Insights

Revolutionize retail with AI forecasting

Peter Charness, VP Retail Strategy - UST

Forecasting for retail is high on the list of applications that can be significantly improved with the use of AI. However, forecasting is neither simple nor monolithic. It becomes truly compelling when we consider the business outcomes or goals of the forecast and the additional steps required to achieve those goals.

The end-to-end forecast process comprises numerous stages, some common across all forecast goals but many specific to particular objectives. Creating effective routines backed by the correct AI methods for each forecast goal requires specialized logic and appropriate feedback mechanisms. To discuss AI in forecasting, we must get granular and explore the types of AI that apply to different forecasting stages, depending on the goal.

 DIVIDER

Specialized forecast goals and logic needs

Placing orders: Determining merchandise needs at various points in the supply and fulfillment chain to create orders for stores or vendors is complex. Once a sales forecast is available, a sophisticated order-building process optimizes for factors such as pack sizes, constrained capacities along the supply chain, and econometric modeling, considering factors like inventory carrying costs. Further complexity arises with planning for promotional buys and floor resets. A forecast that can't be turned into high-quality orders is only mildly interesting.

Managing price and promotions: Using a forecast to determine optimal prices and promotions involves various price-setting scenarios across different locations, customer groups, and supplier rebates, aiming to maximize sales and margin. Getting through price elasticity to a price recommendation requires more modeling than just a forecast.

Financial modeling: Forecasting sales probabilities for financial modeling, typically for pro forma profitability calculations, involves setting accurate predictive Pro Forma Income Statement factors for presentation to the board or the street. Here, the forecast is just the starting point for evaluating and modeling the rest of the P&L.

Planning staff investments and labor scheduling: Determining optimal staffing investments and testing labor scheduling possibilities aim to achieve high task-oriented service levels while optimizing human resource investments. Forecasting, in this case, involves factors beyond sales, including availability, skill sets, union or legal regulations, and more. The forecast barely scratches the surface of the model needed to create a useful staffing plan.

 DIVIDER

The multistep nature of forecasting

Forecasting is a multistep activity, and some steps are common across most goal models. A useful way to look at these basics can be illustrated as follows:

 DIVIDER

Drilling into forecasting steps

Data normalization and causal factors

This step aims to correct historical data, removing significant factors that won't reoccur in the next forecast interval and adding factors reflective of future intentions. Examples include:

• Store closures, renovations, and new openings:
  Forecasts must account for go-forward sales coverage intentions. Closed stores' history needs to be removed or redistributed to adjacent stores. Renovated stores require modeling for the sales impact during renovation and the usual uplift upon reopening. New stores need fabricated bases for forecasting with initial sales assumptions.
• Promotions and weather impact:
  These causal factors need consideration within data normalization.
• Avoid forecasting prior errors:
  This includes obvious mistakes, such as not repeating stockout errors, and subtle ones, like misattributed sales due to backup location fulfillment.

Improved data cleansing and normalization enhance all types of forecasting. Will your AI project address the "find and fix" requirements for these issues, or will you tackle data normalization differently?

 DIVIDER

Strategies and parameters

Influencing the forecast by informing algorithms and downstream systems of corporate goals is crucial. For example, a strategy to increase market share for children's wear might involve signaling this intent properly to bias higher forecasts, increase stock and staff levels, and adjust minimums/presentation stocks or additional facings. The right AI support can help turn strategy into action.

 DIVIDER

Short term, long term, and granularity

Depending on the business goal, a forecast needs to operate at a useful level of granularity that produces workable accuracy. Longer-range forecasts have higher error probabilities. Store replenishment forecasts may only need a few weeks' outlook, while orders from overseas vendor might require a six-twelve month horizon. The business goal drives the level of granularity and effort needed.

 DIVIDER

Completing the process: Turning forecasts into action

Forecasting itself is only mildly interesting. One must consider the purpose and goal of a forecast— producing an order, setting a schedule, etc.—and how AI can support completing the entire scenario and achieving the business goal. An AI project for "forecasting" is a good starting point but should not be considered in isolation from solving the entire business need.

 DIVIDER

The future of AI in retail forecasting

The future of AI in retail forecasting promises to revolutionize how businesses predict and plan for their needs. As AI technology advances, retailers who adopt and integrate these capabilities will be better positioned to navigate the complexities of modern retail and achieve greater efficiency and accuracy in their operations.

By understanding and leveraging the specialized logic needed for different forecasting goals and ensuring a comprehensive approach that includes data normalization, strategy alignment, and appropriate granularity, retailers can turn forecasts into actionable insights that drive business success.

Learn more about retail forecasting at https://www.ust.com/en/industries/retail-and-cpg

DIVIDER

Resources

https://www.ust.com/en/industries/retail-and-cpg

https://www.ust.com/en/alpha-ai

https://www.ust.com/en/our-partners