v191226

As you know, OpenHealth offers you to work on the HUB from extrapolated data.

Concretely, every day we receive sales from more than half of pharmacies, this is our panel of pharmacies, and we model the sales of pharmacies that are not in our panel.

This allows us to provide you with an extrapolated view of national sales on French territory.

Sector sales are sales at a finer geographic granularity than national sales. For example, cut out from your sectorization.

To calculate the sales sector, OpenHealth conducts a reallocation of national sales area . Thus, if my national sales are 100, I will reallocate these 100 sales units on the sectors, with the cardinal rule that the sum of the sectors must make the national, that is to say 100.

The reallocation of national sales to the sector makes it possible to overcome the geographical dimension by being more specific to the sector (your network or targeting no longer depends on UGAs because not all pharmacies in a UGA have the same potential). This also allows you to follow for example 2 different networks for the same sector or to carry out a follow-up by typology of pharmacies.

It would be possible to extrapolate the sales to the pharmacy, then to add the pharmacies to make sectors.

However, this method is less precise than starting from the national and reallocating sales to the sector.

The reason is as follows:

When I extrapolate to the pharmacy, my margin of error (= 'uncertainty') is relatively high. Indeed, we can very well have atypisms on a single pharmacy taken in isolation (eg: very efficient salesperson, doctor's office next to the pharmacy, etc.). On the other hand, if I extrapolate a set of 100 pharmacies, my margin of error is lower. And even lower on 1000 pharmacies, and even lower on the national one. There is therefore a statistical rule which consists in minimizing the margin of error by first extrapolating the national and reallocating it to the sectors, rather than first extrapolating to the pharmacy and adding the pharmacies together.

We use reallocation coefficients. To put it simply, a coefficient is a percentage of attribution of national sales to the sector. Thus, sector A represents 1% (= the coefficient) of national sales, so I make 1% * 100 = 1

We actually use 2 reallocation coefficients:

Imagine that I want to calculate the sales of my product A, which belongs to the food supplements market, in the Île-de-France sector.

First, I will use the pharmacies in our panel to extrapolate national sales of the product, for example 100 sales units.

Then, I will distribute these 100 sales units over all sectors, including Île-de-France.

For that, I will calculate the product coefficient which will deduct the sales of A in Île-de-France, from the sales of A in my panelist pharmacies . Let's say the product coefficient is 3%.

I will calculate a second coefficient, the market coefficient , still from the panel, but which will take into account the food supplements market. Let's say the market coefficient is 3.3%.

I will now be able to deduce the sales of A in Île-de-France, which will take into account my 2 coefficients, product and market , 3% and 3.3%.

Imagine trying to reconstruct the sales of a low lab DNV , for example Aragan, on a small area, eg: Limoges.

Let us assume that there are 60 pharmacies in Limoges and that our panel has 45 (ie 75% coverage).

Unfortunately, Aragan has only 2 selling pharmacies, in Limoges, and they are not in the panel. (= they are part of the 15 pharmacies that we do not cover).

The product coefficient will give us 0 sales for Aragan in Limoges.

But the market coefficient will inform me that Limoges is an important city in food supplement, and that it is statistically probable that there are Aragan sales .

So my product coefficient will be 0, but my market coefficient will be 5%.

The advantage of having 2 coefficients, product and market , is therefore to better represent the sales of products with low Digital Distribution Sales, improve the accuracy of geographic coverage of products in launches, take into account price variations and have finer temporal granularity, with daily and weekly data.

A IPR (PRODUCT INTEGRATION) , of a product A, consists in adding a product to a market.

Nationally , this addition has no impact on sales of other products.

other hand, this addition may modify the allocation of sales of other products at sector level.

Indeed, an RPI of a product A will not modify the product coefficient of the other products B, C, D, etc. On the other hand, it will modify the market coefficient marginally, because the market definition has been modified by the introduction of product A in my market.

And this explains why the sales of products B, C, D, E will be reallocated in a slightly different way at the sectoral level, and therefore make market shares evolve at the margin.

This also applies to a request for MAM (MODALITY-ASSIGNMENT MODIFICATION) in your market, and even a request for deactivation

Yes, there are several possibilities.

The first is to use only the product coefficient , and not the market coefficient . However, this will result in less precision in the restitution of sectoral sales of products with low DNV.

A second possibility, which we believe is best practice, is to use your sell-in to override the market coefficient.
The sell-in is used to identify atypical customers, atypical geographic distributions, as well as atypical variations (upward or downward).

Limitations to know:

Due to a sales threshold that is too low and / or a very fine sectoral breakdown (for example micro-sectors), the sum of the sectors may not be equal to the national when we analyze whole digits.

For example : 10 boxes sold for a product in a given month to the national to be reallocated on 100 sectors.