Range planning: why every number hides a persona

Planning in retail is absurdly complicated.

When you consider every factor that goes into planning a new season’s range, it’s miraculous that any decision gets made. Every assumption has a caveat, which in turn has three more. And every decison hinges on one thing: who the people buying an item are, what their tastes are and how they’re likely to change.

Demand forecasts give you a number, but they don’t understand behaviour. It’s down to the merchandisers, buyers and planners to make the big predictions about how their customers will respond. We’re forever amazed by merchandisers’ ability to navigate the range-planning maze. But with so many factors to consider and so much data to trawl through, insights get missed, predictions end up suboptimal, and costly errors creep in.

An example of how extraordinarily complicated planning in retail is

Let’s take your typical fashion retailer. They might enter a planning meeting for a season’s range nine months in advance of the season. Maybe they have access to the current season’s sales data (though many retailers may start planning over twelve months in advance of the products hitting the shelves). They then need to use this data to figure out exactly which products, and how many of each, to buy or manufacture. Order too many and it costs the business millions in markdowns, storage and written-off stock; too little and it costs even more in lost sales and lost customers.

Understanding last season

The obvious place to start seems to be to look at the current season’s sales. Maybe a style has sold through only 42% of the stock ordered with the season almost over. Alarm bells ring. The instinct is to conclude there’s no demand for this style, so you cut the order or drop it from the range.

However, if you look closer at how it has sold in each of the sizes it might look more like this:

This paints an entirely different picture. The product was extraordinarily popular in smaller sizes and unpopular in larger ones. The sales differences across sizes are only the trace. Behind them is a persona, a type of shopper with their own budget, tastes and reasons for buying (younger, price-aware commuters for example). These shoppers also happen to correlate with certain sizes. The past data tells you that some personas have strong demand and others don’t, and that information flips the decision on what to plan for next season on its head.

The same might occur with colours and locations. You look at an option overall and see it isn’t selling through well; however, a certain colour of the option might have sold extraordinarily well, while other colours sold poorly. Or, perhaps you have an option that appears to sell poorly nationally, but when you look at location-based sell-through it’s sold out in every urban hub. You don’t need fewer of the item. You need to understand which personas are buying it, and capture their demand. It’s an allocation problem.

Predicting next season

Unfortunately, simply looking at the data at a finer level isn’t sufficient, as you now face the question of whether this demand will still be there next year. Sociological research suggests that trends confined to a single persona group tend to stay small and die quickly as fads, while those that cross into other groups reach further.^1,2 The merchandiser must question whether the fact this item is only selling in urban areas is evidence of a fad that will have died out by next year, or whether it’s the early signs of a trend that will spread further in the coming year.

All of this requires understanding who the people are that are buying your product, and predicting how their behaviour will develop. A demand forecast might give you a number to base your calculations on, but it doesn’t understand the behaviour of your customers, so it can’t make these big predictions that buyers and merchandisers have become remarkably good at making.

How much risk to take

Let’s suppose you’ve now done your calculations and have a prediction of how many of that option you’re going to sell. You then face the issue of how much risk you take on.

There’s always going to be an element of uncertainty in any forecast, whether that’s made by a merchandiser’s intuition or a complex machine-learning model. The final decision of what to buy requires taking into account a whole new set of considerations.

You could order a lot and hope demand lands at the high end of the range. If you’re wrong, you have to mark items down, write them off, or tie up capital storing them until next season. If you under-order, you lose sales you would otherwise have made, and some of the customers who would have bought the item.

Perhaps an item costs you £28 to make, you sell it at full price for £40, and you can shift it at £25 in an end-of-season sale. Each lost sale costs you £12 of profit; each unit you overbuy and clear on markdown costs you only £3. The costs aren’t symmetric, so you might favour over-ordering.

The right balance also depends on who is buying. Take the same item and suppose a markdown clears it for one kind of shopper but not another (and that a leftover unit that doesn’t clear is written off for nothing):

Same product, same costs and same demand forecast, but one persona justifies stocking deep and the other stocking lean.

Another option might be storing the item until the next season and selling it again. Again, however, this is a question about the personas buying it. Is this item a staple that will always sell? In which case storing it until next season is an option as you’ll sell it the next year. But what if demand has faded within two years? If you do store it, it might sell so infrequently that it’s not worth the warehouse space and the cash tied up in it.

The problem is all of these predictions require understanding the customers, the very thing that, as our first post highlighted, traditional demand forecasts are especially bad at, and for merchandisers to take them all into consideration they need to balance hundreds of different factors in their heads.

Supplier price breaks and the range

Suppose you settle on 3,500 units and go to your supplier. They say you get a volume discount if you order 4,000. Do you stay at 3,500, or order the extra 500 at a lower unit cost and find more stores, or more personas, to absorb them?

Store space is limited, so if you introduce this product, you have to remove a different one.

Moreover, if you change what products are in each store, there’s a knock-on effect on other products. Perhaps you introduce a budget jacket into a store to hit the supplier’s volume discount, but then this might take demand away from the more premium jacket, as those who might have paid more for a jacket will now buy the budget option. Those who would have paid more might be value-seekers who, given a cheaper option, can’t justify the expense. Once again, knowing the impact requires knowing the customers.

Where it breaks down

The truth of the matter ends up being that, though the planners are remarkable at their jobs, many of these considerations have to be neglected. Every factor needs a prediction of human behaviour, and there’s just not enough time for merchandisers to calculate and consider all that needs to be considered. These decisions are rarely made optimally, and the retailer loses millions.

A better way

At Clinchr, we have a better way. Our models understand human behaviour and can quickly generate data-driven predictions of the impact of different decisions. So after each meeting, you don’t have to commission a data science team to generate new results; you can instantly see the relevant trends in the data and what our model tells you.

On top of this, our models are explainable and adaptable. You know what behaviours in which persona groups are driving the predictions, and if the experts disagree, they can adapt the models via natural language.

We strive not to replace the merchandiser, but to make their lives easier and their predictions better. If this is a problem your business faces, get in touch.