This is the final text from our series about what is most important for a modern e-commerce business that faces rapid assortment changes, daily price jumps and diabolic competition.

Previously on:

1.  Assortment matrix formation

2.  The pricing policy

In the previous episodes, the plot developed as far as an ideal assortment, ideally priced, enters your central warehouse.

This is already a powerful success.

However, there remains the Last Mile problem; from the central warehouse, the goods must be optimally allocated among your retail outlets.

Disclaimers.

1. The Retail Outlet is not to say that topic is limited to retail only. On the contrary, all the approaches and algorithms that will follow are quite versatile and also apply to wholesome and any geographically allocated network sales.
2. If your logistical scheme excludes a central warehouse (the suppliers ship their products to the outlets directly), this text remains relevant — for you still have to form orders for the suppliers to make shipments to specific shops.
3. We assume that the reader uses an IEM System (or those inherited from it). Or any other system that is close to it in terms of functionality and possibility of customization.
4. The practical approaches to be suggested below are applicable to big retail networks with dozens, hundreds, and thousands of shops offering a broad assortment, which includes thousands and tens of thousands of mottled goods items that change quickly. In less complicated cases, our recipes may quite probably be redundant.

As is usually the case, you need to strike a balance between the mutually exclusive requirements of maximum assortment breadth + depth sufficient for maximizing sales vs. acceptable inventory turnover rate, the need to prioritise the supply of well-developed shops that generate the bulk of sales/profit vs. maintaining the best assortment at newly opened and puny shops that produce nothing as yet but an immobile inventory and losses.

To build a good merchandise allocation system, one should address a set of tasks, among which we believe the following to be the main ones:

1. Defining each shop’s assortment matrix, i.e. "what there should be".
2. Each goods item’s assortment depth; "what and how many/much there should be" at the shops, separately for goods with a sales history (and similar goods) and for brand-new goods items.
3. The rules for allocating scarce goods items, particularly across the Developed vs. New Shops dichotomy.
4. What do we do about non-sellers?
5. Goods re-allocation among shops.
6. Advance orders service: its applicability limits and hidden risks.
7. The authority boundaries between the automated system’s features and the specialized human resource. The interaction schemes and material incentives for the personnel involved.

Here we mean that a decisive feature of a good quality system is the situation where the system processes 99,99% of the regular goods flow volume in an automatic mode, while manual interference overriding its formal algorithm only occurs in very rare emergencies (like a truck turning over and the goods lost, or an armed assault on the delivery man).

And now, point-by-point.
 

1. Defining Each Shop’s Assortment Matrix, i.e. "What There Should Be"

In most cases of a chain of shops within one agglomeration (e.g. Greater New York), an individual shop’s assortment matrix will not differ from that of the company as a whole.

Alternatively, the business" geography will be divided into regions, e.g. New York City and State, California, Idaho, etc., whose assortment matrices will be formed independently using the process described above.

Then an individual shop’s assortment matrix (AM) will be inherited from its region’s base AM (with individual differences possible at the local level that will be dealt with below).
 

2. Assortment Depth

Goods with a sales history and new ones are processed using different logic.

In the former case, calculating the Sales Rate (SR) is indispensable.

In the most basic case, this is calculated as the goods volume sold during a period divided by the duration of the period. The resultant value in pieces (kg, m, l) per day is then reduced to the base duration of the industry’s business cycle, mostly a week for trade.

We thus obtain the number of pieces of an individual goods item that must be brought to a shop for a week of sales.

However, the SR thus calculated will only be relevant provided that the goods item is constantly available in stock and for sale throughout the period for which we calculated the turnover rate (so our reference to the most basic case really make sense).

For example: we calculate the Apple iPad sales rate on the basis of February 2018 data.

Six pieces were sold that month, so the weekly sales rate is 1.25 pcs. per week.

But if we go deeper we’ll see that the Apple iPad goods item appeared at our warehouse in the morning of February 10, and four were bought on the first day, while the remaining two were reserved and invoiced for a cash-free payment, which was remitted on February 13, and the shipment was made on February 15.

The SR calculated using the simplest algorithm will obviously be correct in mathematical terms but absolutely useless practically.

Moreover, it will be subversive, for if you assume that just 1.25 pc. will sell on an average week, it is no good bringing in more than two. So we’ll never sell any more — though if we go into details we shall see at once that we sell dozens of those iPads. The lost profit is easy to figure out.

