Merlin ... No Compromises

Parklea Software Ltd, contact email: harvey@lockie.co.nz

Merlin is a set of sales forecasting algorithms that can be attached to existing Point-Of-Sale or Inventory Control software, to more accurately predict usage.

Objective: Having the right quantity of the right items on the shelf at the right time is critical to success in all distribution and manufacturing industries, strongly affecting cash flow, earnings and customer satisfaction. This is a non-trivial task as sales are fickle, driven by many factors including seasons, product life cycles, new product introduction, and changing customer tastes which in turn are influenced by advertising and short term fads.

Links:

NZ Computer Society research paper

Full descriptive text.

Method: The approach adopted by Parklea Software Ltd was to apply a modified form of Bayesian Analysis, developed by Genetic Algorithms using internal feedback mechanisms, in order to produce at the start of each month a prediction of the sales of every item in every site for the coming month.

The Merlin system 'learns' and scores itself gradually refining it's forecasts and develops a known degree of confidence in each forecast. I.e., on a particular product in a particular site it may know that there is a 92% probability of selling between 900 and 1,100 units in the next 30 days.

The initial development work was done many years ago and by the mid 1990s there were some 1,000 sites in Britain, Australia and New Zealand running the Merlin forecasting analysis on around 10,000 line items every month, giving 10 million forecasts a month. As the forecasting software was linked to the PoS software that Parklea was developing and supporting at the time, the actual sales were observed by the forecasting software and at end of month Merlin was able to compare its forecast for every SKU with the actual sales. This high level of feedback enabled the system to be rapidly refined and unused machine time was used to run 'what if' scenarios, resulting in better predictive algorithms.

For 3 or 4 years (around 500 million forecasts and refinements) the system continued to improve, levelling out with around 90% of items being forecast to good accuracy in retail businesses. Further improvements were then gained by the introduction of Genetic Algorithms to handle site-specific quirks in the sales patterns. These forecasts can be considered by a buyer, or passed internally to existing order generation software to help improve the performance of the orders.

The Merlin Predictive Engine is available as a stand alone module that can be handed a set of sales figures (in most common formats), and will hand back a forecast for the coming month that is around 90% accurate in most retail sites.

By placing orders based on the accurate forecasts typical benefits are:

- It typically requires 15% less stock for the same turnover than when following a computer system with user defined max/min ordering.
- The out-of-stocks are halved in most stores, which increases customer satisfaction and avoids work getting items in for someone (getting items 'in' for a customer causes a lot of work and many customers will not wait).
- The dead stock is usually halved, with reduced obsolescence and better usage of display space.

The total effect of these three factors is to increase the net profit per year by around 5% to 7% of the total stock value in many businesses.

Harvey Lockie, IT Certified Professional (NZ)

Parklea Software Ltd, contact email: harvey@lockie.co.nz

     
Keywords: Sales Forecasting, Useage Forecasting, Demand Forecasting, Order Forecasting, Reorder Forecasting, Stock forecasting, Inventory forecasting, Sales Prediction, Useage Prediction, Demand Prediction, Order prediction, Reorder prediction, Stock Control, Stock Management, Inventory Control, Inventory Management, Predictive Engine, Forecasting Engine, Predictive Algorithm, Forecasting Algorithm, Genetic Algorithm,