Technology is advancing at an increasingly rapid pace, many of the things that we thought were a dream a few years ago are reality today. Artificial Intelligence (AI) and Machine Learning (ML) for use in demand planning have been talked about for a long time and have promised much but delivered little. Until now.
It is vitally important that there is a good understanding of what these technologies can bring to an organization and not simply think “great, now I don’t need headcount for demand planners”. If that is your position, then you will be disappointed. Rather, these technologies, that are available now, will release your demand planners and other staff from the tedium of data crunching. Their focus will be on interpretation and application of the data which will bring great value to your demand planning process and S&OP demand review.
A wise man once said, “AI is the answer, now what is the question?” In this article, we look at some of the urgent questions that confront demand planners and show how AI and ML can provide the answers.
1. How do I work out what my “baseline” actually is?
This is a perennial problem for demand planners. Classic demand planning is about defining the baseline and then adding or subtracting the building blocks to reach the final demand plan. Typically, these building blocks are promotions (see below), price changes or new product launches.
The difficulty that demand planners have is calculating exactly what the baseline should be, what should be included and what should not. They can spend many hours sifting through historical data, manually removing the impact of marketing activities, stock outs, one-off events, etc. And at the end, there is still debate about whether this baseline is correct.
Statistical forecasting with outlier correction has been around for a long time and has been used to good effect in many companies, but stat forecasts only work on past data, without analysis and outlier correction will correct significant variations but will miss other, cumulative, demand fluctuations. This is still a good start, but AI/ML can do better.
By analysing your data, understanding when promotions, price changes, stock outs, happened, it is possible for the technology to derive a much more refined baseline than was possible before. And by applying machine learning to the data, the technology can continually review the accuracy of this baseline, thereby increasing confidence in the numbers the demand planner is using but also helping you to understand what really impacts your demand plan and what is simply noise.
One extremely important component of this activity is the analysis of promotions and that leads us to our next two questions…
2. How much uplift should I include for this promotion, and when should it run?
Having a well-defined baseline is only the start of the demand planning process. We now need to add in some building blocks. For this, we will concentrate on promotions, but the principles are similar for other types of demand uplift.
AI is extremely good at analysing data quickly and finding trends and impacts of changes. Armed with the calendar of promotions, the AI engine can spot how successful promotions have been and, based on supporting data and events it can provide accurate data on the likely volume uplift from a future promotion. Finding the optimum time to run your promotion is also possible today.
An oversimplistic example says that you don’t run promotions for ice cream in the middle of winter. AI/ML is much more sophisticated than that and can highlight the optimum time for specific promotions.
This is obviously of great benefit to marketing departments who can struggle to understand the impact of their promotions. Getting a clear picture of why a promotion worked enables marketing to plan more effective campaigns and avoids wasting valuable marketing dollars on campaigns that are likely to fail.
The technology is constantly learning from your data and provides clear analysis of the effectiveness of a promotion, or campaign. As we have said, this transforms the information available to marketing to enable them to understand the likely uplift due to promotions and therefore the ROI for each activity. However, the other advantage of constantly learning about what works and what doesn’t work is that ML can make recommendations about what should be promoted or what campaigns to run. This is available now and is taking the guess work, or gut feel out of marketing plans. Based on the data, it leads us to…
3. What kind of promotions should we run?
Yes, that’s right AI/ML can deliver recommendations on the type of promotion to run, the expected uplift and the timing. This saves hours of work within the marketing department and provides good, data-driven, plans that can be worked into the overall demand plan. Marketers and demand planners are happy!
Just as exciting is the ability to know what campaigns NOT to run. AI/ML is very good at spotting where activities have not delivered results or are delivering below their expected ROI. Untold millions of dollars are spent on sub-optimal marketing campaigns and without AI/ML, it is extremely difficult to identify this and to take action to stop it happening again.
So, with AI/ML, marketing has valuable input that tells them what, where, when and how much for each promotion or campaign. This enables marketing to focus on the real value-added activity of planning and delivering the marketing materials, artwork, literature, etc.
With the marketing calendar linked to the AI/ML technology, timings and volumes can be delivered direct to the demand planner to build into the overall demand plan. This, again, saves a lot of time where manual input to or from a spreadsheet is too often the norm.
4. If a business driver changes, what impact will it have on demand?
We have our baseline and we know what the building blocks of promotions and campaigns look like. But these are not the only things that determine a demand plan. Competitor activity, market share, number of sales people, Forex rates, GDP, Oil Price, weather, are all examples of drivers of demand and are often poorly understood within an organization.
AI/ML has much to offer in this area in providing insight into what is really affecting demand. Organisations can often articulate what the main demand drivers are (although they are sometimes wrong) but they find it much harder to define the impact these drivers have on demand. For example, if the temperature increases by 5 degrees C, what will that do to sales of beer? We all know that demand will increase, but by how much can often be a wild guess (or assumption if you are more refined).
AI/ML is able to analyse your data and identify key correlations with external data. The good thing about ML, is that this can be done many times, very quickly, with different parameters in order to zero in on what really drives demand.
So far, we have mainly looked back at past data to derive some insight into the future, but here we are starting to look to the future to understand how demand might change. AI/ML can be set up to constantly monitor the key parameters affecting demand and highlight changes as conditions move. For example, if AI is connected to long range weather forecasting systems, it can identify demand impacts due to climate changes. This is done today in some organisations, but AI can enhance this activity simply because the collection and analysis of the data is so much faster.
A further advantage of this is that the technology can constantly monitor actual demand versus the plan and quickly spot where parameters are changing that will cause significant changes in demand. Very few organisations have the ability to do this and yet the technology is available now. This transforms tactical planning and demand planning by providing insight back into the long-term planning horizon (S&OP).
This idea of providing insights is our final question that demand planners ask…
5. I know what has changed but why has it changed?
Yes, it is easy to know what has changed, just compare the last plan with the current one. The more important question is WHY has the plan changed?
Demand planners spend many hours searching for the answer to the “why?” question. They often fail or run out of time. As we have already seen, AI/ML can determine what the baseline demand is, what promotions to run and when and also understands the key drivers for demand. A demand plan that is recommended by AI can therefore articulate why the plan has changed.
In the past (and the present, with basic systems), changes were articulated as a set of computer parameters, e.g. alpha changed to 0.02, beta 1.4, etc. Again, this is what not why.
With AI/ML the output can be delivered in plain language, in the format of your choice and can be used as direct input to your S&OP demand review. For example, “Demand is increased P4 to P8, due to extra marketing promotion, starting end P3. Over the planning horizon, we see a 4% increase in demand due to anticipated reduction in oil prices. Demand is reduced in P6 to reflect the later Lunar New Year.”
This kind of input to the demand planning process, increases confidence in the demand plan and releases the demand planner from data searching and admin.
In conclusion
Will AI/ML replace the demand planner? In short “no”. However, used correctly, AI/ML is probably the most powerful tool yet deployed in the demand planning process. It can provide insights into what really matters in the demand plan, recommend changes to the plan and explain why.
What we once thought was a dream is reality.