Forecasting revenue, and putting a price on inaction

Why forecast your revenue

A sales forecast is there to help you decide before you are forced to: how much stock to order ahead of a peak, which ad budget to commit, what cash to plan for, whether this month’s target is still within reach. Without a forecast, those decisions are made on gut feeling, or too late.

Three methods, from the simplest to the most reliable

1. The same period last year

Take last November’s sales to forecast next November. Simple, and it captures seasonality. But it ignores your growth (or decline) since then, and it repeats last year’s accidents: a stockout, an exceptional promotion, an outage.

2. The moving average

The average of recent weeks’ sales, projected forward. It follows your current level but smooths everything out: it does not know that a Monday is not a Saturday, or that December is not February.

3. A recent baseline adjusted by factors

The most robust method combines both: a baseline computed from recent history, weighted to give more importance to the latest days, then adjusted by factors measured on your own history:

• the day of the week effect;
• the seasonal effect (month of the year);
• the underlying trend;
• public holidays, when you sell noticeably less than on a normal day.

One detail changes everything: exclude stockout days from a product’s history. A product that did not sell because it was sold out did not lack demand; counting those days as zero underestimates future sales, and causes the next stockout.

Measuring how reliable a forecast is

A forecast without an error measure is not worth much. The most common indicator is the mean absolute percentage error (MAPE): each day, compare the forecast with what actually happened, and average the gaps. A 10% error is easy to read: on average, the forecast is off by a tenth.

Measuring error also lets you improve: when several methods run side by side, you can give more weight to the one that was least wrong over the last few weeks.

From forecast to recommendation

A forecast on its own is just a chart. It becomes useful when it is cross-checked with the rest of your data to detect signals:

• a best seller whose stock will not cover forecast sales;
• a margin drifting on a product;
• sales unusually low or high compared with the forecast;
• a heavily visited product page that converts poorly;
• a high-margin product that no longer sells;
• regular customers who are not coming back.

Putting a price on inaction

Ten signals at once are ten reasons to do nothing. To prioritise, the question is not “what is wrong?” but “what does each day without action cost me?”.

Take a product that generated €4,200 of revenue over the last 30 days, or €140 a day, whose stock will run out in 5 days. A week of stockout puts about €980 of revenue at stake. A 3-point margin drop on a product selling €10,000 a month costs €300 a month for as long as it is not fixed. A group of loyal customers drifting away represents the sum of what they usually spent.

Expressed in euros, every signal can be compared with the others, and the list sorts itself: start with what costs the most and can still be avoided.

How Lumexio does it

Lumexio forecasts your revenue and each product’s sales over the next 90 days, using the method described here: weighted recent baseline, day-of-week, seasonal, trend and public holiday factors, stockout days excluded. For overall revenue, several sub-models are then combined according to their accuracy measured over the last few weeks.

Detectors then cross-check forecasts, stock, margins, visits and customers to produce prioritised recommendations. Upcoming stockouts are costed as revenue at stake, and customers about to churn as revenue at risk. On PrestaShop, an upcoming stockout on a product that customers are waiting for through back-in-stock alerts goes straight to the top of the list. Discover Lumexio for PrestaShop or for Shopify.