Assortment is a classic optimization problem.

The assortment greatly influences the store's revenue, but is not managed by the store itself. Judging by the latest research , in general, very few people manage the range in Russia as a whole. Simply if you keep the right products in the right quantities in the right places, you can wildly raise the sales efficiency of many stores. Wildly - this is, for example, a third.

Naturally, we are no exception, we have shoals about the same as in the whole country. True, we are able to strike back to these schools. Now I will tell you about how to get revenge on the lack of presence and viciously outrage upon it.
')
The first question is why there may not be any product at all. This is an obvious thing for any person from retail, but extremely illogical for an outsider. The market has operated for centuries, so why the hell are there still inconsistencies?

Why some goods may not be?

Because it's all about the nature of our world. Imagine that you have a static matrix of products, say, 500 items for a grocery store in a residential area. This is already a simplification, because in reality there are no static matrices.

First, look at the product that has no expiration date. For example, it will be toilet paper. Obviously, it is sometimes needed by the residents of the surrounding houses, it is obvious that without it they will experience moral suffering, and it is obvious that since they went behind it, then why not buy sausages, vodka and a cake, right? Well, or at least kefir.

Toilet paper does not deteriorate and does not disappear, does not become obsolete technologically, it is simply incredibly cheap, gives you almost no profit and takes up some space. You want to keep it as small as possible, but always available, so that there is no such that someone has come after it, but it is not. You are lucky that she does not have 10 different colors, textures and sizes like clothes - this is still not a fashionable store.

In an ideal model of the world you want to do this: there are 5 rolls, every evening a car arrives with delivery, if 4 rolls are left - it is necessary that she bring one more.

The problem is that the car comes on Tuesdays and Fridays, and you need to predict the stock of paper for the intervals between these days. Since its consumption is quite creative, you use some function (most likely, historical sales data to varying degrees of averaging) and keep a small margin so that it is not exhausted.

Two new entities appear: demand prediction and stock forecasting.

I’ll now skip the features of the prediction of demand for more complex products such as beer, crunchy loaves and other things with expiration dates, but I’ll just say that there are two objective functions:

- Minimum stock in stock.
- Maximum: covering all customer needs with different fluctuations in demand.

And there is also a function of time, since deliveries are not instantaneous, and they need to be planned. Lag appears.

You can use a simple minimax type engine: determine the minimum rolls in stock and their maximum. For example, "When there are less than 4 pieces, add up to 25 to the next delivery." It works well on small numbers.

Now let's make the task more complicated. The machine with the delivery is not one, but from different suppliers. Everyone brings something different, according to their own schedule. It became a little more difficult, but still be tolerated, right?

The range partially overlaps. Sometimes it is more profitable to take small lots from one, and more from the other. And sometimes it is more profitable to endure until the goods are required by a large batch. Okay, you write another module of mathematics and everything is more or less good.

Suppliers often change conditions and change themselves. One can raise prices, the second can leave the market, the third can be 1-2 days late, the fourth always delays the order ... Is it? Not yet. One wants full prepayment, the second one is ready to trade for implementation, the third one gives you a deferment of payment for 3 weeks and so on.

The expiration dates, marketing trends (fashion), technological obsolescence of goods, juggling with a matrix of color-form models, tarry in logistics (to leave whole boxes, and no loose boxes) fall on top, questions about the possibility of returning to the supplier and so on. So I wrote a little more about the work of the purchaser . In general, the result is a large complex model.

Large model deals with the purchase.

And this model confidently takes over one accident from any of the suppliers' cars.

To prevent this from happening, the network is equipped with a “capacitor” at the entrance, which smoothes out the roughness of the real world - its own warehouse, which collects everything from suppliers and delivers it inside its network. It turns out more stable and with more clear areas of responsibility. But still, the risks of suppliers in the spirit of "we have a marriage throughout the party, and now oops" are in the procurement. Procurement acts as a control: for example, it calls: “Did you send us a car? Does he know exactly where to unload? Will he arrive on time? ” For large networks, for example, the purchase signs contracts with completely draconian measures in relation to any shoal - this is their way to stabilize the model.

As a result, all the same, some goods simply will not be on time, and this is a law of the world. But it can be influenced.

Fortunately, in the real world, the task is not reduced to “100% availability always,” but conditionally, in order to satisfy demand with a probability of 95% by spending a minimum of money. You can catch up to 96, 97 or 98, but each new percentage will be more expensive, and almost an order of magnitude.

Is it possible to achieve permanent availability of the most important goods?

Yes. Otherwise, sometimes in fast food there were no hamburgers, for example. We, conditionally, do not need to comply with 100% availability of goods. It is enough to set the SLA (SLO) by the presence and hold it. We have a product, for example, which gives 1% of the revenue of the entire network. This is the one and only box of the Jackal.