In this case, our problem is that with tens of thousands of goods items on our price lists, "going into details" is hardly possible.

What shall we do?

First of all, we divide the whole set of the goods items with sales histories into two parts: the goods that were constantly in stock and those that were intermittently available.

The former case is quite simple; an elementary calculation will return an adequate SR figure.

For sellers of electronic stuff, this set of simple goods largely includes the Accessories category: keyboards, mice, cases, mobile phone cradles, etc.

No assortment problems arise with these goods categories.

However, problems with the second subset, which includes the demand-generating goods items like those fashionable tablets, notebook PCs, mobile phones, etc., occur quite regularly.

For such goods, IEM System has to calculate the "integral" (our term) sales rate.

This is rather harsh mathematics that need not be thrust on the business reader. To put it in a nutshell, the system will decompose an article’s whole sales volume over a period to each individual sale document, analyse how long each individual piece was on sale, and then use numeric differentiation methods (plus fluctuation smoothing) to calculate the ideal sales rate for the constant availability case.

Although the integral SR for intermittently available goods provides figures much closer to reality than straightforward average-based calculations, this accuracy is usually insufficient.

The next tier of the integral SR refinement business logic comprises all sorts of cascades of if-then operators, simple both in terms of sense and algorithmically, but 100% specific to industry conditions.

Now for new goods items.

There are no universal solutions how to calculate the quantity of any new goods required for a shop accurately enough.

Nor can there be one.

Nevertheless, the following techniques may be rather efficient as applicable to some industries" market specifics:

1.  A "new goods price cap —­ number of pieces to be first shipped to a shop" matrix is created in the system. For example, we send four pieces of something that costs $100 a piece, one $500 item, and if something new costs more than $1000, developed shops will get one piece while new shops will collect advance orders.
2. We proceed from analogues: we select two (or any number of) goods items most closely priced and
   with sales histories and take their average sales rate.

3. We proceed from the new goods item’s popularity in the most relevant digital marketplaceaggregator top rating (see the assortment matrix formation case for details). For example, we bring 10 pieces of a Top 3 item, six pieces of a Top 10 item, two of a Top 20 item, and just one piece, for trial purposes, of something not rated. To be more efficient, we combine this with the (2.1) price matrix.

4. And so on, and so forth.

The diversity of empirical techniques similar to the above-cited ones is unlimited. The specific set of the techniques or combinations thereof is fully determined by industry and regional nuances and assortment specifics.
 

3. The Rules for Allocation of Scarce Goods Items, Particularly Across the Developed vs. New Shops Dichotomy

Like anything else in our naughty world, the development of new shops (given the limited resources of a self-made business as opposed to bubble companies being inflated with external investment) requires long-term goals to prevail over short-term ones.

In the assortment context, this point means that new shops" assortment should at least be no worse than developed ones".

At least.

And to be equitable, we should first supply scarce goods to new shops and then to all the rest. Of course, flagship shops" sales will suffer in this case, while the scarce goods will lie longer at puny ones.

And your product managers" hearts will bleed and belch out bile.

Are you not ready?

Then it’s no use spending resources on network expansion. Nothing good will come of it.

So, after we have supplied our new shops with scarce goods on a priority basis (in quantities determined using e.g. (2.1.) logic or any other), the remainder (if any) must be allocated among our standard shops.

We say straight off: the inter-shop "quota" solution that suggests itself belongs to the "simple, obvious, and wrong" category. Like any other "planning" sub-species.

We suggest that scarce goods should be allocated with maximised profit in mind (the sales maximisation logic is senseless in respect of scarce goods, which are scarce exactly because they sell out fully and quickly).

In retail, maximised profit is in 99% of the cases based on maximised sales of complement products.

We should remind you that (a little less often than always) an assortment can be divided into goods generating customer traffic and sales (locomotives, attractors, base goods — they may be termed differently) that sell at a minimal profit rate and often at a loss, and related goods that are cheap as compared to the base goods but produce a mad profit rate.

Profit is made by selling the latter, but they go "attached" to the attractors.

Examples are: a mobile phone and a case for it. A notebook PC and a bag. A washing machine and an "extended warranty" certificate.

For obvious reasons, virtually always it is the base goods that are scarce.

So, to maximise profit, we have to allocate our scarce base goods, first and foremost, to the shops that are the most successful in selling their respective related goods.

Success may have different mathematical definitions, but the ultimate effect will not very much.

