Google Ads Conversion Window and Conversion Lag in 2026: Why Yesterday’s ROAS Lies
You open the account on Monday morning. Yesterday’s campaign shows a 1.4 ROAS against a 3.0 target, so you cut the budget. Nine days later you look at that same date again and it reads 3.6. The campaign was profitable the whole time — you strangled it mid-flight.
That gap has a name, and it is made of two separate things that people constantly confuse: the Google Ads conversion window (a setting you choose) and conversion lag (a behaviour you measure). This guide covers how the conversion window works in Google Ads, how to build your own conversion lag curve from account data, how to pick a window length that matches your real sales cycle, and which decisions you must not make before a specific number of days have passed.
What a Google Ads conversion window actually is
A Google Ads conversion window is the period after an ad interaction during which a conversion still gets credited to that interaction. Someone clicks on 3 August and buys on 20 August: with a 30-day window that purchase is recorded against the click date, not the purchase date. That is exactly why yesterday’s numbers keep climbing for weeks.
The window is set per conversion action, not per account. Path: Goals → Conversions → Summary → open the action → Edit settings.
| Window type | Range | Default | When to touch it |
|---|---|---|---|
| Click-through | 1–90 days | 30 days | Your main lever — match it to the real sales cycle |
| Engaged-view | 1–30 days | 3 days | Video campaigns; counts views of 10 seconds or longer |
| View-through | 1–30 days | 1 day | Handle with care — longer windows collect coincidences |
Two mechanics break people’s mental model more often than anything else:
- Window changes apply forward only. Extend from 7 to 30 days and the conversions you previously missed do not come back. There is no retroactive recount.
- Assisted-conversion reporting is separate. Assist data is shown across 30-, 60- and 90-day lookbacks regardless of the window you set on the action itself.
Conversion lag is not the window — it is your customers
The window is a configuration. Lag is a fact: how long it actually takes your audience to move from click to purchase. You choose one and you measure the other.
The practical difference shows up fast. If 80% of your sales land within 24 hours and your window is 30 days, you are simply waiting longer than necessary — harmless. If 45% of your revenue arrives between day 8 and day 21 and your window is 7 days, you never see almost half of it — and neither does the bidding algorithm that learns from that data.
A short window does not make an account faster. It makes part of your conversions invisible — to your reports and to Smart Bidding at the same time.
The time-lag report: where to find it and what to extract
Google ships the tool so you do not have to guess:
- Go to Goals → Conversions → Summary.
- Open the time-lag reporting view (days-to-conversion distribution).
- Pick a range of at least 90 days — anything shorter cuts off the tail you are trying to measure.
- Segment by a single conversion action. Mixing “Lead” and “Purchase” into one curve produces a number that describes neither.
- Write down the distribution by bucket: <1 day, 1–2, 3–6, 7–13, 14–30, 31–60, 61–90 days.
The report splits click-through and view-through lag separately. What you actually want is the cumulative share — that is the line that answers “after how many days is this data finished?”
Building your maturation coefficient
The distribution is the raw material; day-to-day you want a single correction factor:
K(d) = final conversions for day D ÷ conversions visible d days after D
Take six to eight “cold” days (older than 45 days), record how many conversions were visible on day 1, 3, 7 and 14, and average them. You end up with something like the table below — the numbers illustrate the method, not a benchmark for your vertical:
| Days after click | Visible share of final | Coefficient K | What you may do with it |
|---|---|---|---|
| 1 | 52% | 1.92 | Raw. No decisions at all |
| 3 | 71% | 1.41 | Direction only, not level |
| 7 | 88% | 1.14 | Segments can be compared to each other |
| 14 | 96% | 1.04 | Working point for budget decisions |
| 30 | 100% | 1.00 | Closed |
The rule that follows is simple: when you look at a three-day slice, multiply visible conversions by K(3) before comparing anything to target. An account with K(3) = 1.41 showing a raw 2.2 ROAS is already tracking toward 3.1 — nothing there needs cutting.
Choosing a window length: four rules
Rule 1. Cover 90–95% of the cumulative curve, not 100%. The 3–5% tail arriving on day 60–90 costs more in waiting than it returns in insight.
Rule 2. Actions that mean the same thing get the same window. The classic wreck: “Purchase (web)” on 30 days and “Purchase (offline import)” on 7 days. Adding those two numbers together produces nonsense.
Rule 3. Do not set 90 days “just in case”. A long window smears the link between cause and effect: edits you made three weeks ago keep landing in today’s report, and experiments become painful to read.
