What Are Qualified Future Conversions? QFC, Explained — and the Baseline to Capture First.

Sometime in the next few budget cycles, a CFO is going to approve a Q3 reallocation justified by conversions that haven't happened yet. Most won't know that's what they approved.

At Google Marketing Live 2026, Google announced Qualified Future Conversions — QFC — and most of the coverage filed it under measurement housekeeping, somewhere below Asset Studio's video models and the AI Brief. That filing is wrong. QFC is the first time a major ad platform has moved from estimating conversions that already happened to crediting conversions that haven't happened yet — and it's scheduled to flow into Meridian, the open-source MMM that was supposed to be the independent check on the spend. I made the structural argument about that collision last week. This piece does the other half of the job: what QFC actually is, where it genuinely helps, where the line is — and the one thing to capture before the integration ships, because it cannot be captured after.

What exactly is a Qualified Future Conversion?

A Qualified Future Conversion is a Gemini-generated forecast that links today's ad spend to a sale expected later, inferred from early intent signals — a brand search this week, an engaged visit that didn't convert — with no custom tagging required (Search Engine Land). Run a brand-awareness campaign that produces zero immediate purchases, and QFC will report the conversions Google's models expect that campaign to produce next month, surfaced as measurable value today.

Three properties matter for how you should read it. First, it's a calibrated probability, not an invoice — Google is not inventing sales, it's forecasting them, and the early-signal research underneath (branded search lift as a leading indicator of purchase) is legitimate. Second, it's zero-friction by design — no tagging, no implementation project, which means adoption will be near-universal by default. Third, and this is the property the trade coverage keeps missing: it travels. QFC isn't staying in a reporting column. Google's stated direction is integration with Meridian — which is itself being folded into Google Analytics 360. The forecast is en route to the measurement layer.

QFC vs. modeled conversions: past tense vs. future tense

The reflex dismissal — I've heard it from smart operators already — is that platforms have reported "made up" conversions for years: consent-mode modeling, view-through, engaged-view, cross-device estimation. If you've accepted modeled conversions, the argument goes, QFC is just more of the same. It isn't, and the difference is grammatical.

Modeled conversions are past-tense estimation. Something probably happened — a purchase behind a consent wall, a cross-device journey the pixel lost — and the model fills in the gap. The event is real; the observation failed. Estimation error is bounded by reality, and it washes out as actuals arrive. GA4's defaults have plenty of problems, but they are problems of reconstructing a past that actually occurred.

Qualified Future Conversions are future-tense credit. Nothing has happened. The model is projecting forward from early signals and booking the projection as performance. There is no underlying event to bound the error — only the model's confidence in its own leading indicators. When the forecast is wrong, nothing arrives later to correct the record, because the record was the forecast.

What QFC gets right

Let me steelman this properly, because the legitimate version is genuinely good and pretending otherwise gets the whole argument dismissed.

Short-window, click-based attribution structurally undercounts upper-funnel work. A YouTube campaign that moves someone to search your brand three weeks later has, under last-click, produced nothing — which is exactly the accounting failure that pushed the industry toward MMM in the first place. QFC attacks that failure at the reporting layer: it gives delayed value a number, early, in the console where budget conversations actually happen. The leading-indicator science is real. Branded search volume is predictive of purchase. Engaged visits do convert at knowable cohort rates. If you run brand-heavy spend and have spent years losing budget fights to a last-click column, QFC will feel like justice — and partly, it is.

A Google measurement DS would add, fairly: the predictions are calibrated, the methodology will be published at least in outline, and Meridian remains open-source — you can inspect the priors and configure the inputs. All true. The problem was never honesty. The problem is architecture: you can audit the courtroom and still lose if the evidence is written by the defendant. That argument — separation of the optimizer from the evidence layer, the correlated-evidence failure mode, what Meridian-in-GA360 defaults will do at scale — is the spine of You Can't Let the Optimizer Own the Evidence Layer, and I won't repeat it here. This essay's job is narrower: given that QFC is coming and most accounts will take the default, what do you do this month?

