How many SKUs should a demand planner manage?
In short: 100,000+ SKUs per planner.
Lokad employs a team of ≈30 supply chain scientists (and it’s growing). Each scientist is essentially acting as a supply and demand planner, although the role goes well beyond the planning angle. The scientist is responsible for the data pipeline, the forecasts1, the economic modelling and the end-game decisions2.
Each supply chain scientist manages:
- 100,000+ SKUs
- 50+ million USD of stock
- A full daily refresh
These numbers represent the cruise speed that Lokad reaches 6 months to 1 year down the road after starting a supply chain initiative with a client. Furthermore, supply chain scientists don’t grow on trees. Training a junior supply chain scientist takes 3 to 6 months. Lokad typically hires engineers, although not necessarily supply chain engineers.
Half a dozen of our most talented senior scientists individually manage:
- 1+ million SKUs
- ≈500 million USD of stock
- Multiple daily pipeline refreshes
Our supply chain scientists try to maintain a 80/20 work ratio: 80% of the time spent on implementing and improving numerical recipes, 20% spent discussing and getting feedback from the client company.
In order to achieve this productivity, a series of simple elements are involved.
Defragmenting the decisions. In many companies, figuring out the right quantity to put in a purchase order involves many people: the planner, the inventory manager, the ERP specialist, the BI specialist, a supervisor, etc. By putting all those elements back in the hands of a single person, productivity increases massively. This is exactly what Lokad does through the role of the supply chain scientist.
By way of anecdotal evidence, during the last decade, I have met several founders turned CEOs who’ve told me almost the same story: During the first 2 decades, I was managing all inventory decisions on my own on Saturday afternoons. When the business reached 50 million USD, I hired my first supply chain manager to handle this for me. We are now at 100 million turnover, and it takes 5 full-time employees to do the job.
Correctness-by-design at the platform level. The majority of supply chain woes are mundane: incorrect dates, incorrect stocks, incorrect constraints (e.g. MOQs), IT changes, biases of all kinds, etc. Programmatic expressiveness is needed, and this is why spreadsheets are so effective (Excel is very much programmatic). However, as spreadsheets offer little in terms of correctness-by-design, mundane glitches remain all over the place. Teams spend all their time firefighting, processing alerts and exceptions, instead of improving the numerical recipes themselves. Eliminating this entire class of problems is exactly what the Lokad software platform is about.
Real-world optimization paradigms. The bulk of the classic supply chain theory (point forecasts, safety stocks, EOQs, min/max, ABC analysis, etc.) is not appropriate for coping with real-word supply chains. Those elements look great on paper and work poorly in the field, leading to even more firefighting. Probabilistic forecasts, algebras of random variables, differentiable programming, array programming, data versioning, … are as many paradigms that are essential in practice to achieve real-world results. Being a one-stop shop for all these paradigms is exactly what the Lokad programming language3 is about.
On a tangential note, when we negotiate a monthly subscription fee, we try to make sure that the client company can offer us a single point of contact who will act as the coordinator for the whole initiative. Indeed, while we are not philosophically opposed to talking to sales, finance, marketing, production, … there are limits to the degree of productivity that can be achieved in this sort of work no matter the tooling involved. We try to keep our supply chain efforts focused on where Lokad delivers the most.
There are multiple kinds of forecasts involved. The two most common are the demand forecast and the lead time forecast. However, other sources of uncertainty such as returns or production yields may have to be forecast depending on the context. ↩︎
There are multiple kinds of decisions involved. Purchase orders, production orders, dispatch orders are the most common ones. However, other decisions such as price steering, stock divestments or assortment optimization may also be covered. ↩︎