Rule 4. Change the window once, and log the date. A window change is a break in the data series. Every before/after comparison across that point is invalid until at least one full new window has elapsed.
| Business type | Typical behaviour | Starting point for click window |
|---|---|---|
| Impulse purchases, food delivery, low-ticket e-com | Most volume inside 24 hours | 7 days |
| Mid-market e-com, $50–300 AOV | Peak on days 1–3, tail to two weeks | 14–30 days |
| Local services, home repair, healthcare | Fast enquiry, slow close | 30 days on the lead, separate action for the sale |
| B2B, high-ticket, subscriptions | 3–8 week cycle | 45–60 days plus offline conversion import |
| Apps and subscription products | Install now, pay later | Different windows for install and first payment |
Treat these as hypotheses to verify, not settings to copy. The only valid source for your window length is your own time-lag report.
What happens to bidding when the window is shorter than the lag
This is where the money goes. Smart Bidding optimises toward the conversions that fall inside the window. Cut off 40% of the outcome and the algorithm sees a distorted funnel, then does three things:
- It underrates slow segments. Queries that bring a considered buyer who converts on day 10 look unprofitable and lose impression share.
- It shifts budget toward fast, cheap conversions. In lead gen this reliably means drifting into junk leads: they register instantly, and quality is invisible unless you feed the quality signal back.
- It demands a higher tCPA than it should. You end up paying for the algorithm to hit a target using half the evidence.
The fix is measurement, not bid tinkering: a sane window, enhanced conversions to recover matches you are currently losing, and offline conversion import for deals that close in the CRM. When deal sizes differ, add conversion value rules so a $5 enquiry and a $900 contract stop looking like the same event.
A worked example: the same decision, two moments in time
Take an account spending $7,000 a week across four campaigns with a $40 CPA target. On Monday morning the manager looks at the last three days and sees the left-hand numbers. Two weeks later, those exact same days read like the right-hand columns.
| Campaign | Spend | Conversions, day 3 | Raw CPA | Conversions, day 17 | Final CPA |
|---|---|---|---|---|---|
| Brand search | $640 | 34 | $18.8 | 37 | $17.3 |
| Generic search, broad themes | $1,980 | 29 | $68.3 | 52 | $38.1 |
| Performance Max | $2,450 | 48 | $51.0 | 71 | $34.5 |
| Remarketing | $510 | 19 | $26.8 | 21 | $24.3 |
Look at the shape of the error. Brand and remarketing barely mature at all — that audience has already decided, so the lag is short. Generic search and Performance Max add 45–80% more conversions, because those are the campaigns bringing in people who need time to think.
A manager cutting budgets on Monday’s raw CPA removes exactly the two campaigns that end up hitting target and keeps the brand campaign, which would have harvested existing demand anyway. That is the standard way an account quietly turns into a machine for re-buying customers you already had: the dashboard looks great and growth stops.
You can test this on your own account in ten minutes. Take any day from a month ago, pull its current numbers, and compare them with whatever you logged for that day in your daily report at the time. If the spread across campaigns is more than 20 percentage points, your campaigns have materially different lag and one account-wide cutoff no longer works.
When different campaigns need different cutoffs
Split the cutoff when the account mixes fundamentally different intent: brand search next to cold prospecting, remarketing next to first touch, cheap SKUs next to high-ticket ones. A workable split is three tiers:
- Fast (brand, remarketing, high-intent queries): cutoff at 3–5 days.
- Medium (generic search, Shopping, PMax over an established catalogue): 7–14 days.
- Slow (new geos, high-ticket lines, B2B): 21–30 days.
Different cutoffs do not mean different conversion windows — the window can stay uniform. The cutoff is your internal rule for when a decision is allowed, and it is perfectly reasonable for that to vary by campaign type.
Seven mistakes that cost the most
- Deciding on yesterday’s data. Pausing an ad group “because of yesterday” is the fastest way to kill a working setup. The minimum basis is a period no shorter than your K(7) point.
- Comparing a fresh period to an old one. “Last 7 days vs previous 7” compares unfinished data with finished data. Shift both ranges back by your lag first.
- Changing the window mid-test. Any Google Ads experiment that has a window change in the middle of it is unreadable. Change before the start or after the finish.
- Different windows on duplicate actions. Especially dangerous when both are marked primary and both feed bidding.
- Ignoring adjustments. Refunds, cancellations and offline corrections arrive even later than conversions. An account measuring ROAS without them always looks better than it is.
- A 90-day window on a short cycle. Technically safe, practically blinding — you stop knowing which change did what.
- Automated rules on daily thresholds. “Pause if CPA > X” evaluated on one day of data systematically kills everything that converts slowly. If you need rules, run them on a 14-day window and apply the correction factor.
