A measurement error you can tolerate in August becomes a budget decision you cannot reverse in November.
The reason is arithmetic. In an ordinary month, a tracking gap that misplaces five per cent of installs costs you a rounding error. In a six-week window where you spend a quarter of your annual budget, the same gap misdirects real money, and you find out in December when the numbers stop reconciling. By then the campaigns are over.
Diwali falls on Sunday 8 November 2026. The auction starts getting expensive from late September with Flipkart Big Billion Days. That gives you until roughly mid September to get this right, and after that you should not be touching it.
Why does festive traffic break attribution that works the rest of the year?
Festive attribution is the practice of verifying and freezing your measurement setup before a compressed high-spend period, so install and revenue credit stays reliable when volume, auction competition and fraud pressure all rise together.
Three things change at once, and each one exposes a different class of problem:
-
Volume spikes expose race conditions. Postback delays, duplicate event firing and timeout handling that never surface at 500 installs a day become visible at 5,000. These are not new bugs. They were always there, below the threshold where anyone noticed.
-
The window is short. A broken week in March is two per cent of your year. A broken week in the festive window can be a fifth of your festive spend, and there is no second attempt.
-
Fraud follows money. Click flooding and injection scale with the value of the inventory. Festive is when it is most worth someone's while, and when your anomaly thresholds, tuned on quiet-month traffic, are least likely to catch it.
An MMP that has been quietly slightly wrong all year will be loudly wrong in November.
The attribution window problem nobody plans for
This one deserves its own section because it is specific, common, and almost never discussed.
Big Billion Days runs roughly ten days. Your default click-through attribution window is probably seven days.
So consider a user who taps your ad on day one of the sale, browses, does nothing, and converts on day nine. Under a seven-day window, that conversion is unattributed or credited elsewhere. The campaign that actually created the demand shows a worse return than it earned, and you may well cut it mid-window on the strength of that.
Now stretch it further. A user clicks during Big Billion Days in late September and finally transacts during Dhanteras in early November. That is a six-week consideration gap, entirely plausible for a considered purchase, and invisible to any standard window.
Your options, and their costs:
-
Leave the window at seven days. Simplest, and consistent with the rest of your year. You will under-credit long-consideration campaigns and should say so out loud before anyone reads the report.
-
Widen the window for the festive period. Captures more genuine conversions, but also captures more coincidental ones, so precision falls. If you do this, widen it before the window opens and leave it alone, and record the change so the year-on-year comparison is not silently broken.
-
Keep the window and add a longer-window view alongside it. Best of both, if your reporting supports two views. Decide in advance which one drives budget decisions.
There is no correct answer. There is only a decision made in September with the team's agreement, or a decision made accidentally in November by whoever last touched the settings. The travel app playbook on 90-day consideration windows works through the same trade-off for a category where it is a year-round problem.
The pre-festive checklist
Nine checks. Run them before mid September.
-
Freeze your SDK version. Confirm the version in production is current and stable, then stop. No SDK updates during the window, no exceptions. Verify the integration works using the integration testing flow rather than assuming.
-
Verify postback mapping per network. Each network separately, not "we set this up last year". Check Meta and Google configuration against a live test conversion. Our complete postback setup guide has the per-network detail.
-
Confirm Meta Install Referrer is configured on Android. This is the most consequential single setting for Android-heavy Indian apps, and the most commonly missed. Without it, Meta-driven Android installs are systematically under-attributed, which during festive means quietly defunding your best-performing channel. The configuration requirements are documented.
-
QA deep links against every festive landing surface. Sale pages, category pages, offer pages, anything built specifically for the campaign. New surfaces are exactly where routing breaks. Use the deep link QA checklist.
-
Check revenue events fire with correct values and currency. Test an actual transaction end to end. Currency mismatches are silent and they corrupt every downstream return calculation.
-
Agree how reattributions are treated. Decide now whether a returning user credited to a new campaign counts towards acquisition targets, and make sure everyone reading the dashboard knows the answer.