For example:

Suppose your network includes three shops. Mobile phones of the X model are scarce now.

Then, the IEM System robot that forms internal goods movements from the central warehouse to shops will come to the X goods item, find its quantity to be insufficient for all the shops, and then proceed to calculate a period’s (e.g. last month’s) sales volume of mobile phone-related goods (for all models, not just X, to smoothen local fluctuations) at each shop, relate the former to the latter (to obtain three quantities of the USD/piece dimension, as many as there are shops), sort the shops in this quantity’s decreasing order and mutually normalise the three values so that they add up to a unity, which will result in a ratio like 0.55:0.32:0.23.

And exactly in this proportion the scarce X article will be allocated among the network’s three shops.

Instead of the ratio between related goods revenue and the base goods sales count, we can use the quantity-to-quantity, sales-to-sales, profit-to-profit ratios plus any mixed options, with the optimal candidates for the numerator and denominator depending on the your individual network’s assortment specifics.

Still, the final result will not differ much.

Instead of proportionate allocation of scarce goods, we can use the "winner-takes-all" principle: the shop that leads the sales of related goods has its need for the scarce goods item satisfied as much as possible, and the remainder (if any) goes to the second rated shop within its calculated need, and so on.

Any other formalisable algorithms for allocation of scarce goods are possible, but the above ones are optimal, in terms of both common sense and economic logic, for most practical applications.
 

4. What Do We Do About Non-Sellers?

There is a practical case about how to efficiently address the problem of non-selling goods.
 

5. Goods Re-allocation Among Shops. The Authority Boundaries Between the Automated System’s Features and the Specialised Human Resource

The issue of goods re-allocation among the network’s shops usually arises where

(a) some outlets and the central warehouse have run out of the X goods item while at others it lies still or does not sell at a satisfactory rate, and/or

(b) your logistical capability enables your business to move the goods without much red tape or high explicit costs.

At first glance, the solution is obvious: the system re-calculates the needs and generates the waybills required for internal goods movement; sales and profit increase, and turn-round grows with working capital efficiency.

But, at second glance, such practice virtually eliminates the shops" responsibility for their assortment.

Suppose that this responsibility is at least officially declared.

The shop manager’s optimal modus operandi in these conditions is to drag as much everything as possible to his shop; what won’t sell will be taken back.

It is easy to guess how such a behaviour pattern will affect the financial performance of the business as a whole.

Quite otherwise than it seemed to at first glance.

Also, well-oiled logistics that permits such complicated operations without headache for their participants and beneficiaries has its dark side: all that internal transportation, however brilliantly organised, costs money. Which costs, in turn, are so difficult to correctly charge to the shops that in the practice they are nearly always borne by the central office (as part of the business expenses as a whole).

This results in virtually runaway and stupid logistical costs — "privatised profits and nationalised losses".

In general, where it comes to re-balancing assortment among our shops by means of internal transportation, we observe again the usual antagonism between the short-term and long-term interests of your business.

The Opinion (in the Spherical Vacuum Perspective of an Abstract Retail Network and the Priority of the Long-Term Interests of a Business):

- if you regard your network as a sum total of shops that are also business units managed by responsible businessmen, then you must renounce assortment levelling. What gets into a shop must be sold there. Non-sellers and sales at a loss should be reflected on the shop’s general balance sheet and profit and loss account and on its manager’s income in particular;

- but if your network comprises simple sales outlets fully managed from the headquarters (and, consequently, the goods are allocated by the automatic functionality of an system — because it is physically impossible otherwise), then levelling is indispensable, or it is the only available tool for minimizing the consequences ​of inevitable projection errors.

All the trade networks known to be successful belong to the first type, with minor variations. On the other hand, the technology underlying second type networks only appeared some years ago.

It is for the shareholders to choose how to build their business. We shall dwell upon this issue in a somewhat different perspective in the next section.

And now a few words about the sale of all sorts of substandard goods, including those in damaged packing or otherwise in a non-marketable condition, returned from warranty repair, etc.

We believe that after being duly re-priced and separated into a separate goods item (like MARK-DOWN, lady’s high boots: a scratch on the heel" in the Discounted Goods category), problematic goods must be sold at that very shop where they first appeared.

The verbally tempting idea of a single Discount Centre always degrades, in reality, into a more or less gigantic hole in your business where all the crap is gathered from the whole network and whose cheerful personnel bury the problems that they are not interested in addressing or being responsible for.

Abundant theft is a free attachment.