Now imagine that we removed the Jackal. How much do you think we will lose? All 1%? 0,7%, because those who have come will take something else, for example, Imaginarium? Not. We will lose 1.5–1.6% of revenue, because those who came for the Jackal take something else from the goods, finishing it up with an average check of 1.6 positions.

If you remove one product from the top of sales - you lose money. More than they would give you sales of this product.

That is, in our interests to keep the availability of this product and the like. For example, those 5% of goods that give 20% of the network’s revenue. Or those 20% of goods that give 80% of network revenue.

But this is not enough. The range is not only revenue. This is another reason to go to the store. The larger your range, the more people will be at the entrance. And vice versa. Few people come just to pop around, and you already covered them all. On the other hand, it is impossible to keep in stock all the goods of the price. Therefore, smart people have developed two models: ABC analysis and XYZ analysis.

And this is the beauty of mathematics.

Now how to determine what you need to purchase and keep in stock?

First, an ABC analysis is done on revenue - this is an estimate of how much money the product brings to you (for example, A: 20% of the range; 80% of revenue; B: 30% of the range; 15% of revenue, C: 50% and 5%, respectively). Then an ABC analysis is made for profit (you can sell a bunch of Monopolies, but you can earn a little because of the low margin). Then an XYZ analysis is done, it is an analysis of the stability of demand (and its predictability) - for example, the same Uno can be asked very, very often, but it brings quite a bit of money (because it is small and cheap).

The result - you get for each product category in terms of revenue, profit and availability. For example, AAZ (A is a large share in revenue, A is a large share in profits, Z is very poorly asked) is a product that does not sell itself, but will bring you a lot of money. This could be some Space Alert from our games, for example. AAX is Jackal or Imaginarium, a product that is profitable to sell, and people want it. Something like ABY (a lot of profit, average revenue, they normally ask) is, for example, Genga. It should obviously be in the store, but if it is not there, the loss is not so great (because there are many towers in this particular segment). And so on, here are some more examples:

• AAX - the best products in the world for you.
• ACX - products that you attract people, but almost do not earn them.
• CCX - it can be any cheap type of battery garbage, for example: it needs a lot, but in the finmodel it is not visible.
• CAX - products that ask, on which you make a huge margin, but which cost little. We have this packaging (it is worth the time of the seller in fact).
• CAZ / CBZ - products that no one asks, which bring you a large margin per unit, but almost never sold. It may well be that this is some kind of super-megadelux box set up for contrast with the price of others, or in case of a sudden appearance on the threshold of the oligarch.

Further, after such an assessment, it is necessary to take and purchase all products that have the letters A and X in the analysis. And install SLA on them.

At this moment, as experience in the regions shows, the store matrix will suddenly straighten out, and the magic of purchases will increase point revenue by 15-20% for sure. Because it’s rare for someone to do this with their hands. It would be too clever, logical and correct, so they don’t.

Then you need to closely look at those where the letters B and Y are. You should try to get them, but not to tear the navel for this.

What's the Difference? The fact that when Vladivostok was left without goods for the New Year, I would probably recommend “good” letters A and X to order air delivery to the edge of the country, then B and Y are unlikely.

Then you need to try to get rid of the goods that you have been hanging for years - the genes C and Z in the diagnoses. More precisely, reduce their stock to the minimum possible. Naturally, for today, because after cutting the matrix, new C and Z will appear again. The analyzes are comparative. But this process (as well as the process of entering and evaluating a new product) should not be stopped.

Where do these reports come from?

To get the ABC analysis and XYZ analysis, you need to have some kind of sales statistics, where the entire range already exists, and everything works correctly. Well, or use mathematics, allowing to repair holes in the presence of something. To do this, there is a special software, and various methods, often associated with demand forecasting.

We use our math and our crutches (this historically). We have two main models: a matrix with letters for Moscow and a matrix with letters for regions (still, these are two markets). There are under each store, but not in real time.

Further simple action is this: take the first 100 ABC products. What goes there should be ironic, otherwise the store works with low efficiency. More difficult action - to take the ABC / XYZ-analysis on the region. And finally buy this product. It is important that in X and A there should not be a single hole. It is believed that the purchaser does the work poorly if there is an exit for the SLA, for example, in the absence of any A product during a season for any reason (including disasters and wars) for more than 1 day. Simply, it must be obtained at any cost from any supplier. If you still have a hole in A - you just lose money every day.

There is another feature with AAX products: they can be stored in the warehouse for as long as possible. There is no reload on them, because it is precisely known that they will be sold. Products such as CCY are questionable in this regard, you can gnaw cardboard with them (and sometimes it is cheaper to throw them out than to keep in stock until the party leaves).

For now. I remember that I promised to show more what happens in the display specifically, and why it is important. In particular, we are interested in a number of goods that are not for sale, do not generate revenue, but can work in a common model. About this next time.