Last-click attribution is hiding your Microsoft Ads value

Most performance teams already know last-click is imperfect. What fewer have worked through is that its imperfection is not random. It is biased in a specific direction, and that direction systematically penalises smaller, earlier-funnel platforms. Microsoft is the platform most often on the wrong end of that bias, which is why so many teams conclude it does not work when their measurement was never built to see it working.

The volume bias in last-click

Last-click attribution awards 100% of the credit for a conversion to the final touch before it. It is simple, auditable and easy to defend in a meeting, which is exactly why it persists. But its simplicity encodes an assumption that does not hold: that the last interaction is the one that mattered.

In any multi-touch journey, the last touch is disproportionately likely to be the highest-volume, lowest-funnel channel. For most advertisers that is Google brand search, the place people go once they have already decided. The platforms that did the earlier work (that introduced the brand, framed the consideration, or reached the buyer during research) handed the baton along and then watched the channel that caught it take all the credit.

This is a structural bias, not a measurement error. Consider how it compounds against a platform like Microsoft:

• Microsoft often plays an earlier, research-phase role in a desktop-first, considered journey, so it is less likely to be the final click by design.
• The bigger platform’s brand campaigns harvest the last click on demand the smaller platform helped create.
• Once last-click numbers come in, budget flows to the apparent winner, which captures even more final clicks, which deepens the bias on the next read.

The result is a feedback loop that makes the highest-volume platform look ever more efficient and the assisting platforms look ever more disposable. The data appears to justify concentrating spend, when in fact the model was always going to produce that conclusion regardless of where the value was created. This is one of the quieter ways concentration risk gets built into an account: not by decision, but by a measurement default no one questioned.

Higher rates, better LTV

The bias matters more for Microsoft than for most channels because of who Microsoft reaches. If Microsoft delivered a near-identical audience to Google, undercounting it would cost you efficiency but not insight. It does not deliver an identical audience.

A meaningful share of Microsoft’s audience is higher-income and desktop-dominant. Add to that LinkedIn Profile Targeting (job title, company and industry), which is exclusive to Microsoft Advertising. Together these mean Microsoft frequently reaches a more senior, higher-value prospect than the blended average of your other channels.

That has a direct consequence for measurement. A higher-income, more senior customer is often worth more over their lifetime: a larger deal, a longer retention, a higher-tier product. A pure conversion-count view, even a perfect one, would still miss this, because it treats every conversion as equal. Last-click compounds the problem twice over: it undercounts Microsoft’s conversions and then values the few it does count at the same flat rate as everyone else’s.

So the platform that may be sourcing your best-fit, highest-value customers is the one your default reporting is most likely to dismiss. The combination of a structural undercount and a flat valuation of high-value customers is why Microsoft so often reads as marginal in exactly the accounts where it is quietly doing important work. The fix is not to argue with the dashboard. It is to change what the dashboard measures.

Incrementality as the fix

The honest answer to “what is Microsoft actually worth” is not a better attribution model. Every attribution model is a set of assumptions about how to divide credit after the fact, and you can argue about the weights forever. The answer is incrementality: a test that measures what changes when Microsoft is present versus absent.

The question incrementality answers is the only one that matters to a budget owner: if I turn this spend off, how many customers do I actually lose? Not how many last clicks, not how many assisted conversions on someone’s preferred model, but how many real, incremental customers the channel is responsible for.

In practice, this looks like:

• A controlled on/off or geo test. Hold Microsoft out in a comparable set of regions or audiences, run it in others, and measure the difference in total conversions and qualified pipeline, not just the conversions attributed to Microsoft itself.
• Watching the whole account, not just the channel. The signal of incrementality is what happens to your total results, including the conversions other channels claim. If pausing Microsoft quietly drags down your Google-attributed conversions too, you have just measured the assist that last-click hid.
• Running it long enough to clear the sales cycle. For considered purchases, a short test measures noise. Let it run past a full decision cycle so the earlier-funnel contribution has time to show up downstream.

Incrementality testing is more work than reading a dashboard, and that is precisely why it is worth doing. It replaces a structural bias with evidence. Our method builds incrementality into how we evaluate Microsoft rather than leaving it to a model that was always going to undercount the channel.

What to change in reporting

You do not need to throw out last-click entirely. It remains a useful, stable operational metric, and finance teams are right to like that it is auditable. The change is to stop treating it as the verdict on a channel’s worth and surround it with measurement that sees what it cannot.

Practical steps, in order of effort:

• Add an assisted-conversion view. Even within your existing tools, look at how often Microsoft appears anywhere in converting paths, not just at the end. It is the cheapest way to start seeing the assist.
• Adopt a data-driven or multi-touch model alongside last-click, and read them side by side. Where they disagree most is where the volume bias is doing the most damage, and that gap is information.
• Carry reporting through to value, by source. Track conversions to revenue, retention or lifetime value, so a channel that sources higher-value customers is not flattened to a conversion count.
• Commit to one incrementality test before you cut a channel. Make it a standing rule: no platform gets defunded on last-click numbers alone. A single controlled test changes more minds than any amount of model debate.

The deeper point is that measurement is not a neutral lens. The model you choose decides which channels look valuable, and a model chosen for its convenience will quietly reward the platforms that suit its assumptions. Last-click suits the highest-volume platform. If you want to know what Microsoft is really contributing, you have to measure in a way that can actually see an earlier-funnel, higher-value channel. Most teams have simply never given it the chance. If you want help designing a test that does, get in touch.