The use of historical data, statistical models, and machine learning to predict future customer demand for products or services across time periods, markets, and channels. For agencies, demand forecasting is the analytical foundation for campaign timing decisions, budget allocation, and the coordination of media plans with client inventory and operations.
Also known as sales forecasting, demand prediction, predictive demand planning
Demand forecasting models learn patterns from historical sales or conversion data and project them forward. Classical approaches, including ARIMA, exponential smoothing, and seasonal decomposition, model time series structure explicitly and work well when demand patterns are stable and regular. Machine learning approaches, including gradient boosted trees and recurrent neural networks, learn patterns from many variables simultaneously and handle irregular patterns and large feature sets better than classical methods, at the cost of requiring more data and more expertise to configure correctly.
Modern demand forecasting incorporates external signals alongside internal sales history: search volume trends, macroeconomic indicators, weather data, competitor pricing, and event calendars. The addition of these signals improves forecast accuracy for products with strong external drivers and makes the model interpretable to stakeholders who recognize the causal mechanisms being captured.
Generative AI is beginning to change demand forecasting in two ways: by enabling natural language interfaces that allow non-technical stakeholders to query and explore forecasts, and by using language models fine-tuned on time series data to generate probabilistic forecasts that handle edge cases and regime changes better than purely statistical approaches.
Agency media plans and campaign timelines are built around assumptions about when demand will materialize. A working ad agency that builds those plans without access to the client’s demand forecast is guessing at timing. One that integrates demand forecasts into media planning is building plans that reflect the actual shape of expected consumer demand rather than calendar conventions and industry benchmarks.
Media plans misaligned with demand forecasts waste budget. Running high-pressure acquisition campaigns in low-demand periods depletes budget on harder-to-convert audiences and produces lower efficiency than the same budget deployed when demand is rising. Aligning campaign activation to the demand forecast, rather than the campaign launch calendar, is the operational change that produces better returns from the same budget.
Agencies can build demand forecasts on client data. Clients with historical sales data rarely have formal forecasting capabilities. Their predictions are manual estimates from the sales team and the finance team’s annual targets. An agency that can run even a basic machine learning demand forecast on the client’s own data is providing capability the client does not have internally and producing analysis that directly informs media strategy decisions.
Demand forecasting connects agency work to supply chain reality. Campaigns that drive demand spikes the client’s supply chain cannot fulfill produce customer disappointment and wasted ad spend. Agencies coordinating campaign timing with client demand and inventory forecasts are functioning as operational partners rather than just media buyers, which changes the nature and value of the relationship.
An agency manages media planning for a consumer packaged goods client across six product lines with different seasonal demand profiles. Rather than using the previous year’s media calendar as a template, the agency builds a demand forecasting model trained on three years of the client’s sales data, supplemented with search volume trends for each product category. The model produces weekly demand probability distributions for the next 52 weeks, segmented by product line and retail channel. The resulting media plan front-loads investment in the two weeks immediately before each product line’s forecast demand peak rather than at the start of the season. The client’s sales team reports that the shifted timing improves retail inventory coordination: stores are stocked and promotions are live when the model predicts consumer demand will arrive.
The generative AI foundations module of the workshop covers how to build analytical models on client data that agencies currently do not have internally, including demand and performance forecasting approaches that produce better decisions than historical averages.