Suppose an appliance is returned on warranty. It is repaired, then taken back to the shop that first sold it, and now let them try and persuade buyers, offer a discount — in short, do something about it. The shop manager, who subsists on a percentage of the profit in some form or other, is interested in solving such problems with minimal losses for the shareholders.

And to organise a single virtual discount centre on your website, for your customers" convenience, is no problem at all. After ordering a goods item on your website for pick-up, the buyer will collect it at the shop where the required sub-standard item is physically lying.

And if s/he orders goods with delivery, no questions will ever arise.
 

6. Advance Orders Service: Its Applicability Limits and Hidden Risks

In general, the inherent risks of the advance orders system are ultimately of the same origin as the problems with automatic levelling of shops" assortment, examined in the previous section, namely: weakened local personnel’s motivation and responsibility for working with the shop’s assortment.

What’s the sense of all the fuss, if they may tear anything out of your assortment at any time? And usually something that you hoped to sell at a good profit. Not a non-selling item, anyway.

Although, again, in the short run the advance orders service boosts sales by expanding the available assortment. The effect is especially visible if the assortment was initially fragmented.

And in the long run the assortment will ultimately become even worse.

As will your sales, of course.

This is the same antagonism between the short-term and long-term goals.
 

7. The Authority Boundaries Between the Automated System’s Features and the Specialised Human Resource. The Interaction Schemes and Material Incentives for the Personnel Involved.

Here we shall avoid describing different models, for the material to be sited is endless.

We shall proceed right away to describe our vision of an ideal model.

For clarity we’ll take a simplified example of a retail network comprising just two shops (Shop1 and Shop2) and the central warehouse (CWarehouse), to which the headquarters, or the head office, is logically attached.

The network’s assortment, in turn, consists of two goods categories: GoodsCatA and GoodsCatB.

Sitting at the headquarters are two purchasers/category managers: PurchCatA and PurchCatB.

They are responsible for the sales of their goods categories.

At each shop there are two local category managers (LocalCatA/Shop1 and LocalCatB/Shop1, LocalCatA/Shop2 and LocalCatB/Shop2) who are responsible, respectively for the sales of those same categories at their shops.

The assemblage of a central category manager and his shop-level reflections constitutes a category team (or goods line team). In our case, there will be two teams, comprising::

TeamA = PurchCatA + LocalCatA/Shop1 + LocalCatA/Shop2

TeamB = PurchCatB + LocalCatB/Shop1 + LocalCatB/Shop2

The local category managers are the nodes of an organisational matrix structure; administratively, they report to the shop manager, while functionally they belong to the category team and report to their respective pcm.

In a correctly organised system, all "reporting" boils down to human resources decisions following a "two keys" logic: the candidate must be satisfactory to both the shop manager and the head of the category team.

Functionality

The head of the category team (besides procurement proper, here we talk about goods allocation only) sets up the base rules for forming internal goods movements.

The shop-level categorist adjusts them for his shop.

Neither may physically get into the waybills and massage them manually; they may only alter their system-wide formation rules.

Only the shop manager or chief procurer may get into the waybills on force majeure occasions.

Motivation.

The shop-level category manager:

a % of the category’s gross sales profit (a salary replacement in a sense. Also included are losses from the sale of all sorts of substandard goods) × assortment quality factor

+

a fat bonus for higher than standard profit rate in the category (a profit rate higher than usual is indicative of good work by the goods line team to which the that shop-level category manager belongs).

–

a malus for the category’s standard profit rate not attained (no explanations required)

–

payment for the working capital immobilised in the goods in stock at the shop (based on the period’s average inventory, calculated as a percentage of the amount in stock — IRR, commercial lending rate, or an suitable arbitrary rate)

The central category manager — head of the goods line team:

• exactly the same duties as the shop category manager’s, but he is responsible for network-wide performance.

Again, to avoid being too verbous, we shall not ground this exact choice of KPIs here. We shall just say that it is NOT the ultimate truth and NOT the only working option.

And now a couple of words about a seemingly non-obvious but exceptionally valuable practical consequence of this: a self-learning system.

Any positive innovation found by one shop’s lcm is automatically reproduced by others. For it is in the direct interest of both the central product manager (information exchange centre) and other shops" local product managers.

This is easy to compare with the ordinary administrative pangs of implementing innovations in the "revolution from above" format.

