by company size

The right approach scales with your spend and complexity, so match the method to your stage rather than copying an enterprise playbook. Early-stage and smaller businesses should keep it simple: blended CAC as the north star, clean UTMs, last-click for tactical decisions while knowing its bias, and a self-reported "how did you hear about us" field that punches far above its cost. You do not have the volume for MMM or the spend to justify elaborate incrementality programmes yet. Mid-sized companies with meaningful budgets across several channels should add CRM-connected reporting, view campaigns through multiple attribution models to stress-test conclusions, and begin simple geo holdout tests on their largest channels to check what is incremental. Large advertisers with substantial, multi-channel spend warrant the full stack - MMM for strategic allocation, a regular cadence of incrementality experiments, server-side data infrastructure, and blended metrics tying it together. The mistake at every size is buying capability you cannot yet feed with data or act on. Spend should pull sophistication, not the reverse. Start with the basics done well, and add rigour as the budget at stake justifies the effort.

Decide well under uncertainty

The hardest skill in attribution is not statistical - it is psychological: getting comfortable acting on incomplete, conflicting information without freezing. Some leaders respond to imperfect data by demanding ever more precision, delaying decisions while the market moves on. The better response is to accept that all your numbers are estimates, weight them by how trustworthy each is, and decide anyway, then learn from the result. Lean on the metrics hardest to game - blended CAC, payback period, validated incremental lift - and treat platform-reported ROAS as directional rather than gospel. When models disagree, that is a signal to run a small test, not a reason to stall. Build a culture that treats measurement as a way to reduce uncertainty rather than eliminate it, and that rewards good decisions made on reasonable evidence even when the evidence was murky. The marketers who win in the post-cookie era are not the ones who found a magic perfect-attribution tool - it does not exist. They are the ones who triangulate honestly, hold ambiguity without panic, and keep moving.

What good measurement is worth

Attribution done honestly is not an accounting exercise - it is how you allocate budget toward what actually grows the business, and that is one of the highest-leverage decisions a marketing leader makes. When you stop trusting last-click blindly and start triangulating, you typically discover you were starving the channels that create demand to overfund the ones that merely harvest it. Reallocating against a truer picture routinely improves blended CAC and accelerates growth without spending an extra rupee, because the same budget simply works harder. It also changes how you are perceived internally: a leader who can explain, with credible blended metrics and the occasional clean incrementality result, why the mix is what it is earns the trust and the budget that vague platform screenshots never will. The payoff is compounding - better measurement leads to better allocation, which produces better results, which justifies more investment. You will never have perfect attribution, and you do not need it. You need a measurement practice honest enough to point you in the right direction more often than not, and the confidence to act on it. That is what turns marketing from a cost the CFO tolerates into a growth engine the business funds.