Trade promotion forecasting is difficult: getting it right involves factoring in many variables. Marketing people, who are often responsible for it, may excel at creating and running promotions. Forecasting demand data, not so much. This can leave supply chain professionals the tough job planning for, and responding to, promotions.
To do this, the first hurdle is creating a good statistical baseline forecast. This allows you to identify the uplift the promotion will presumably deliver on top of the normal business performance. Many organizations fall down at this first hurdle because their baseline forecasts lack precision and rely too much on gut feel, guesswork and oversimplified calculations.
The next step is to accurately predict the uplift from baseline expected to be generated by the promotion so that supply chains can optimize inventory to properly support the promotion. This includes pre-filling inventory, forward locating inventory to respond to unexpected spikes in demand, and gracefully exiting from the promotion in a way that doesn’t leave a lot of excess inventory, especially if there is a period of slow demand immediately after the promotion. The Nielsen Company calls this a “mortgaging” effect, where bringing forward sales from a later period reduces demand in later periods.
Most companies lack the skills and tools to produce the forecasts needed to make these kind of judgments. Some still manage their promotional data in spreadsheets simply because they are familiar with them. But two-dimensional spreadsheets, however transparent, are incapable of handling the many different variables and data sources involved.
As stated before, the first step up is good statistical forecasting software and process that allows you to create a reliable baseline forecast. But Industry leaders like Danone (click here for case study) are going to the next step, using machine learning enhanced advanced analytics for trade promotions planning. Machine learning offers the added horsepower to improve demand forecasting accuracy by going beyond historical sales data to blend in data about the uplift and ROI of past promotional campaigns. An added benefit is that machine learning systems automatically improve themselves over time as they gain new insights from a growing history of consumer responses. Your returns from investing in the technology continue to grow.
How else can machine learning improve trade promotion forecasting outcomes?
Imagine a system that could proactively suggest a new promotion that might increase your revenue by drawing on past promotion data. You provide the system your target revenue or profit and it recommends a promotion plan to reach it.
Here’s another scenario. You are running a promotion and stock is getting dangerously low. At a certain point, the promotion will start cannibalizing your profit margin. A machine learning-enhanced system could continually monitor and analyze the balance between promotions and inventories, sending you an alert when it is time to stop or alter the promotion or change before it becomes less profitable.
Now imagine being able to factor in external data like weather forecasts and social media sentiment to make trade promotion forecasting even more accurate. This ‘social sensing’ enhanced promotion forecasting is actually in use today.
Despite the benefits, the prospect of reaching advanced technical maturity can seem overwhelming – especially for those still operating in spreadsheet mode. Fortunately, supply chain teams can make their cases for step-wise investments in skills and tools. In Gartner’s report Enhance Demand-Planning Maturity through a Scalable and Sustainable Technology Strategy, the analyst firm recommends building a technology foundation to support advanced demand planning. It describes a maturity model that delivers increasing ROI over each of five stages. As you move into the higher stages of maturity, you increasingly rely on advanced analytics and machine learning for demand shaping and promotion forecasting.
We have seen firsthand how companies’ bottom-lines improve as they invest in more accurately forecasting trade promotions and more effectively managing their impact. As Gartner explains, you should advance your demand planning maturity and forecast promotion planning in stages at your own pace. But don’t wait too long. We hear from customers who have made the journey that their main regret was not getting started sooner.
Click below for more information on forecasting trade promotions:
Gartner classifies Demand Planning maturity into five stages:
Stage 1 typically includes alerts, dashboards, import and export of spreadsheets, report generation and workflow management.
Stage 2 includes statistical forecast generation and baseline data cleansing. Hierarchies can be aggregated and disaggregated. Forecasting can be top-down, bottom-up or middle out. Various forms of exception management become available. Other capabilities often found at this maturity level are reference data maintenance, time bucket utilization, unit conversion and demand-class identification.
Stage 3 requires a much more robust planning platform. Trade promotion forecasting usually first appears at this maturity level. Other capabilities include what-if analysis, item phase-n and phase-out, multilevel and multidimensional hierarchical planning. Other capabilities may include causal forecasting, role-based profiles, scenario comparisons, planning of intermittent demand, assumption management, demand-characteristic identification, and collaboration.
Stage 4 adds multi-enterprise coordination, demand sensing, demand shaping, predictive and prescriptive analytics and granular demand-plan analytics. Other possible capabilities include segmentation definition, proportion profile planning, statistical model blending, cost-to serve analytics, and point-of-sale analytics. Machine learning has usually been introduced by this stage of maturity.
Stage 5 maturity adds demand pattern analytics to the above capabilities.