Performance marketing that proves itself: attribution basics for non-marketers

Attribution is the practice of deciding which marketing activity deserves credit when someone buys, signs up or fills in a form. No attribution model is perfectly accurate, because people rarely follow a straight line from first ad to purchase. The goal is a consistent, honest method that lets you compare channels fairly and move budget towards what actually works.

Why attribution matters if you are not a marketer

If you approve marketing budget, attribution is the difference between a report that says "we generated 400 leads" and one that says "paid search generated 180 qualified leads that became 31 customers, and email brought back 40 lapsed ones". The first is activity. The second is evidence you can act on. Performance marketing is the discipline of spending against measurable outcomes, and attribution is how the measuring gets done.

The vocabulary you need

• Conversion. The outcome you care about: a purchase, a qualified lead, a booked call, a membership renewal. Define this first; everything else depends on it.
• Touchpoint. Any interaction before the conversion: an ad click, an organic search visit, an email open, a social post view.
• Channel. A grouping of touchpoints: paid search, paid social, organic search, email, direct, referral.
• Click-through versus view-through. Credit for a click is straightforward. Credit for an ad that was merely seen is where platforms get generous with themselves.
• Lookback window. How far back a platform looks for a touchpoint to credit. A seven-day window and a thirty-day window will report very different numbers for the same campaign.
• UTM parameters. Tags added to links so your analytics knows which campaign, channel and creative a visitor came from. Inconsistent tagging is the most common reason attribution reports cannot be trusted.
• First-party data. Information you collect directly, such as CRM records and consented email lists. As third-party cookies fade, this is what makes accurate attribution possible.

The main attribution models, in plain terms

An attribution model is simply a rule for splitting credit across the touchpoints that came before a conversion.

• Last click. All credit to the final touchpoint. Simple and widely used, and it systematically flatters branded search and direct traffic while starving the channels that created the demand.
• First click. All credit to the first touchpoint. Useful for understanding what introduces new people to you, useless for understanding what closes them.
• Linear. Equal credit to every touchpoint. Fair-minded but blunt.
• Time decay. More credit to touchpoints closer to the conversion. Sensible for longer sales cycles.
• Position-based. Most credit to the first and last touchpoints, the remainder shared in the middle. A reasonable default for many organisations.
• Data-driven. A statistical model estimates each touchpoint's contribution by comparing paths that converted with paths that did not. The most defensible option, but it needs enough conversion volume to be meaningful.

None of these is "right". Pick one as your reporting standard, state it on every report, and change it only deliberately.

Why the platforms disagree with each other

Ask three ad platforms how many sales they drove last month and the total will exceed your actual sales, sometimes by a wide margin. Each platform counts any conversion it touched, using its own lookback window and its own view-through rules, and none of them can see the others' touchpoints. Your analytics tool sees more of the picture but relies on tracking that consent banners, browser privacy features and ad blockers now interrupt, so parts of it are modelled rather than observed. Your CRM sees revenue but often not the source.

The honest approach is to treat platform numbers as directional, hold your analytics tool as the neutral referee, and reconcile against the CRM or sales ledger, which is the only place a sale definitively exists.

What "proving itself" actually looks like

Attribution assigns credit for conversions that happened. It cannot tell you whether those conversions would have happened anyway. For that you need incrementality testing, which asks the only question a budget holder truly cares about: what would we have lost if we had not spent this?

• Holdout tests. Pause a channel in some regions or for a random slice of the audience and compare outcomes with the rest. If sales in the paused group barely move, the channel was taking credit for demand it did not create.
• Conversion lift studies. The platform-run version of the same idea, available on the larger ad networks.
• Blended measures. Total marketing spend divided by total revenue, tracked monthly. Individual channels will always argue about credit; the blended number cannot be gamed.

A setup that works for most organisations

• Define two or three conversions that matter, and rank them. A newsletter sign-up is not worth the same as a qualified lead.
• Agree a UTM naming convention and enforce it on every link, including email and offline campaigns with QR codes.
• Move conversion tracking server-side where possible and collect consent properly so your data survives browser changes.
• Connect the CRM so you can attribute revenue and customer quality, not just lead counts.
• Choose a reporting model, most often position-based or data-driven, and keep the others visible for context.
• Run one holdout test a quarter on your largest channel.

Questions to ask whoever runs your marketing

• Which attribution model does this report use, and what would the numbers look like under a different one?
• Are these platform-reported conversions, or conversions from our analytics or CRM?
• What share of these conversions is view-through?
• When did we last run a holdout or lift test, and what did it show?
• Do we own the ad accounts, analytics property and tag manager, and can we log in today?

That last question matters more than it sounds. If an agency owns the accounts, you cannot audit the history and you cannot leave without losing it.

An anonymised example

A provincial professional association was running paid search, paid social and email for its annual membership drive and reporting on last click. Branded search appeared to produce most new members, so the board wanted to cut paid social. Before doing that, we switched the reporting standard to position-based and paused paid social in two of the province's regions for four weeks. New memberships in the paused regions fell 24% against the control regions, and branded search volume fell with them. Paid social had been creating the demand that branded search was collecting. The budget stayed, and the reporting changed.

Where OlDevs fits

OlDevs is a full-stack technology studio in Vancouver, British Columbia, building software and running performance marketing since 2014. Our marketing work covers SEO and content, paid search and shopping, paid social, email and lifecycle, analytics and conversion rate optimisation, and creative and media. Because we also build the websites and web apps that campaigns land on, tracking is designed in rather than bolted on.

Every client owns their ad accounts, analytics properties, data and creative from day one. Reporting is bilingual in English and French where needed, and campaign landing pages are built to WCAG 2.2 AA. We do not publish rates because the right plan depends on your conversions, your channels and your sales cycle. Describe what you are trying to grow and we will reply with a quote within one business day.