September 21, 2020
Forecasting: predict anything you can think of (and have data for)
Revenue. Sales numbers
Apart from the weather and election results, these are the metrics most people imagine when they hear the word Forecasting.
Forecasting, as defined by Wikipedia, “is the process of making predictions of the future based on past and present data and most commonly by analysis of trends.
As per the Oxford Dictionary, the words similar to Forecasting are – Predict, Prophesy, Foretell, Foresee, Guess, Speculate, Estimate. Now when you hear these grammatically different, but similar meanings you suddenly have a different picture of the kind of things that can be predicted, or estimated or speculated.
Suddenly we start imagining stock indices, currency exchange rates, traffic patterns, disease and healthcare trends, etc.
Forecasting allows you to do exactly the same – Everything you can imagine and have the data for – can be forecasted. Many things are easy to forecast, but a few things are more difficult than some others – since they do not follow logical trends – and can be very irrational in their behaviour (ever tried predicting election results?) But for most, if you have historical data, and clear influencers of that metric – you can forecast the same with acceptable levels of accuracy.
Revenue, sales, and demand become the easiest and the most obvious choices since they frequently follow historical patterns, and in many cases have strongly defined trends and seasonalities.
Sophisticated algorithms have ensured that even complex metrics can be forecasted to a high degree of accuracy. The following are a few examples of how different industries use Machine Learning frameworks to forecast not just revenue/sales, but also predict various metrics which can create a large impact on businesses.
- Forecasting Demand:
Being able to accurately estimate how many units of your product will sell in the next day, week, month, quarter or year is a very powerful tool for any business. Demand Forecast is probably the single most discussed metric in boardrooms and business plans for almost every company. Today, Machine Learning allows us to build sophisticated demand forecasts which factor not just historical data – but also combine many external determinants of demand (weather, holidays, pricing, discount, etc) to get a fairly close idea about how the future demand is going to look like. The current approach in most legacy forecasting models works on the assumption that historical data is an indicator of the future trend. I.e the future is determined by the past.
This model works as a good approximation framework – but would fail in case there are complex factors influencing demand. Hence these models have very moderate accuracy levels. Machine Learning allows us to easily include factors that are too complex to comprehend either in manual or in simple forecasting processes – Things like seasonal patterns, pricing influences, availability metrics, etc can be included in demand frameworks. This gives a very good picture of how the future sales/revenue/demand is going to be like.
- Predicting the number of visitors per store:
The number of visitors in a store (if you’re an online vendor – the number of visitors on your site) is probably the starting point of your planning process. Getting a good forecast of your daily footfall/traffic by the store is clearly a very important but often hard-to-estimate metric. Over a period of time companies try at best to get a generic estimation of the high, medium, or low footfall scenarios and plan accordingly. Traffic in a store depends on many factors including what day of the week it is & the time depending on what you are selling.
Weather, pricing, special offers, availability/out of stock situations too plays an important role. Machine Learning algorithms can allow you to co-relate all these (and more) fairly accurately to determine the level of traffic/footfalls in the upcoming days/weeks. This allows you to plan your logistics, operations, and inventory better. This leads to not just better customer satisfaction – but also increases your operational efficiency, reduces wastage, and overall better profits.
- Forecasting number of orders per day by city/state/region:
Driving traffic to your store is one thing, but getting your customers to buy more and frequently is totally a different challenge. The footfalls to the store can be an input to forecast the number of orders which the store (offline or online) will get. But this metric also depends on the demand for topical/seasonal goods, their availability (or being out of stock), the pricing and discounts offered, payment options available amongst many other things. Getting this metric right allows you to do more precise revenue forecasts, align your logistics (delivery teams, carts etc) – hence impacting efficiency, satisfaction, and revenue.
- Forecasting number of orders by new users and existing users per day by city/location:
Do you think your existing customers will come back for more? Do you think you are attracting enough new customers? Or are they turning towards the competition? Is your marketing strategy aligned for you to acquire new customers or persuade existing customers to buy more, or buy more often? Is the existing strategy even doing you any good?
Most enterprises would love to be able to answer all these questions precisely- but most often struggle as new/existing user counts are usually analyzed through the rearview mirror. Today, Machine Learning can allow you to easily co-relate the drivers of new/existing customer engagement with your historical data and predict with a very good degree of accuracy on how that metric would work for you in the future.
You can very easily generate future scenarios by controlling different levers of marketing promotions to see the best combination which would yield your required results. You can finally have a strategy and a decision-making criterion which is backed by data – and not just rely on gut feel and historical patterns.
- Forecasting supplier variability
Much of the forecasts are done for finished goods, leading to accurate predictions of the requirements on things like raw materials, packaging materials, etc. But a major component of the supply chain is the supply side variability. You may have a world-class supply chain framework but if you have not considered the variability on the supply side you will probably still get stuck with shipments not reaching on time. This will lead to loss of sales and revenue and you will even incur other costs, especially if you are heavily dependant on importing finished goods and raw materials.
If you have uncertain timelines from suppliers you would know how important it is to be able to estimate the variability and plan for it well in advance. The safest solution is to order significantly more than you need and stock up but that comes with its own downsides – blocking cash flow, storage limitations, expiry dates, damages, and other challenges.
Machine Learning algorithms can work on the historical data from your supply chain and co-relate it with external factors that will help predict the variability of the supply-side framework allowing you to plan the ordering schedule very precisely: HOW MUCH of WHAT needs to be ordered and WHEN for every item in your inventory.
Forecasting the supply-side variables and combining them with the future demand prediction can give you a very good idea of how your supply chain should run.
- Forecasting auction pricing for commodities
There are many businesses that have variable pricing based on agricultural produce, climatic variations, and global demand. If you are in the business of commodities, metals, tea, crops, etc you would know what this is about. There are global exchanges that auction (on spot and futures) and a company’s ability to be able to secure the right quantities at the right prices determine their sales, profitability, and business viability. While these auctions are tough to determine as there are very many factors that could determine the final selling price Machine Learning algorithms allows you to get a good idea of the price range and allowing companies to plan their business better, and spread out their risk if needed. Many companies now rely on such tools to get a general forecast of what the prices could look like before planning their purchases and orders.
These are just a few examples of how companies are using new-age forecasting frameworks to get ahead in the game and build a competitive advantage.
If you have or can your hands on the data for it you can virtually forecast anything in the business world today. Some of these metrics are easy to predict, some of them need more complicated processes to get future outcomes – but at the end of the day almost everything can be predicted!
Forecasting is the bedrock of every business. It brings agility in business planning which is amplified by accuracy and scale.