This document continues the discussion on how AI could influence end to end workflow in Revenue Operations. Part 1 covered Proposal Workflow. Part 2 of this series examines Inventory Forecasting.
A couple of caveats.
- This explores inventory forecasting’s impact on direct sales (including programmatic guaranteed) – but does NOT cover other programmatic buy types. I’m acknowledging that while programmatic may be the ultimate future for us all, direct sales is not dead yet. So how can we improve their use of inventory?
- I want to acknowledge the industry contributions of Tom Shields who founded Yieldex, the platform that challenged the notion that publishers were chained to the limitations of GAM inventory forecasting.
What’s wrong with inventory forecasting today for direct sales? Are we discussing a problem that doesn’t exist? After all, GAM has a 24 month lookback at inventory, to help incorporate the impact of seasonality or events on availability. GAM has a manual inventory adjustment feature that allows publishers to “tweak” the forecast based on factors that may be unique to their domain.
But there is room for improvement. If programmatic has taught direct sales anything, it is that at scale, incremental changes in how inventory is managed can result in significant increases in revenue.
I think AI can play an important role here, by not only ingesting data that occurred in the past, but quantifying the impact that 1) real time events and 2) future events will have on inventory. Thus, providing more accurate data for proposals and campaign optimization.
Predictive Inventory Alerts
Perhaps direct sales can learn a lesson from programmatic. When there are unexpected spikes in ad inventory, programmatic is used for backfill, which is fully automated.
I don’t believe ad platforms do a good job of leveraging the data we can derive from CMS (content management) platforms such as page views, which correlate to ad impressions.
Perhaps we can use AI to intercept CMS data, analyze spikes in activities, and send an alert to a custom field in GAM which in essence says “hey, for the next 3 days, you can sell X% more inventory in the business/finance section”. An inventory manager could use that intel to engage with direct sales to sell the overflow.
Forecasting Based on Future Events
Depending on the publisher, here are some events that will take place in the future that could influence inventory forecasting:
- Editorial features, scheduled throughout the year
- Midterm Elections, national and local
- Tentpole events, awards ceremonies
- Retail promotions and seasonal influences
- New product releases
- Sporting events. Superbowl, FIFA, Olympics, etc. .
- Weather events and seasonality
Combining this future facing data with historical lookbacks could provide a more intelligent inventory forecast. By training AI to ingest this large data set (or “corpus” as it is sometimes referred to), I think we’d have a better method of prediction.
Do we know what the 2026 midterm election will be on certain types of inventory? Do we have to do a “back of the envelope” guess to arrive at that number? We might get a more accurate reading if AI was able to combine historical lookbacks with the future events, unlocking the potential for more direct sold inventory.
Is all this wool gathering even worth it? Here’s an example.
A publisher with 150,000,000 sold ad imps / month @$25 cpm = $3,750,000 ad revenue
If the methods above yielded a15% uplift = $562,500 a month, or $6,750,000 per year.
Another $6MM a year through more effective inventory management would be worth it.
DMW MediaBlog 