Operationalising lag in your weekly process
- Set a cutoff date. Today minus your 95%-coverage day. Nothing fresher than that enters a decision report.
- Split your dashboards. An operational one (spend, impressions, clicks, CTR, anomalies — daily) and a decision one (CPA, ROAS, cost of sale — cold data only).
- Add a corrected column. In custom columns and the report editor keep a “projected CPA” = current CPA ÷ K(d). It removes most of the panic from Monday stand-ups.
- Fix an edit cadence. Bids and budgets once a week on cold data. Structural changes no more than every second week.
- Keep a change log. Date, what changed, expected effect, and the date on which the data about that change becomes readable.
Seasonal peaks deserve their own note: lag compresses before a sale (people buy faster) and stretches afterwards. If you intervene manually around a promotion, do it through seasonality adjustments and data exclusions rather than by yanking targets around.
How lag connects to learning and traffic quality
Learning. Changing a conversion window changes the data your strategy learns from, so the strategy effectively re-enters learning and behaves erratically for a while. Which edits trigger that and how to schedule changes without restarting the algorithm every week is covered in the piece on the Smart Bidding learning period.
Source quality. If part of your traffic comes from surfaces with different behaviour, one averaged lag curve stops describing reality — one source converts within the hour, another never converts at all. Where your ads run outside Google Search and how to audit it is in the article on the Google Search Partner Network.
Attribution. The window decides whether a conversion is counted; the attribution model decides which interaction gets credit. The difference is unpacked in GA4 attribution for media buyers.
If you run several projects and need stable ad accounts with measurement wired correctly from day one, that groundwork is already handled with Google Ads agency accounts from PPC Rebels.
A 40-minute checklist
- Export the time-lag report for 90 days, one primary conversion action at a time.
- Find the day where the cumulative curve reaches 90–95%. That is your candidate window.
- Verify that every same-meaning primary action shares the same window.
- Calculate K(1), K(3), K(7), K(14) and put them in a shared team document.
- Add the cutoff date to your weekly report and ban decisions on fresher data.
- Audit automated rules and scripts for daily thresholds; rewrite them on a 14-day basis.
- Set a quarterly reminder to recalculate K — lag moves with season, pricing and product mix.
Related reading: Customer type labeling and lifecycle goals: Google now classifies your lists
FAQ: conversion windows and conversion lag
What is the default conversion window in Google Ads?
30 days for click-through conversions, 3 days for engaged-view and 1 day for view-through. All three are configured per conversion action.
What is the maximum conversion window?
90 days for click-through conversions and 30 days for view-through and engaged-view. The minimum in every case is one day.
If I extend the window, will past conversions be added back?
No. Window changes apply going forward only. Conversions missed under the shorter window will not reappear.
After how many days is the data final?
Whenever your cumulative curve reaches 95–100%. In fast e-commerce that is three to seven days; in B2B it can be 45–60. There is no cross-industry answer, and borrowed averages are actively harmful here.
Does the conversion window affect Smart Bidding?
Directly. The strategy optimises on conversions that landed inside the window, so a window that is too short hides part of the outcome and biases bids toward fast, often less valuable conversions.
What if conversions arrive after 40–60 days?
Split the funnel: keep a fast qualifying action (enquiry, registration, qualified lead) as the primary conversion with a short window, and import the final sale as an offline conversion with a long window and a real value attached.
How is a conversion window different from an attribution window?
The conversion window decides whether the conversion is counted at all. The attribution model decides how credit is distributed among interactions. They are independent layers.
Why did yesterday show 12 conversions and now the same day shows 17?
Normal maturation — conversions are stamped to the click date, not the event date. Recent days are always understated, which is precisely why you cannot act on them.
Should I set 90 days so I never lose anything?
Not if your real lag is shorter. Long windows stretch the link between edit and effect, make experiments unreadable and delay your detection of problems.
How do I explain lag to a client or a manager?
Show two columns for the same date: “how it looked after 24 hours” and “how it looks now”. One screenshot beats any explanation of attribution theory.
How often should the lag curve be recalculated?
Quarterly, and always after a change in pricing, assortment, landing pages or a launch in a new geo. Lag is a property of demand, and demand moves.
Do refunds and cancellations affect these calculations?
More than people expect. Adjustments arrive later than the conversions themselves, so accounts with high return rates need a separate net-revenue curve or ROAS will read permanently inflated.
The takeaway: establish when your data becomes true, then build decisions on top of it. An account making decisions on cold numbers beats an account with smarter bids and raw ones.
Next steps: read the Smart Bidding learning period guide and the breakdown of Google Ads bidding strategies to see how edits and data quality interact. For a broader tune-up, the account audit checklist covers the measurement layer end to end.
