A Data-Driven Model for Inventory Planning in a Multi-Site Food Manufacturing Network

Abstract

The thesis addresses inventory planning in a multi-site food manufacturing
network where inventory decisions are currently made using differing local
routines, planning parameters, and levels of analytical support. The single-
company case study is conducted at a bakery subsidiary of a large agricultural cooperative based in northern Europe that operates over 30 bakeries. As a result of recent bakery acquisitions, routines, software usage and ordering decisions aren’t standardized across the network. The purpose of this thesis is therefore to develop a data-driven decision-support tool that facilitates more standardized ordering decisions.

The thesis follows an adjusted operations research framework spanning problem formulation, data analysis, model construction, and backtesting on unseen sales data. Weekly sales histories for products in Denmark, Norway, and Sweden were filtered based on data availability, coverage-period usability, and stationarity. After the filtering process, the data was fitted to the Poisson, negative binomial, normal, and gamma distributions using maximum likelihood estimation and evaluated through goodness-of-fit tests and the Akaike Information Criterion. These were then applied to a periodic review order-up-to policy implemented in Microsoft Excel with VBA, and its recommendations were compared with the company’s current planning decisions.

Achieved service levels and recommended order quantities varied considerably between individual products, with some falling below the 95% target
and others exceeding it. At the country level, however, aggregates remained
close to the target and within three percent of planner orders. Performance was strongest for products with stable, data-rich demand. Products that performed poorly were characterized by intermittent sales, shifting demand, and increasing variance. The main limitation to wider adoption within the company’s bakeries is not the modelling approach itself but inconsistencies in master data and planning practices across sites. The tool should be interpreted as a second opinion based in inventory control theory to be used alongside planner judgment and experience.

Popular Abstract

Can a spreadsheet help a bakery network decide how much bread to produce?
A master’s thesis at LTH explored how data-driven inventory planning can bring more consistency to a large bakery group with more than 30 sites across Europe. The result: an Excel tool that standardizes the decision-making process - and reveals where standardization stops working.
Walk into any freezer aisle at any supermarket and you’re met with a choice of hundreds of different breads, pastries and rolls. But behind every product in those freezers sits a tougher choice someone had to make weeks or months earlier, namely, how much of those products should be produced to ensure the shelves stay full without filling the warehouse with bread nobody buys.
For a European bakery group with more than 30 bakeries, that question was answered in more than 30 different ways. Years of buying up smaller bakeries had left every site with their own routines, procedures and experience of what “enough” looks like. One planner’s excessive overproduction is another planner's conservative amount, and nobody could easily tell who was right.
Inventory control within the food industry is a tightrope. Hold too much and you tie up cash, fill freezers and risk throwing goods away, hold too little and customers find empty shelves at their local grocery store. That balance plays out across thousands of products and dozens of bakeries every single week. With energy and transport prices climbing, getting this balance right has become a serious money and sustainability question. To assist in tackling this problem, we built a tool that anyone at the company can run in Excel. It studies the historical sales data for the company’s products, learns each product’s typical demand pattern, and recommends how much to order each production run to hit a chosen service level based on a periodic review inventory policy.
Out of all the products in the Nordic portfolio, there were a lot of products who turned out to have a poor fit for a standardized periodic review model. This had to do with that the demand was too irregular, intermittent, or shifting too fast. When backtesting the model on unseen sales history, the resulting service levels varied considerably between products, but for items with predictable, stable demand, the model performed well. Its order quantities also differed from those of the planners, even when both aimed at the same target service level. The point, however, isn't that the model is always right. It's that the decision-making process becomes explicit and standardized, instead of being hidden in local routines and gut feeling. The tool therefore serves its purpose as a second opinion based in inventory control theory.
The biggest takeaway wasn't about the planning, inventory policies or mathematics. What actually held back a smarter, more consistent way of working was inconsistent data and planning routines between the bakeries. Until that's addressed, no tool can be used widely for all products. Once it is, the tool can do the routine thinking, and the planners can focus on the products that don't behave predictably.



Authors: David Engvall & Felix Malm
Original title: A Data-Driven Model for Inventory Planning in a Multi-Site Food Manufacturing Network
Supervisor: Danja Sonntag Examiner: Johan Marklund
Programme: MSc Mechanical Engineering — MIOM05 Degree Project in Production Management, LTH, Lund University, 2026

• author: Malm, Felix and Engvall, David
• supervisor: Danja Sonntag
• organization: Department of Industrial and Mechanical Sciences; Production Management
• course: MIOM05 20261
• year: 2026
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• keywords: inventory control, demand distribution fitting, order-up-to policy
• other publication id: 26/5337
• language: English
• id: 9228875
• date added to LUP: 2026-06-03 17:13:21
• date last changed: 2026-06-03 17:13:21