A forecast that gets corrected in time

The problem it solves

The forecast is built in January, carefully, and filed. In December somebody opens it to see how wrong it was, observes that the year was unusual, and closes it again. Between those two dates the forecast did no work at all, even though those were precisely the months when something could still have been done.

The format carries part of the blame. A single number cannot be corrected: it is either met or missed, and when it drifts there is no way to tell which assumption failed, because the assumptions were never written down anywhere. The October conversation becomes an argument about whether the year was strange, which is an argument with no exit and no lesson in it.

Meanwhile the decisions that depended on that forecast, a large purchase, inventory, a hire, a payment commitment, were all taken on the January version. A company with seasonal demand or a handful of large accounts can carry a drift for months without seeing it in the income statement, and then see it all at once in the bank.

What it covers

The workshop builds a forecast from the history the company already has. The first section separates three things that usually travel together in the same line: seasonality, trend and noise. Telling them apart changes how any series reads, and it heads off the hasty correction made when a weak month gets mistaken for a change of direction, which is the most expensive correction of all.

Then the single number gives way to a range. Each participant builds a low and a high scenario for their own business and writes down what each end depends on, against a demanding standard: somebody else should be able to argue with those assumptions without having to ask anything first. An assumption nobody can argue with is not an assumption, it is a belief, and a belief does not get corrected when reality disagrees with it.

The third block separates revenue from cash. They are two different forecasts answering two different questions, and a company can have the first in good order and the second in trouble for months without noticing. The group works on where in the year the two curves part company, which decisions depend on each, and what warns before that gap becomes a cash problem.

The close defines the correction rule. Which event forces a revision ahead of schedule, who declares it, and what happens that day. The group also agrees the format for reviewing forecast against outcome: a comparison that looks for which assumption failed rather than for who got it wrong. The session deals plainly, too, with the cases where a forecast should not be attempted at all.

What participants learn

• Build a revenue and cash forecast from the company's own history.
• Interpret a series by separating seasonality, trend and noise.
• Develop the result as a range and explain what each end depends on.
• Define the event that forces a revision ahead of schedule, and who declares it.
• Compare forecast against outcome looking for the assumption that failed rather than the person.

Agenda

1. 01What history you need, and what can be forecast without it
2. 02Seasonality, trend and noise
3. 03From a single number to a defensible range
4. 04Written assumptions: the part worth arguing about
5. 05Forecasting cash and forecasting revenue are different questions
6. 06Events that force an early correction
7. 07Reviewing forecast against outcome and learning from it
8. 08When a forecast should not be made at all

What the organisation leaves with

• Your own forecast expressed as a range, with the assumptions written beside each end.
• The list of events that force a revision of the forecast ahead of schedule.
• A forecast-against-outcome review format, with the question asked at every deviation.
• A revision calendar for the forecast, with an owner and a date for each checkpoint.