AI Demand Forecasting for S&OP Starts with a Different Approach to Data

How Impactive AI examines data structures across industries and identifies indicators that can support forecasting

“Can AI understand the specifics of our industry? It’s very complex.”

This is a question we often hear from companies considering AI demand forecasting. Their concern is understandable. Even within manufacturing, demand patterns can vary significantly depending on where a company sources raw materials, whom it sells to, and when its customers decide to place orders.

Historical sales data captures these processes and their outcomes. To forecast future demand, we need to examine not only how the numbers change, but also the business structure behind them.

At Impactive AI, we use our in-house analytics system to map each client’s supply chain and commercial relationships, then explore indicators and time lags that may influence demand by product and channel. This helps us determine which demand streams to forecast separately, what data to obtain, and which relationships to validate.

This article introduces how Impactive AI’s data science team develops an understanding of each client’s business and identifies the data needed for forecasting.

‍

1. Identifying Distinct Demand Patterns Hidden in Total Sales Volume

If order volume increased this month, will that trend continue next month?

To answer this, we first need to understand why orders increased. The rise may reflect actual demand growth, but it could also result from customers building inventory in advance or bringing forward planned orders.

The same applies to revenue. When revenue rises, we need to distinguish between an increase in sales volume and a change in price. For products sold through distribution channels, the volume shipped by our client may also differ from the volume purchased by end consumers.

Order volume typically combines several types of demand. To distinguish them, we examine why customers buy and what determines when they place orders.

In a recent client analysis, we found that some demand streams were declining while others were growing. Total order volume appeared relatively stable, but breaking demand down by type revealed distinct trends.

As this case illustrates, aggregate sales volume alone can make it difficult to understand demand when total orders show little change. Based on our analysis of the client’s demand structure, we established criteria for forecasting demand streams separately and examining the indicators and time lags relevant to each.

Improving forecast accuracy requires more than adding variables. It is essential to distinguish which orders reflect underlying demand. We need to determine whether an increase in orders reflects higher consumption, inventory building, or earlier ordering, and examine the appropriate data accordingly.

‍

2. Connecting Supply Chain Analysis to a Data Acquisition Plan

Before building a model, Impactive AI examines the client’s industry and business structure. We conduct this process through our in-house system, starting with the following questions:

• Where does demand originate, and who makes purchasing and ordering decisions?
• Through which commercial relationships and sales channels are products and services supplied?
• Through what pathways do market changes affect the client’s business performance?
• What conditions and time lags exist between changes in demand and actual orders?

We organize the answers into a supply chain map, candidate indicators, and a plan for obtaining the necessary data.

1. Map the supply chain: We trace the flow from raw material sourcing to end markets. We also examine the criteria that suppliers, direct customers, and distribution channels use to make ordering decisions.
2. Map influence pathways and candidate indicators: We identify how changes in customers’ production, sales, and inventory may translate into orders for our client, then look for indicators that can be examined at each stage.
3. Develop a data acquisition plan: We classify the identified indicators by source—public statistics or the client’s internal data—and specify where and in what format the data can be obtained.

Even within a single company, demand can originate through different pathways depending on the sales channel and the industries it serves. Raw material prices and exchange rates may affect purchasing timing as well as costs and selling prices, so we examine their roles in the context of the client’s commercial terms.

‍

3. Selecting Relevant Indicators for Validation

A wide range of data may be relevant to a client’s market. However, not every industry indicator is equally useful for forecasting that company’s demand. An indicator that captures the direction of the overall market may not explain changes in orders for a particular product. The same indicator may also affect different sales channels at different times.

For the client introduced earlier, we defined separate forecasting targets by product and sales channel. We then mapped dozens of candidate indicators that could influence those targets, along with hundreds of causal relationships to examine.

What matters is the rationale for examining each indicator, rather than the number of indicators collected. Impactive AI uses the following criteria to clarify the role of each candidate indicator and what needs to be validated.

When organizing indicators, we also analyze the pathways that connect them to the client’s order volume. For example, a change in market conditions may affect a customer’s production or sales, leading to adjustments in purchasing plans and order quantities.

Our internal system displays these indicators and relationships in a single graph. The example below is adapted from an actual system screen. The graph helps us examine the connections between forecasting targets and candidate indicators, including both direct relationships and effects that may emerge through several intermediate steps.

We also examine when these effects may appear. Even for the same indicator, time lags can vary by product and channel depending on customers’ inventory levels, purchasing cycles, delivery lead times, and contract terms. Relationships with insufficient supporting evidence are flagged for subsequent validation with data.

‍

4. Checking Whether External Changes Also Appear in the Client’s Results

External data helps us understand broader market trends, while the client’s internal data shows how those trends have appeared in results by product and customer. Connecting external indicators with internal business results allows us to examine which products and channels are affected by market changes, and when.

For example, if an external demand indicator is rising while orders through a particular channel are declining, we need to examine that channel’s inventory and customer mix. If orders suddenly increase, we check whether channel sales have also risen, whether orders have simply been brought forward, or whether sales terms have changed.

_{These are examples of data that may be examined. The actual scope depends on the client’s business characteristics and data availability.}

Organizing data requirements around specific questions also gives each data request a clear purpose. Early in the project, we can determine which data to review first, what additional information to request, and which relationships to validate.

‍

5. Defining Forecasting Granularity and Model Variables Based on the Analysis

Business structure analysis informs how we configure forecasting models. For the client discussed earlier, the findings led to the following design criteria.

Even companies in the same industry may need different data and forecasting granularity depending on their customer mix, sales channels, and contract terms. Impactive AI analyzes these differences to establish model design criteria for each client.

The resulting forecasts inform production, purchasing, inventory, and supply planning decisions. Users therefore need to understand not only the forecast values, but also the factors and conditions considered. The same increase in orders may call for different purchasing and inventory decisions depending on whether it reflects temporary early ordering or sustained demand growth.

Forecasting targets and units of measurement must also be clear when comparing AI forecasts with the sales team’s plan. Specifying the products and channels covered, and how volume and price are treated separately, allows teams to assess differences on a consistent basis.

Impactive AI establishes these comparison criteria and the basis for interpreting results during the data analysis stage. This enables teams to compare forecasts with sales plans by product and channel, investigate differences using demand types, influencing indicators, and time lags, and adjust plans through the S&OP process.

‍

Connecting Industry Analysis to Forecasting Data and Model Design

Through its in-house analytics system, Impactive AI connects supply chain and commercial structure analysis with candidate indicator identification, data acquisition planning, and model design. We specify why each indicator should be examined, which products and channels it relates to, and which time lags need to be assessed, while distinguishing relationships that still require validation.

The core of this process is selecting the data a client needs and defining how it will be used in forecasting. We examine the same industry indicator differently depending on the client’s customers and contract terms, and forecast demand streams separately when they follow different patterns. Impactive AI’s approach turns an understanding of industries and businesses into practical criteria for data selection and model design.

Which indicators should you examine alongside your company’s sales and order data? If you are exploring what you can do with your existing data or how to introduce AI into your S&OP process, talk to Impactive AI. We connect data assessment, demand forecasting, sales planning, and inventory review to make S&OP easier to manage.