Baselines: understanding normal sales

Thom Iddon-Escalante

Marketing Director
  • To evaluate a campaign, retailers need to compare actual sales data with an estimate of what they would have sold if a promotion wasn’t running.
  • When planning a new campaign, retailers need to estimate how much they could sell without a promotion to know if it is worth doing.
  • Regardless of whether the objective is volume, profit, or both, baselines are retail’s benchmark.

How do retailers estimate baselines?

Most retailers use one of three baseline models:

  1. Mean of previous 10 non-promotion weeks (PMT-10)
  2. Discounted mean of previous 10 weeks (TPM-10)
  3. Same non-promotion period last year (HIST)

These models are easy to implement, draw on accessible data, and don't require large analytical capacity. They may also prove sufficient for replenishing everyday sales stock. But they are inadequate for evaluating promotions.

Formulate experiments

We put each of the models to the test estimating baselines for 4 different item types, for which baselines could be reliably determined by independent means.

  1. Life cycle
  2. Seasonal
  3. Upward trending
  4. Clearance

Figure 1 shows how sales volumes vary over time for each item type.

Figure 1:  4 Items Types : Lifecycle ,  Seasonal ,  Upward trending ,   and  Clearance

The results

Figure 2 shows how much each model under or overestimates baselines for each item type.

Figure 2:  % mean average baseline estimation error for 3 models across 4 item types (How much the models over or underestimate baselines

What does the test show?

  • For each model the margin of error is too wide to determine whether or not a given promotion fulfilled its objective
  • The HIST model is especially poor. It under or overestimated baselines by 34 percent on average, and by 53 percent for clearance products specifically.
  • Even the performance of the most accurate model, PMT-10, varied significantly across item types. It was more than twice as likely to under or overestimate life cycle (17%) than upward trending products (7%).

Why is estimating baselines so difficult?

#1 Demand changes

It is difficult to predict demand. Tastes and preferences aren’t static. Products go in and out of vogue. Some products, like ginger shots and protein puddings, display steep preference trends few could have predicted. Other products flop when nobody thought they would.

#2 Supply changes too

External factors affect supply. Climate, for instance, affects harvests, makes commodities rarer, drives up prices, and stifles demand. New products routinely enter the market, for which retailers have little or no data against which to benchmark baselines.

#3 Scale and complexity

Retailers run thousands of promotions across hundreds of stores each week. This generates a vast amount of data that needs to be processed, analysed, and acted upon in very short order. Most retailers simply cannot keep up.

#4 Inferior tools

Merchandising teams’ toolkits are not up to task. Excel was not built for this kind of volume and complexity. Consultants run pricing exercises on dead data sets, but can’t offer tools for operational use. Enterprise software providers offer tools, but they’re outdated, unwieldy, and often overly technical.

Stop using baselines you can’t trust

Retailers know strikingly little about their promotions, even though a major chunk of their revenue is generated by them. Bad baselines are the main culprit.

Enter Formulate

We repeated our experiment, but this time included Formulate’s proprietary baseline algorithm.

Figure 3 shows the results. Formulate accounts for changes in supply and demand, and it scales across the entire assortment. It is also between 2 and 10 times more accurate than other models at estimating baselines.

Figure 3:  % mean average baseline estimation error for 3 models  + Formulate  across 4 item types (How much the models over or underestimate baselines)

Formulate is between 2 and 10 times more accurate than other models at estimating baselines.

Why does this matter?

Take a seasonal item like ice cream selling 1000 units a week. The HIST model has a 26% error margin, is off on average by 263 units, and estimates a range of 737 to 1263 units in sales. Formulate, on the other hand, is off by just 50 units, and estimates a range of 950 to 1050 units.

As a buyer how many units should I purchase? Too few, I risk underestimating sales, and miss out on profit. Too many, I risk overestimating sales, decimate my margins and create waste.

The wider the range, the harder it is to plan. The harder it is to plan, the more likely a scenario with either waste or shortage. One scenario is bad for the environment. The other is frustrating for customers. Both are costly for retailers.

2x the business impact of your promotions with Formulate

By using Formulate retailers can double the business impact of their promotion decisions by:

  • Leveraging several percentage points profit improvement on every promotion
  • automating campaign management and reducing duplicate work
  • facilitating communication across the organisation.

At a company level, this translates into a profit improvement of up to 10%; a major profit contribution for such a high-volume, low-margin business.

Book a demo

Like the article? Spread the word

Sign up to receive stats, data, and insights that will help you keep up with retail trends

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.