Also, being able to swap lcm’s both between shops (at least within an agglomeration) and between goods categories, company management can try cheap and illustrative experiments to detect the root causes of some problems: is the shop really poorly located? Or is it the local personnel’s failure? Do teapots really fail to sell at our company for some mystical reasons or is the respective pcm underqualified?

The above theoretical scheme can be applied to any number of goods categories and shops.

If, as we have already written in the first section, the outlet geography includes regions with vastly different market conditions, then the two-tier "central product manager — shop product manager" system will be transformed into a three-tier one, "central product category manager — regional product category manager — shop product category manager".

For convenience of administration, the number of each node’s subordinate links should not exceed ten (ideally, seven).

A scheme with more than three levels is inefficient in the practice.

And now let us apply the theoretical scheme to practice and discuss the most frequent difficulties.
 

7.1 Situation: a big network with a central warehouse and regional second-level allocation warehouses. How do they fit in?

Solution: managing the assortment at regional warehouses is the competence of regional product managers. In this case they are positioned in relation to the central product category manager exactly like the shops" local product managers are in relation to themselves.
 

7.2 Situation: pure arithmetic says that following the principles of "not more than three levels in the scheme" and "not more than ten persons reporting to a mode" limits the number of your shops to 1000. What do we do if we have more?

Solution:

• in the case of a network comprising big shops, we most probably deal with a trans-national company (and how many IKEA shops are there in the United States?) that is organised as a holding company in which the operating companies catering to local markets are responsible to the headquarters in terms of financial performance only.
• There just a few cases like Walmart worldwide. But, on the one hand, they have little to do with the e-commerce business issues under review and rather belong, with their base assortment, to traditional retail (where things are much simpler); and, on the other hand, those super-powerful guys don’t need our advice at all.
• A network of thousands of small shops that look more like kiosks, is quite a real-life situation. But as those outlets are often located opposite one another and their assortment is strictly standardised, optimising supply at kiosk level makes no sense. In this case, it is reasonable to introduce local product managers at the level of, say, a town/city with its district/region, with the exception of a few big cities. These will be interpreted as "shops" in the theoretic model’s terminology.
   

7.3 Situation: we have several dozens of goods categories. Should we seat several dozens of product managers at every shop? There are just as many workers there now, including loaders.

Solution: we have already discussed that category managers, both centrally and at the shops, must optimise the rules RATHER THAN create waybills manually. After they are trained and things get streamlined, this will not take more than half an hour a week.

This is exactly why the lcm role is nothing but additional workload (only voluntary, of course) on the shop’s personnel already on the staff. And the smartest and most resourceful ones, for obvious reasons.

Incidentally, successful lcms are a ready-made internal human resource for promotion the central product managers" posts. Cheap, 100% trained, and with proven efficiency. Ordinary salesmen can thus see their career prospects and what they should do to attain them.

As their work volume is smaller than that of a central product category manager, one lcm at an individual shops may cater to several goods categories. Moreover, at different shops the sets of goods categories in charge of an individual lcm may be differently re-assembled.

Most importantly, each central product manager must know which specific person he/she can contact at a specific shop, and vice versa.

Moreover, not all goods categories can be included in this scheme at all. Only the most important and/or problematic ones, where you expect the greatest effect — and at the protracted stage of starting and debugging business processes this is indispensable.

 A sine qua non condition is that a local shop’s assortment should be divided at least into two parts at least one of which should be assigned a special lcm.

We’ll say straight off that the formal appointment of a "single" lcm responsible for a whole shop’s sales and assortment is a sham. Which, depending on how profound the initial assortment catastrophe was, may produce tangible benefits at the first stage, although in the practice you will most probably get an additional sponger. Strategically, it is a dead end — sure as hell.

This seems about all the big things.

Much, difficult, complicated?

Quite so.

Start-up alone will take you several months and a lot of nerves and time for training and persuasion, finalisation, honing, answering idiotic objections like "that won’t work" and so on.

Probably, even more than one year before you reach the stage of being fully able to take advantage of the proposed scheme.

However, your inspiring goal is a self-supporting, self-learning and highly profitable organisation that ultimately requires one thing of top management: not to spoil.

To finish this extensive narration with a parallel, let us remember orthodontics in dentistry.

Malocclusion rarely manifests itself as uneven teeth over somebody’s lips.

The problems that it causes are usually delayed ones — for years and decades.

On the contrary, treatment is obviously nasty, with the extraction of teeth, wearing dental braces, and regular visits to the orthodontist.
 

But...