Custom machine learning
See a demand forecast, its benchmark and the uncertainty behind the model.
0:52 · Product walkthroughDemonstration data and modelled examples. Results shown are not achieved customer results.
Video transcript
The demand forecast uses a custom gradient-boosted machine learning model, tested against a seasonal benchmark. Thirteen historical weeks were held back from training to assess performance on unseen data. At company level, weighted weekly forecast error was three point two nine per cent, compared with three point six three per cent for the benchmark. Detailed product and site forecasts have greater uncertainty, so the level of the comparison matters. The future view extends the planning horizon by twelve weeks. This gives the team a tested basis for planning, with uncertainty visible alongside the forecast. We define the decision, test the model and agree how its output will be used in the business.
Explore this capabilitySee another workflow.
8:34FP&A: from performance to a plan
Follow a performance gap into budgeting, rolling forecasts, workforce costs, cash planning and approval workflows.
8:58Aurelia: a business application in action
Explore reporting, financial evidence, custom models and email briefings in one connected business application.
0:41From the performance gap to the next action
Follow an EBITDA gap into its supporting accounts, then connect the response to initiatives and accountable owners.