How far back a new client's marketing data goes.

Every platform decides how long it keeps what, and how far back it will serve a request. The windows differ by platform, and inside a single platform they frequently differ by metric and by breakdown.

There is a pattern worth knowing. Totals are kept longer than detail, and the breakdown you care about is usually the first thing to expire. So the honest answer to how far back a client's data goes is not one date. It is a different date for almost every column.

A backfill is an ordinary fetch for old dates. That is all it is, which sets its limits precisely.

• Anything still inside the window, at the grain still offered

  This is the case that works, and it is the reason to run a backfill at setup rather than later. The window moves forward whether you use it or not.

• Not detail that has already been aggregated away

  Where a platform now serves only monthly figures for a period, no request produces daily ones. The daily rows are not hidden behind a setting. They are not there.

• Not a definition that has moved since

  Old dates arrive under today's definition of the metric. That can produce a single series whose meaning changes partway along, which is worse than a gap because it looks continuous.

• Not anything that was never recorded

  A standard breakdown you were not looking at is usually still there, because the platform was recording it regardless. A custom field or a naming convention you introduced last month genuinely does not exist for last year.

Almost everything in reporting can be added later. A connector, a dashboard, a metric definition, a whole warehouse. Late is more expensive than early, but late is possible.

History is the exception. You cannot go back and collect last year, and no amount of budget in six months buys the rows that expired in the meantime. That makes starting collection the highest value thing available for the effort involved, and it is worth doing before you have finished deciding what you will do with it.

Year on year is the comparison clients ask for most and the first one a new setup cannot support, because it needs thirteen months of data before it can be made at all.

The predictable version of this is an agency building a full reporting setup for a new client, presenting it, and discovering in the second month that the headline comparison it was built around is a year away. If the backfill cannot reach far enough, tell the client at the start, and agree what you will report on until then.

Do not paper over it. A series whose grain or definition changes partway along should say where it changes, on the chart itself, where the person reading the chart is.

Record the date collection started as a fact stored beside the data rather than in anybody's memory. In two years nobody will remember, and a flat line at the beginning of a chart is indistinguishable from a genuine period of no activity. That ambiguity has a way of surfacing in front of a client.

Backfill runs from the date you choose when the client is onboarded, as far back as each source will still serve, which is the most that can honestly be offered given the windows belong to the platforms.

After that the history accumulates in a warehouse you own. Syncs add to the tables rather than replacing them, so when a platform restates a day that had already closed, the original figure and the correction both stay on record. That ownership is the part that matters most here: if you stop using the platform, the history is already yours, and it is the one thing in this chain that could not have been rebuilt.