Qualified Future Conversions: Can You Trust the Prediction?

Over the next few weeks, three columns are going to show up in some Demand Gen reports, and one of them counts conversions that have not happened yet. That is the whole idea behind Qualified Future Conversions (QFC), the metric Google unveiled at Google Marketing Live 2026: an estimate, produced by Gemini, of the conversions a single ad interaction should generate up to 180 days later. Let me say it straight away: the idea is clever for upper-funnel campaigns, and it is also a perfect trap for anyone who steers a budget by the number on screen.

Because a QFC is not a sale. It is not even a measured conversion. It is a prediction. And a prediction has to be validated before you hand it any money. This guide does exactly that work: understand how the metric is calculated, separate what it promises from what it actually measures, and give you a four-step validation framework, BigQuery query included, to compare predicted against realized on your own data.

What Google announced at Google Marketing Live 2026

Let me set the scene without the hype. Qualified Future Conversions are a new metric announced at GML 2026. For a given ad interaction, it estimates the number of conversions expected over a window that can stretch up to 180 days. The intended use case is Demand Gen campaigns, and upper-funnel work more broadly, where conversions land long after the first touch and where classic reports, tuned to short windows, show a gap.

Three things are worth keeping in mind. First, Google stresses the “zero setup” angle: no extra tagging, no configuration, the metric is computed on the platform side from signals Google already holds. Second, it shows up as three columns in your campaign reports: QFC, cost per QFC, and QFC rate. Third, the rollout is very gradual: a restricted global pilot at the time of the announcement, with a wider beta planned for late 2026.

The most important point is the one that gets read the least. Google explicitly frames the QFC as a supplementary metric, not a replacement for conversion tracking. Ginny Marvin, on Google’s side, calls it a “directional signal.” Translation: it is a compass, not a cash register. In the same spirit, Google keeps pushing its causal measurement pieces, with Meridian now integrated into GA360 (paid tier only). The QFC lives at the very bottom of that stack, in the “estimate” layer, and that is where you should file it mentally.

As always with a rollout in progress, the right reflex is not to take my word for it: go and check whether the columns have appeared in your own interface, and on which campaigns.

How a Qualified Future Conversion is calculated (the “Leading User Action”)

The engine is Gemini. In concrete terms, the model looks through a user’s journey for what Google calls a Leading User Action: an action that, historically, precedes a conversion. A brand search after seeing a Demand Gen ad, an engaged site visit, a strong intent signal. The model learns from history how often that kind of action leads to a conversion in the following months, then extrapolates to the interaction at hand.

Why reach so far into the future? Because the upper funnel has a timing problem. According to figures Google put forward, only around 40% of Demand Gen conversions land within the first 30 days. Put differently, most of them arrive later, sometimes much later. A short conversion window therefore badly underestimates these campaigns, and that is exactly the void the QFC claims to fill by projecting what comes next.

The intent is defensible. But never lose sight of what you are looking at: a probabilistic model output, not an observation. The QFC answers the question “how many conversions should this interaction produce within 180 days, if the past repeats itself.” It is a conditional estimate, with everything that implies about assumptions on the stability of behavior, of the market, and of your own tracking.

What a QFC is not (and why that changes everything)

Here is the section not to skip. A Qualified Future Conversion is not a real conversion. It is not measured data. And it is certainly not an incremental sale. It is a prediction, and a prediction may never come true.

The trap is the same one I described for another recent Google metric: mistaking a number produced by the platform for a commercial reality. I unpacked that mechanism for the Data Strength Uplift, which measures recovered data and not sales. The QFC pushes the cursor one notch further: where the uplift reconstructs conversions that already happened, the QFC bets on conversions that have not happened yet. So the risk of misreading is higher, not lower.

Put yourself in the shoes of the busy reader of a report. They see “120 Qualified Future Conversions” on a campaign, they read “120 conversions,” and they allocate a budget on that basis. They have just steered on a black box. The number might be right, might be optimistic, might be calibrated on a history that no longer looks like your current quarter. Until you have cross-checked it, you simply do not know.

QFC, measured conversions, modeled conversions: do not conflate them

Three different objects now circulate in Google’s reports under the same word, “conversion.” They do not share a source, a reliability, or a legitimate use. The table below exists so you stop mixing them up.

The middle row is the one most often forgotten. A modeled conversion is still the estimate of a real fact that slipped past your measurement. A QFC estimates a fact that does not exist yet. The boundary between “real but unmeasured” and “not yet real” is not a semantic detail: it is the line past which you stop counting and start betting.

Validate the QFC in four steps before you trust it with a budget

A non-auditable number provided by the platform is not a reason to throw it out, it is a reason to cross-check it. Here is the framework I would apply before letting a QFC influence any budget decision.

