Skip to content

Forecasting, scenarios and budget recommendations

Projects one recorded figure forward with a range around every day, measures how an outcome has moved with spend, and recommends where spend could go. Refuses by name whenever the figures cannot support the question.

Updated

Forecasting takes one figure already recorded for your workspace and projects it forward. Every projected day arrives with a range around it and the sentence saying how that range was worked out. A projection without its range is not a forecast, and this product will not show one.

Nothing here is produced by a language model. Every figure is arithmetic over figures platforms have already reported into your performance records, and the page formats what the server computed.

BEFORE YOU START

  • Figures recorded for the workspace over several days. A forecast is a statement about a series, and one observation is not a series.
  • Enough of those days settled. Recent days are still being revised by the platform that reported them, so a forecast will not lean on figures that are still moving.
  • Figures that are daily. Some platforms report a running total that only ever rises rather than a figure for each day. A line fitted through a running total climbs whatever your audience did, so this refuses instead and says which kind of figure it found.
  • View permission. Nothing on this page writes, schedules or spends anything.

STEPS

  1. Open Forecasting from the main navigation, or press Ctrl-K and search for it.
  2. Choose the workspace and the figure you want projected, and how many days ahead.
  3. Read the projection together with its range. The range is the answer as much as the line is.
  4. To ask what has gone with a different level of spend, use the scenario control and enter the spend you want to compare against.
  5. To compare what each channel has returned per unit of spend, ask for a budget recommendation.

WHAT YOU SHOULD SEE

A projection is a table of days. Each row carries the projected value, the range around it, and the full sentence stating the method, the window it was fitted over and how many observations it used.

A scenario reports the slope it measured with a confidence interval. Where that interval includes zero, there is no counterfactual at all and the page says so rather than drawing a line through noise.

A budget recommendation ranks channels on non-overlapping intervals and proposes a bounded reallocation, marked as a recommendation.

Where the figures cannot support the question, you get a refusal in prose - no chart, no axis and no zero. An empty chart reads as a forecast of nothing happening, which is a different and false claim.

IF IT DOES NOT WORK

  • "Not enough recorded days" means exactly that. Time or a reconnection will fix it; nothing on this page will.
  • "Not enough settled days" means the days exist but the platform is still revising them.
  • A refusal about mixed granularity or an undisclosed day boundary means two sets of figures cannot honestly be added. Where a platform does not disclose the hour its day begins, its figures are not combined with another platform's, because two days that are not the same day are not the same day.
  • A refusal about a running total means the figure you chose is cumulative. Choose a daily figure.
  • If the page says whether a forecast was possible is unknown, that is a network fault reaching the server, not a finding about your data. Try again.

WHAT THIS WILL NOT DO

  • It will not change your advertising spend. A budget recommendation is a recommendation; there is no endpoint here that could act on it, and a test pins that absence deliberately.
  • It will not tell you that spend caused an outcome. A scenario describes what moved together, and that caveat travels with the answer.
  • It does not know about a campaign you are about to run, a platform changing how it counts, or a week your competitors behaved differently.

WORTH KNOWING

Where the range around a projection is wide, that width is the honest part of the answer rather than a fault in it. A narrow range on three days of data would be the fault.

COMMON QUESTIONS

Why is the range so wide?

Because the days it was fitted over do not agree closely with one another. The width is the measurement, not a fault in it - a narrow range drawn through three noisy days would be the fault, and would read as confidence nobody has earned.

Why did it refuse when I can see figures on the reporting page?

Reporting shows what was recorded. A forecast needs enough of those days settled, at a daily granularity, from sources whose day boundaries agree. Any one of those can be missing while the figures themselves are perfectly real, and the refusal names which one it was.

Can I make it forecast anyway?

No, and that is deliberate. There is no override, because a projection produced from days that cannot support one would look exactly like a projection that could - same table, same shape, no way for a reader to tell them apart.

Does a scenario prove that spend caused the outcome?

No. It reports what moved together over the days you have. The caveat is attached to the answer rather than left to the reader, and where the slope interval includes zero there is no counterfactual offered at all.

Will a budget recommendation change my spend?

No. Nothing here writes to any advertising platform, and there is no endpoint that could. A test pins that absence so it cannot be added without somebody deciding to.

  • Watching performance for changes

    Standing watches over the figures platforms report, and the anomalies and viral opportunities they find. Every number shows the window and denominator behind it.

  • Reading your reporting

    What the workspace report totals, why a figure can say "Not reported" instead of zero, and the two ways a figure reaches it.