The seasonality calendar

The problem the calendar solves#

A marketing measurement model learns the usual shape of your sales and then explains departures from that shape. Now suppose the week before Nowruz doubles your sales — and that you also raised a channel's budget that same week. A model that does not know about Nowruz sees only one thing: a jump. And it credits the jump to the only thing that changed that week: your budget.

The result is an ROI that misdirects budget decisions for months. The reverse is just as damaging: a campaign that ran during Muharram looks weaker than it was, because the general level of buying is lower in those days.

The seasonality calendar closes that gap. Every named event becomes a control for the model, which learns separately how much Nowruz lifts, how much Muharram suppresses, and how Ramadan shifts the shopping pattern — and that amount stays in the season's share rather than a paid channel's.

Which events ship by default#

AdPix ships a default set for the Iranian market. Each event carries three things: the calendar it is computed in, the length of its window, and — where it applies — a lead period for buying that starts early.

The category is a description, not an instruction: surge means sales usually rise, suppressed that they fall, pattern that behaviour shifts without the volume necessarily changing. The real size and direction of each event is learned from your data; if Nowruz is a quiet season for your business, that is what the model learns.

The lead-up window#

The lead-up column encodes what retail already knows: Nowruz shopping starts three weeks out, not on the morning of the first day. AdPix builds that period as a ramp that grows stronger as the event approaches, so the model can separate the pre-holiday peak from the holiday itself.

Solar and lunar dates#

Jalali events land on roughly the same Gregorian date every year. Lunar events move about 11 days earlier each year, and occasionally one falls twice — or not at all — inside a single Gregorian year. AdPix computes the Gregorian date of every occurrence for any year, past or future, so the calendar stays current with no annual maintenance.

When an officially announced date for a given year differs from the computed one, that single year is pinned with its exact dates and the other years are left alone.

Where you see the effect#

The calendar has no page of its own in your console; its effect shows up inside the two measurement tools.

In the marketing mix model — in the What drove conversions (baseline vs. paid channels) chart, the Nowruz wave or the Muharram dip sits inside the Baseline layer rather than on any paid channel's contribution. If you look at holiday weeks and see the baseline rise while the channel layers stay roughly flat, the calendar is doing exactly its job. ROI and marginal ROI come out correspondingly cleaner.

In lift tests — if your test window overlaps an event, that event enters the counterfactual as an exogenous control. The dashed Counterfactual line therefore rises through the Nowruz days on its own, and your lift is measured against that higher level. Without it, any action taken a few days before Nowruz would have produced a fake success.

Only events that actually occur inside the modelled period or the test window are used; the rest are dropped so the model is not needlessly crowded.

If your season is not on the list#

The calendar is maintained by the AdPix platform team, and the default events apply to every property. But some seasons belong to one business: your own annual sale, a production shutdown, a long outage, a heatwave that empties the air-conditioner shelves.

Ask support to add it and include three things with the request:

• The name and nature of the event — recurring or one-off.
• Exact dates — start and end for a one-off; day and month, in the calendar it is computed in, for a recurring one.
• A lead-up period, if buying starts some days earlier.

The event can be scoped to your property alone so it does not crowd other customers' models. Once it is added, the next model run or test picks it up; earlier results are not rewritten, so start a fresh run if you want to see the difference.

Frequently asked questions#

Related#