Mexico is a vast and diverse market. It is the world’s 16th-largest economy based on GDP, with a population of 130 million. For many years, it has been attractive to domestic and foreign investments, regardless of its political condition and environment.
Domestic and international retail companies doing business in Mexico have established hundreds or even thousands of points of sale across the country, from Tijuana in the northwest to Cancun in the Yucatan peninsula in the southeast. Regardless of their technology adoption level, many of them are still planning and forecasting their product sales with spreadsheets, manual methods, or other simple tools. These simplistic methods will invariably become a limitation and liability to most of them, ultimately negatively impacting their bottom lines.
Consider this simple example: Imagine a retailer with 100 points of sale that carries an average of 300 SKUs per store. This product matrix would contain 30,000 items that must be fully understood and planned. Let’s talk about sporting goods, for example, tennis shoes — product sales performance is different from region to region, town to town, and store to store due to a diverse set of factors including, but not limited to, climate, customer purchasing power, genotype, culture, and demographics.
Determining all the correlations and interdependencies between different combinations of stores and products can be overwhelming or simply impossible for a human. However, it is a fact that white, size 28 sneakers will sell more in a Monterrey store than in an Oaxaca store; and conversely, red, size 25 will sell more in Oaxaca.
What’s the point?
It should be self-evident. SKU/store level demand planning is necessary to maintain adequate inventory, stocking efficiencies, and high levels of customer service and satisfaction.
Certainly, some may argue they can get a “high level,” or “pretty good” sales estimate using Excel by studying overall quantities of high rotation SKUs, which according to the Pareto rule will constitute around 60 out of 300 SKUs following the same example. Even if true, how about the 240 remaining products and all the costs related to reallocating goods between stores to avoid sales loss, and all the discounts subsequently applied to get rid of product excesses? It is hard if not impossible to manually keep “the right goods at the right place at the right time, all the time.” Pure and simple.
Now, let’s step up the complexity with a practical example. Let’s factor in sneaker sizes. We’ll keep it simple by adding S, M, L, and XL as if it were clothing. Now, imagine how large the product matrix/catalogue becomes. I’ll let you do the math.
Let’s be realistic. You could hire five to 10 demand planners and instruct them to crunch numbers all day long using almighty Excel, Access, and other similar tools. But this exposes decision-making to too many potential errors. Ultimately, these practices end up being expensive, painful, and risky.
Now, some of you have mastered these legacy tools, and may have solved the first order of problems and managed to cover most of the product matrix (SKU/POS/Colour/Size). You already have the workforce and the tools to reasonably forecast sales. But now comes the dilemma: How accurate could that sales forecast be? To answer this question, you must put together an intensive effort and resources, start working with a sales forecast, act upon it, measure it, and inevitably notice that the accuracy is poor because you are still getting similar amounts of overstock and out-of-stocks as before.
So, what have we done wrong?
To forecast sales accurately, we need “good history sales data” with relevant time horizons of historical information. This is a good beginning but will not be enough — unit sales history simply does not tell the whole story. It will not tell you when sales were affected by an out-of-stock, by promotion, by weather,or other relevant variables that impact sales performance.
The point is this: There are subtle trends in unit sales history that humans and simple forecasting formulas simply cannot detect from analysing raw one-dimensional data in a spreadsheet and certainly will be linked directly to the quality of the forecast.
Now think, what would I do if I had a very accurate forecast? If I sell 10 pineapples today in my store and my sales forecast says I will sell only four tomorrow for whatever reason with a high level of confidence, would I replenish 10 again knowing that I will throw six away at the end of the day? Of course not, thus I will just let you fantasise about this possibility.
The fact is outstanding levels of forecast accuracy are now possible. AI and machine learning platforms are available and can help you analyze and interpret millions of data sets and variables. This gives you the capacity to automate your forecasting processes and cover all your product catalogs no matter how extensive they are. This new technology can help you solve difficult tasks like forecasting new products by easily linking them to the sales history of existing and similar ones to generate their first rollout forecasts. Moreover, these new technology platforms enable you to maximize the use of working capital through a more efficient inventory mix that increases sales, minimizes out-of-stocks, and reduces overstocks.
In my experience, adequate demand planning and inventory management must be a strategic operational priority for all companies selling products. Inefficient demand and inventory management strangles profitability.
I encourage organizations, not only retailers, but also distributors and manufacturers, to implement and support modern demand and sales forecasting practices, managed by planning professionals, and supported by state-of-the-art AI technology.
What piece of the puzzle are you missing?