1. Build a cohort of interactions and accept that you have to wait. The QFC is judged over time, by construction. Freeze a cohort of ad interactions over a given period, note the predicted QFC at time T, and let the clock run. If your conversion windows are badly set, your realized figure will be biased before the comparison even starts: first check how to choose a GA4 conversion window suited to a long cycle.

2. Compare predicted and realized at 30, 90 and 180 days. This is the heart of the validation, and it happens on your GA4 export in BigQuery, not in the Google Ads interface. You reconstruct the cohort’s realized conversions at each milestone and set them against the QFC predicted at the start. The query below does precisely that; to go further on this kind of analysis, my library of BigQuery queries for GA4 covers neighboring cases.

3. Test incrementality, not just the prediction. Even a perfectly calibrated QFC does not tell you whether the campaign created those conversions or whether they would have happened anyway. An accurate prediction and real impact are two distinct questions. The only clean answer is still an experiment, via a geo incrementality test with GeoX and GA4. The QFC tells you “how many, tomorrow”; GeoX tells you “thanks to what.”

4. Keep the QFC out of client reports until it is calibrated. Until you have seen, on your own data, that the prediction lands reasonably on the realized figure, the QFC stays an internal steering indicator, not a number you present as a result. The distinction protects your credibility the day the prediction gets it wrong.

A BigQuery query to compare predicted and realized

The idea is simple: take a cohort of ad interactions, then count the conversions actually recorded for those users at 30, 90 and 180 days. You then set that realized figure against the QFC Google showed at the start. Here is a skeleton to adapt to your dataset (table and event names need adjusting):

-- Cohort: users with an Ads interaction over a given period
WITH cohort AS (
  SELECT
    user_pseudo_id,
    MIN(event_date) AS interaction_date
  FROM `project.analytics_XXXXXX.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260401' AND '20260430'
    AND event_name = 'session_start'
    AND collected_traffic_source.manual_source = 'google'
    AND collected_traffic_source.manual_medium = 'cpc'
  GROUP BY user_pseudo_id
),
-- Real conversions for those users, timestamped
conversions AS (
  SELECT
    user_pseudo_id,
    PARSE_DATE('%Y%m%d', event_date) AS conv_date
  FROM `project.analytics_XXXXXX.events_*`
  WHERE event_name = 'purchase'
)
SELECT
  COUNTIF(DATE_DIFF(conv_date, interaction_date, DAY) BETWEEN 0 AND 30)  AS realized_30d,
  COUNTIF(DATE_DIFF(conv_date, interaction_date, DAY) BETWEEN 0 AND 90)  AS realized_90d,
  COUNTIF(DATE_DIFF(conv_date, interaction_date, DAY) BETWEEN 0 AND 180) AS realized_180d
FROM cohort c
JOIN conversions v
  USING (user_pseudo_id)
WHERE v.conv_date >= c.interaction_date;

You then compare realized_30d, realized_90d and realized_180d against the QFC recorded for the same cohort. If the prediction roughly matches the realized figure at 180 days, the metric has earned your trust for that campaign type. If it drifts systematically in the same direction, you have an exploitable bias: you will know by how much to deflate or inflate the number before using it.

Should you steer a budget with the QFC?

My position is nuanced, not binary. The QFC answers a real problem: the upper funnel is structurally under-measured by short windows, and having a projection beats steering blind on a truncated realized figure. To arbitrate between two Demand Gen campaigns, as a directional signal, it is useful.

But until it is validated on your data, it should neither feed an automated bid nor appear as a result in a client report. A directional signal points; it does not decide and it is not billable. Treat the QFC as a quantified hypothesis: interesting, testable, and promoted to steering-number status only once it has passed the four steps above. The good news is that the gradual rollout gives you time to do that work before the wide beta arrives in late 2026. Take that time: it is exactly what will set you apart from the advertiser who took the prediction for a sale.

FAQ

Does the QFC replace conversion tracking? No. Google presents it as a supplementary metric. Your measured conversion tracking stays the basis for optimization and reporting; the QFC comes on top, on the upper funnel.

Do you need extra tagging to get it? No. Google’s pitch is “zero setup”: the metric is computed on the platform side from existing signals, with no configuration on your part.

What is its prediction window? Up to 180 days after the ad interaction, which is aimed squarely at the long cycles of Demand Gen and upper-funnel campaigns.

When will it be available to everyone? At announcement time it was a restricted global pilot, with a wider beta announced for late 2026. Check whether the three columns have appeared in your own account.

Can you bid on the QFC right now? Not advisable until you have compared predicted against realized on your data. An uncalibrated prediction steering bids is the best way to optimize toward a number that does not exist yet.