Demand Sensing vs Demand Forecasting and How AI fits in the picture

?

Demand Sensing ≠ Forecasting

AND

Black Magic ? {Forecasting,Demand Sensing}

?

Recently I was discussing with one of my prospects about how customers are being misled in the name of Demand Sensing, Machine Learning and AI. So I thought of penning down my thoughts.

I can explain everything in Bullet points explaining the Supervised and unsupervised learning algorithms planners can use. But where is the fun in that? You can read that anywhere on the internet. Besides, humans learn better with stories instead of cold hard facts. So lets use the most widely used analogy of Party Planning to describe forecasting and Demand Sensing.

?Imagine you have a party in your house next Friday, and you must get drinks and food. How do you estimate the quantity so that you don’t run short or end up with excess on the next day?

You look at the expected number of guests expected, their age, dietary preferences, etc and procure the products from the store.

This is forecasting in its simplest form.

If you could buy all the items from the store next to your house, why would you need a forecast in the first place? You know about the store 30km away where you shop for grocery because it offers value for money. Also, instead of going to the store specially for the party purchase, you decide to combine it with your weekend grocery shopping because the party is still 7 days away and you can shop on Saturday or Sunday.

Simple, isn’t it?

This is taking advantage of a better forecast over lead time which is lowering your purchasing cost and write-off risks.

Now imagine that instead of inviting 6 to 8 people on just one weekend, you run a business of organizing party for your town and the party happens every Friday.

This means you are repeating the forecasting and procurement process every week.

?Instead of manually going through preferences of each person every week, you hire a robot, lets call him Marvin,(No, not the same paranoid android from Hitchhikers guide, but rather a happy one), to help you. In other words - to help you automate that decision and adjust the final number based on your budget. It also tells you the best store to procure it from and optimises to host maximum number of guests.

This is Forecasting Automation helping in Demand Planning and S&OP.

A simple statistical engine will not be able to do this, but an AI engine can do it. As a result, you are focusing on day-to-day activity instead of bothering about the accuracy of the number.

?

As the day of party gets closer, people RSVP the invites.

Before your shopping run, which is 5 days before the party, only 50% of the people confirmed. This information is assimilated by Marvin to adjust the final list of items that need to be purchased. This is Order Consumption allowing you to adjust your operations depending on your lead times. Marvin also detects that every week, 2-3% of people show up without RSVP, hence he recommends contingency forecast. Marvin also tells you that people who show up in party for consecutive weeks, do not show up for the next 1 month. Furthermore, Marvin observes that people eat more hors d'oeuvres if they arrive early. In other words Marvin is now sensing the Demand, not simply forecasting anymore. This, beginning from the Order Consumption, is Demand Sensing with Pattern Recognition. In other words, Demand Sensing is "Order Consumption on Steroids".

On the day of the party, people start showing up. On that day, people who arrive early, start eating more hors d'oeuvres as Marvin predicted, but you still run out of them. This, is the ability to quantify the impact of influence factors on demand /consumption.

Hence you must quickly head out to the next-door shop and buy more. This is Expedite cost which is unavoidable. In other words, Forecasting is not Absolute.

?

Let’s take this a step further and categorize the products into groups. Certain products have shorter shelf life hence you can only buy it for the next party but others ′have a longer shelf life. So, you decide to buy 1 year worth of stock. Marvin immediately interrupts your chain of thought and reminds you that you are on a budget and have only so much space in the house. Looking at all the factors, you decide to buy only 4 weeks worth of stock and then tell Marvin that you are going to buy 4 week forecast. Marvin stops you again and reminds you about variations. You boast about your statistics knowledge and decide to use Z × σLT × D calculation to cover 98% of the demand.

The robot which only felt one emotion, started to feel amused and tells you he has a more accurate way of calculating risk.

Until now, Marvin was only telling you about the next week’s forecast, but you had told him to forecast for longer periods in order to ask for budget. Marvin had stored all those previous versions and he compared each of the 4-week out forecast with actual consumption in the party at the end of the 4 week cycle. This gave him an estimate about what should be the risk-hedging stock if you start buying products using 4 week forecast. This is Forecast error over Lead time being used to calculate Safety Stock and Optimise inventory.

?

Unlike his namesake in the Hitchhikers guide to Galaxy, the Marvin in our story is pretty efficient and doesn’t suffer from depression, however there is one similarity. The full potential of Marvin has never been realized because people always see Marvin as a job killer instead of a tool to improve productivity. AI, in the context of Supply Chain Planning is not Black magic or HAL 9000 (Yet), but instead more like the Electric motor in your Bicycle . It is meant to augment planners' work.

More importantly, the purpose of this story is to help buyers understand the difference between Demand Planning and Demand Sensing to make better purchase decisions.

?I hope you find this useful.