Why Tim?

Philip Duplisey, VP – Strategy and Business Development

I recently shared my excitement for a reliable new automated forecasting solution. Everyone needs it, and producing a plan is a fundamental part of every line manager’s job function. It can also be argued that the effort of producing an accurate time series forecast should be one of the easiest costs to tie back to tangible ROI for the business.

However, it’s not something we see very often because it’s not easy to do.

Right now, you have a few options.

Intuition-driven approach

Most managers start with the business target that comes down from senior leadership, which is often derived manually, and by combining some business experience, a look at history and perhaps recent trends, a time series forecast is built in Excel. This is largely a manual approach.

Univariate approach

The next level is using a statistical method like Holt-Winters, ARIMA or some other smoothing approach. This is definitely a huge improvement, and many desktop tools offer these solutions “out of the box.” It’s also easy for IT to implement with many open source libraries for R or Python available. The problem with these methods, however, is that no other factor except the series is considered while forecasting. They are univariate. We all instinctively know that anything we are trying to predict depends on how something has trended historically, as well as many other factors.

Machine learning Approach

Machine learning methods allow you to include multiple factors in the final forecast, and overcome many of the shortcomings of univariate forecasting methods, but are complex to implement, require skilled data scientists, and are compute-intensive, i.e. expensive and generally take minutes, hours, days, weeks and months to produce, depending how you wish to measure the effort.

Introducing TIM

This is why I am so excited about this new solution. TIM takes a completely different approach, and instead of trying to find the right algorithm, it determines what factors (features) in the data are important. This is a game-changer. TIM can produce a more accurate time series forecast in seconds and is very easy to use and integrate with many other tools: you can use it right inside of Excel, or your favorite analytics platform.

An interesting side effect is that to build a forecast, TIM finds the most predictive features in your data. Sometimes that could be more important than the actual forecast itself, because if the features change, it tells you something in your business has changed. That’s the anomaly detection feature provided by TIM. Imagine what you could do with that?