The pre-QFC baseline: four numbers to freeze before the integration ships

The Meridian integration is announced but not yet universal. That gap is not a reason to relax — it's the entire opportunity. Right now, and only right now, your measurement still runs on observed outcomes. The moment future-tense credit starts flowing into reporting and then into the MMM's inputs, the clean "before" disappears, and with it your ability to ever quantify how much of platform-reported performance is forecast rather than observation. The pre-QFC baseline is four artifacts, frozen and timestamped:

One: actuals-only conversion-lag curves. For each campaign type, the empirical distribution of time-from-click (and time-from-impression where you have it) to observed conversion, built from your own server-side data. This is the curve QFC will quietly replace with a model. Freeze yours first, because the gap between your curve and the platform's forecast is the single most informative number you'll own next year.

Two: branded-search-to-purchase cohort rates. QFC's flagship leading indicator is the brand search. So measure, from your own data: of the people who searched your brand in a given week, what fraction purchased within 30, 60, 90 days — by cohort, by season, by campaign mix. When QFC starts crediting campaigns for brand-search lift, this table tells you what that lift has historically been worth in actual margin, not modeled expectation.

Three: a versioned, actuals-only MMM pass. Run Meridian (or your MMM of choice) now, on first-party outcome data only, and version the configuration — priors, inputs, exclusions — in writing. This is the reference model. When the QFC-fed version disagrees with it later, the disagreement isn't noise; it's the measured size of the forecast's thumb on the scale.

Four: your current incrementality results. Whatever geo-holdouts or lift tests you've run in the past year, collect the results and the methodology in one place, dated. Counterfactual measurement is the only evidence QFC can never generate — a forecast cannot tell you what would have happened without the spend. This file is the start of the independent evidence layer, and it's also the cheapest insurance you will ever buy against a board-meeting question you can't answer.

None of this requires new tooling. It's two weeks of analyst time and a decision to treat the current state of your measurement as an asset worth preserving. Decision latency usually measures the cost of acting slowly; this is the rarer case where the cost of slowness is permanent — the baseline isn't delayed by waiting, it's destroyed.

Should you opt in?

Yes — with the line drawn in advance, in writing. QFC as a reporting lens is a real correction to a real undercount; use it to defend upper-funnel budget the way it's designed to. QFC as optimization input is what optimizers are for; prediction is their native language, and refusing it there is performative. QFC as evidence — as an input to the MMM, the incrementality narrative, the number that walks into the board deck — is where the answer is no, permanently, for the reasons the evidence-layer essay lays out. The opt-in isn't one decision; it's three, and the teams that get hurt will be the ones who made all three with one checkbox.

If you hold that line, QFC is close to free upside. If you don't, you'll never see the moment it goes wrong — that's the nature of self-confirming measurement. The error doesn't announce itself; it just compounds politely inside numbers that keep agreeing with each other. This is the same architecture I flagged in The Sealed Auction: one prediction layer, more and more surfaces reporting through it, and the advertiser's view narrowing to what the layer chooses to show.

What this looks like from the CFO chair

Strip the measurement vocabulary and here's the transaction: the vendor selling you media will now also forecast the future revenue of that media, book the forecast as reported performance, and — once the Meridian integration lands — supply that forecast to the model your finance team uses to validate the spend. In any other procurement category, a CFO would reject that arrangement in the first sentence. It survives in marketing only because it arrives one convenience feature at a time, each defensible, none announced as what it adds up to. Cost Per Decision exists because marketing's credibility with finance depends on outcomes finance can verify; the P&L translation chain only holds if the numbers entering it are independent of the vendor they evaluate.

There is exactly one action this essay asks of you, and it has a deadline you can't see: freeze the pre-QFC baseline while there is still a "pre." The teams that do will spend next year knowing precisely how much of their reported performance is forecast — and negotiating with Google from that knowledge. The teams that don't will spend next year inside the forecast, with no way to know how far it sits from the floor. The feature is useful. The default is not your friend. The baseline is the difference.