Demand, Sales and Workload Forecasting Software

Small guide to forecasts retrieval

Support » Here

Small guide to forecasts retrieval

With Lokad, as soon as you specify a forecasting task, you are able to retrieve the forecast results. However, we would like to emphasize that the forecast quality increases over time. Since this point is slightly complex, we provide some explanations below.

Upload your data as soon as you can. 
Retrieve your forecasts as late as you can.


A time-series forecast improves over time for 3 main reasons:
  • When the time-series is updated, the forecasts are updated too. The Lokad forecast algorithms leverage the most recent data to make the forecast as accurate as possible, thus any change in the time-series data has an immediate impact on the time-series forecast.
  • Lokad is using, not one, but several forecast algorithms. These algorithms represent a distinct tradeoff between forecast accuracy and computational complexity. Intuitively, very fast algorithms (providing near real-time forecasts) are less accurate than highly complex algorithms (providing the best accuracy that Lokad can achieve). Thus, Lokad will first apply the very fast algorithms (so that the forecast results are instantly available) but then, as soon as results from slower (but more accurate) algorithms become available, they replace the previous results.
  • Lokad is using not only your data but all of the community data to build your forecasts. Even if your time-series data stays unchanged, the community data, which includes all the other Lokad accounts, evolves continuously. Thus, the forecasts get regularly updated to leverage the latest time-series updates within the community.

As a consequence, the sooner you provide the time-series data, the better the forecasts when you retrieve them, because more time enables more powerful algorithms. Also, the later you retrieve the forecasts, the more leverage you get from the community data because of the continuous updates occurring on the Lokad network.

As a rule of thumb, you can expect the following delays
  • 24h to get the optimal forecast algorithm running on your data.
  • One week to get an optimal integration of a new time-series within the Lokad network.

Those delays are only indicative (actual delays may be shorter in practice). Additionally, Lokad follows a best efforts approach: the forecasts can be retrieved at any time, and we do our best to provide the best possible forecasts within the available period.