Inventory optimization for SMB



Small and medium businesses (SMB) dealing with a physical flow of goods may not have a large supply chain network to manage, but they certainly need to get their stock levels under control - especially when growing fast. Inventory control is a twofold problem: first, asset management, second, stock optimization. Lokad delivers the latter through its predictive optimization technology. Think of Lokad as your inventory copilot that tells you when to buy, how much to buy, where to dispatch, and what to do about those slow movers that might turn into dead inventory if nothing is done about them. Our technology is precisely designed to cope with inventory rotations that are both limited in volume (few units sold or bought in the first place) - and in depth (with a history much shorter than a decade).

Drawing of fashion models


There are tons of inventory apps out there that look good and yet fail to deliver any added-value compared to a nicely organized Excel sheet. If your app does not deliver an inventory performance that would be impossible to achieve through Excel, you should really question why you need such an app in the first place.

Joannes Vermorel, Founder of Lokad



Your predictive inventory copilot

Stock levels are a balance: too little and your customers are not properly served, and too much and your carrying costs skyrocket. As soon as an inventory management software is in place, there is an opportunity to start optimizing the stock levels. Lokad delivers this through a predictive optimization, which means lower stock levels, better service, less dead inventory, and increased productivity. This is particularly critical in smaller companies that cannot afford a large clerical staff to deal with mundane operations such as replenishments. In particular, in young, dynamic companies, time liberated from tedious tasks tend to fuel an even faster growth.

Usually, we start by addressing the problem of “when and how much to purchase”, by generating a daily report containing the suggested quantities to reorder. This exercise may entail many subtleties, such as ordering schedules, supplier MOQs, price breaks, inconsistent lead times or multi-sourcing. Lokad accommodates all these constraints and more.

However, depending on your needs, Lokad also addresses a variety of inventory-related problems such as:

  • Pulling back stocks from FBA in order to avoid long term storage fees
  • Balancing the stock between two or more locations
  • Evaluating reliability of suppliers with scorecards
  • Deciding to keep an item in stock or not, and whether to serve through dropshipping
  • Identifying slow movers and promoting them to avoid dead inventory
  • ….

copilot's hands on a cockpit

Lokad delivers a “software + service” combo. When you subscribe to our managed services, a Supply Chain Scientist is assigned to your account. They take care of turning your historical data into actionable figures such as suggested reorder quantities. Under the hood, this expert leverages our webapp and Envision, a domain-specific programming language dedicated to predictive supply chain optimization and makes sure you get the most out of Lokad’s technology, without you having to turn into an AI/IT expert.


You remain in the driver's seat but have someone who translates your business expertise into code. Critical knowledge is automatically turned into automatic decisions delivered by a system that can be further refined by multiple contributors over time.



Beyond classic forecasting

Forecasts looking only at a single averaged future work poorly for SMBs. Inventory costs are concentrated on extreme situations: stock-outs happen when the demand is severely under-estimated, and conversely dead inventory happens when the demand is severely over-estimated. In between, inventory rotates gently.

Yet, the vast majority of the software products on the market completely miss the point, opting for classic time-series forecasts, which are unfortunately not suitable to cope with the problems faced by SMBs.

Lokad features a probabilistic forecasting technology: we assign a probability to every single possible future. We forecast not only demand, but also lead times and returns whenever it’s relevant. More generally sources of uncertainty need to be forecast.

Probabilistic forecasting vastly outperforms outdated approaches like classic safety stock calculations, oversimplify reality and work poorly when demand is either intermittent or erratic, which is often the case for SMBs. Assigning a probability to each possible future - i.e. quantifying the harm that extreme scenarios could bring -, is the first step towards performing a predictive inventory optimization.

The second step consists of looking at all possible options, for example, every single quantity to be reordered - unit by unit. We do not think of replenishment through a min/max policy per SKU. Instead, we look at all SKUs, that is, we look for the one single extra unit of stock that will bring the most return on investment for the company while accounting for operational constraints as MOQs and batch sizes.

Which brings us to the final step: the economic prioritization. The opportunity to buy every single extra unit of stock should be assessed in dollars or euros of profits and losses. We refer to these factors as the economic drivers: gross margin, carrying costs, stock-out penalties, etc. The final result of the optimization is reorder quantities that are completely aligned with the uncertain futures and the strategy of your company.




Lean inventory performance

For replenishments, the Lokad webapp delivers a tabular report that gives you exactly the quantities that you need reorder today, plus, the KPIs, in dollars or euros, that explain why those quantities.

This report can be accessed through a web dashboard, downloaded as an Excel spreadsheet, or even scheduled for automatic import into your ERP.

Lokad delivers numbers that do not need any further post-processing, and no further manual tweaks of any kind. Achieving this feat is a twofold challenge; it requires:

  • a technology featuring state-of-the-art probabilistic forecasting models and numerical optimization solvers.
  • a talented expert who builds the end-to-end numerical recipe leveraging your historical data, and mitigating all the unavoidable data issues.

Indeed, many classic planning solutions are the opposite of lean: historical data needs manual “cleaning”, forecasting models need manual “tuning”, ordered quantities needs manual “tweaking”, etc. All those operations treat your staff like consumable resources.

Lokad delivers the opposite: efforts are invested and capitalized into bespoke numerical recipes, that are just right for your company.



Classic solutionIssues with the classic solutionSolution adopted by Lokad
Classic forecasts (i.e. daily, weekly, monthly averages) Doesn’t work for erratic or intermittent demand Probabilistic forecasts that embrace uncertainty
Forecasting model tuning and forecasts editing Very time consuming, payback is very poor Self-calibration of machine learning models
ABC analysis Crude categorization of SKUs, tons of edge cases Embraces the full complexity of every single SKU
Min/Max inventory method Ongoing generation of dead inventory Prioritizes every single extra unit of stock against its ROI
Safety stocks Unsafe approach that does not ensure fill rates Robust optimizations reflecting direct stock-out penalties
Configurable lead times Doesn’t work on varying lead times, and time-consuming Learns and forecasts lead times with probabilistic forecasts
Configurable stock covers (i.e. days of stock) Alternative unsafe approach that does not ensure fill rates Economically optimize stock covers per SKU
MOQs, price breaks, multi-sourcing Not supported, time consuming manual overrides Native support through bespoke logic
Configurable per-item prices Demand is assumed to be independent from price Demand forecasts that leverage product prices