Tuning a Statistical Forecast Part 1: Resources

Are the results of your Statistical Forecast not good and getting worse? Then your Statistical Engine needs tuning! Changes in Organisation, Data, Resources, Strategy and things like Global Pandemics will change your data and can impair the quality of statistical forecasts. How can you go about tuning the machine that creates your forecasts I hear you cry? I'm, glad you asked; there are a two different resource options:

Outsource & In-House

Outsource

Use a Third Party to undertake tuning on your behalf. Solution Providers, Integrators and specialist statistical tuning companies can be hired to perform analysis, recommend changes and implement them. If you don't know who they are check your Solution Provider website for 'Partners'.

Pros: There will be limited risk to internal resource capability using this approach; you won't need to remove any of your planning resources from the standard forecast cycle. Any Engine Tuning environments will only be needed for the period of tuning undertaken by the 3rd Party which will save on IT infrastructure or Cloud service costs.

This is the easiest and perhaps, most common approach. You can expect the tuning to be conducted and completed quickly since the specialists will have the necessary tools and experience; there should be very little trial and error. Partnerships created through such a 3rd Party could provide useful resource and advice opportunities in the future.

Cons: A clear direction or expectation will need to be created and provided to the tuning agent. You will need to indicate what it is that you want achieved, Performance, Short Horizon accuracy, Data Cleansing, Advice on Hierarchy Changes? You must be clear with your requirements and of course, there will be higher costs and the more you demand - the more expensive it will be.

It is unlikely that the 3rd Party will make any firm guarantees about improvements, so there is the risk that the gains you desire cannot be achieved in the time frame allotted. Assuming that this worse-case scenario does not happen, and you receive a beautifully serviced engine that is fit-for-purpose again, you will not have gained any knowledge or expertise! The analysis will be part time - not ongoing. There is also the risk of data security.

In-House

There are two variances to the Insource approach; either get your planners to perform the tuning or consider changing organisation structure somewhat and creating a specialist's engine tuning team to own the task on a permanent basis.

Demand Planners

Task your planners with performing the tuning analysis. After all, they know the data better than anyone! This could be a constant task inside the standard cycle or perhaps a once or twice a year activity or possibly a small project run to impart maturity growth.

Pros: This approach enables resource development and, if combined with a strategy for planning maturity growth it can drive and derive more than statistical improvement since it will improve the data understanding and quality of the Demand Planning team.

The planning teams are a 'sunk cost' and the tuning will be achieved for less outlay. The ownership of the expertise will be retained in the company and tuning can be applied as regularly as desired. Even if the tuning isn't as efficiently applied as through a specialist firm the education and insight gained by the planners will likely be significant.

Cons: Training to perform the tuning will be required - so at the very least you will need assistance to get started (or read this series of articles :-)). A Permanent Engine Tuning environment will be required or at least, a strategy to make one available when needed.

The size of the activity could be extremely large and time consuming. Unless you have spare internal resource, no matter how you manage it, the standard cycle activities will be impacted and trying to squeeze extra activity onto stressed planners could do more harm than good.

Dedicated Engine Tuners

Consider setting up a Centre of Planning Expertise. A specialist department could be resourced with Data Scientists to perform all manner of analytics for the company including tuning the Demand Planning Statistical Forecast.

Pros: This approach enables even greater resource development. The Demand Planners could have a clear career path from the Demand Planning Team to Data Scientists Team. The Tuners will likely have less product or customer bias which will create a more accurate baseline. Ownership of the expertise will be retained in the company and the tuning can be applied regularly and potentially within the standard cycle.

Cons: Training will be required and maintained internally. Permanent Engine Tuning environments will be required for the tuners to use. There will be the extra cost of new department with all the related resource requirements. New processes will be required; to remove statistical ownership from the Demand Planning team and to update the forecasts cycle inside S&OP.

Only Two Approaches?

Actually, there are always more options. Here are some alternatives:

1. Leave it Alone

Perhaps it will get better? If it doesn't, just instruct the planners to override more of the forecast.

2. Turn it Off

Perhaps Statistical generation is not the right way? Consider turning the engine off and doing something else instead. You could learn to use alternatives methods of generating a forecast; The Previous Forecast or Last Year with a factor or even going down the Demand Driven Pull route instead of Forecasting Push.

3. Statistical Forecast as a Service

Go the full outsource approach and pay a 3rd Party to run your Statistical Forecasts for you. There are a number of specialists firms (like Chainalytics or Houston- Analytics) that will take your data and use their engines and experts to create forecasts that you can then import and use.

4. Young Wizards

Get some eager advanced mathematics graduates from a nearby college or university. Consider partnering with a University to get the company on their curriculum for work experience. They will be keen to apply their knowledge and step from education into business. The first task could be to assess point alternative options 1 & 2.

Honestly, this is a win-win situation for all concerned. The Demand Planning team will gain their enthusiasm at graduate wages and they will get to use their mathematics skills to resolve real world problems. The graduates won't have any internal bias and could well become essential new lifeblood in your company.