AI fixes many things, but you still need human judgment to make sure that you are solving the right problems.
In digital publishing, it’s incredibly tempting to hand the keys of inventory forecasting over to an algorithm and hope for the best. But an algorithm doesn’t understand your business strategy, your changing ad layouts, or the tectonic shifts happening in audience traffic. It just looks at historical numbers. If you rely purely on the machine, you risk leaving millions on the table—or worse, facing the asymmetric risk of over-forecasting your premium inventory.
To get forecasting right, you must be the “Human in the Loop.”
Here are the 5 main questions you need to address to ensure you are guiding the technology, rather than letting it blindside your revenue teams:
1. What is the business decision you are trying to support?
Before looking at any data, you have to define why you are doing the supply forecast in the first place. Are you trying to ensure you have the inventory available to deliver on guaranteed managed campaigns? Are you trying to make sure your ad server paces correctly? Or are you trying to meet your quarterly financial plan? A forecast built to verify campaign delivery requires deep granularity, while a forecast built for financial planning needs aggregation. The human must match the model to the specific business question first.
2. What time periods are important to your business?
The use case directly informs the level of detail you need in your forecast. While an AI can calculate traffic fluctuations by the hour, minute, or second, this level of granularity is completely meaningless for the vast majority of digital publishers. More detail is not always better; over-specifying just introduces mathematical noise. As a human operator, you need to look at how your sales, finance, and yield teams actually operate and focus the model on the daily, weekly, or monthly intervals that drive actual business value.
3. What is the right level of product aggregation?
An algorithm can generate a forecast at almost any arbitrary slice of data, but the human has to decide what aggregation matches how the business actually sells. You must guide the model to align with the inventory products represented on your rate card. At a bare minimum, this means separating your forecasts by ad formats (display, native, video), ad sizes, and individual sub-sites or sub-apps (such as distinct sections for sports vs. news).
4. How does your unique audience dynamic impact cycles?
To ensure the algorithm is picking up the right signals, you have to feed it the context of your specific audience behavior. Does your traffic peak during the week or on the weekend? Does your content have a massive annual cycle—like a sports or fantasy football site that experiences huge uplifts during a specific season and goes dead the rest of the year? Algorithms are great at patterns, but the human operator must validate that the machine understands the underlying drivers behind those traffic waves.
5. When do long-term structural trends and business updates override historical data?
This is the ultimate place where historical data can mislead you, and where pure business judgment must take over. A model will see what happened last year and project it forward. But if your audience is structurally shifting away from traditional search bars toward AI-driven agentic search, the past is no longer a guide to the future. Whether traffic trends are shifting, a product is being redesigned, a major distribution partner is leaving, or a new app release is altering user behavior—these are strategic realities that only a human can build into a future forecast.
The Bottom Line
Stop treating inventory forecasting as a passive math problem. It is a core pillar of your yield management strategy. AI is a phenomenal force multiplier for data processing, but it requires a pilot to point it in the right direction.