-
Review fraud thresholds. Anomaly rules tuned on quiet traffic will either miss festive fraud or scream at legitimate festive volume. Both are bad. Our click flooding prevention guide covers what to look for.
-
Configure alerting on volume anomalies. You want to hear about a tracking break within hours, not at the Monday review. Setting up MMP alerts takes an afternoon.
-
Capture your baseline. Current install volume, activation rate, cost per activated user, by channel. Write it down in August. Without it, every festive number is uninterpretable.
What to freeze and what to leave flexible
The governing rule: never change how you measure while you are changing how much you spend. If both move at once, you cannot attribute the difference to either.
Freeze from mid September:
-
SDK version
-
Event names and event taxonomy
-
Conversion value mapping
-
Attribution window settings
-
Postback configuration
Leave flexible:
-
Creative
-
Budget and bids
-
Audiences and targeting
-
Campaign structure
The freeze list is everything that defines the ruler. The flexible list is everything you are measuring with it. Teams get into trouble when a well-meaning engineer ships an event rename in week two of the sale, and nobody connects it to the conversion drop until December.
Daily checks during the window
Three numbers, every morning, five minutes:
-
Installs against yesterday and against the same day last week. You are looking for a cliff, not a wobble.
-
Activation rate. The share of installs reaching your onboarding event. This is your earliest signal that either traffic quality or tracking has changed.
-
Network-reported conversions against MMP-reported conversions. You are watching the size of the gap, not the gap itself. A gap is normal. A gap that doubles overnight is not.
Distinguishing a genuine anomaly from festive noise comes down to one question: did the ratio change, or only the volume? Volume moves constantly during festive and means little on its own. Ratios moving sharply, particularly activation rate and the network-to-MMP gap, almost always mean something broke.
Reconciling after the window closes
Network numbers and MMP numbers will diverge more than usual in November. Expect it, and do not treat it as a failure.
The divergence widens because self-attributing networks claim credit under their own rules and windows, view-through activity rises with impression volume, and both effects scale with spend. Our explainer on why Meta ROAS and MMP ROAS do not match sets out which to trust for which decision.
Wait at least 14 days after Bhai Dooj before reconciling. Late postbacks and delayed revenue events keep arriving, and reconciling too early produces numbers you will only have to redo.
Then write down four things while you still remember them: what broke, what you changed mid-window and when, which decisions you made on incomplete data, and what you wish you had set up in September. That document is worth more to next year's team than the campaign report.
Start the checklist now
The nine checks take a couple of days spread over a week. The window for doing them calmly closes around mid September, after which you are debugging live at peak CPM.
If you want to run these checks against your own setup and see what the numbers look like before spend climbs, request a demo and we will go through the configuration with you.
FAQ
What should I check before launching a festive campaign?
Verify postback mapping per network, confirm Meta Install Referrer is configured on Android, QA deep links against every new festive landing surface, test revenue events end to end for value and currency, review fraud thresholds, set up volume alerting, and capture your pre-festive baseline. Then freeze the SDK and event taxonomy until the window closes.
Should I change my attribution window for a multi-day sale?
There is no universally right answer. A ten-day sale against a seven-day click window will under-credit long-consideration campaigns. If you widen the window, do it before the window opens, leave it alone once spend starts, and record the change so year-on-year comparisons stay honest.
Can I update my app's SDK during the festive season?
Avoid it. An SDK change alters how you measure at the same moment you are changing how much you spend, which makes any difference in the numbers impossible to attribute to either. Ship updates before mid September or after the window closes.
Why is ad fraud worse during festive periods?
Fraud follows the value of the inventory. When budgets and bids rise, so does the incentive to intercept them. The additional problem is that anomaly thresholds tuned on ordinary traffic tend to misfire during festive volume, in both directions.
How long after the sale should I wait before reconciling numbers?
At least 14 days after the window closes. Late postbacks and delayed revenue events continue to arrive, and reconciling before they land produces figures you will have to revise.