Can AI predict advertising results?
AI can predict advertising results within a range when there is enough historical data, such as past click costs, conversion rates and seasonal patterns. It cannot reliably predict events outside that data, such as a new competitor, a price change or a sudden shift in demand. Good AI forecasts are probabilistic ranges, not guarantees.
Where does AI forecasting work well?
- Seasonality: predicting how search volume and costs rise and fall through the year.
- Budget response: estimating how leads and cost per lead change as budget increases.
- Conversion patterns: spotting which audiences, times and placements convert.
- Anomaly detection: flagging when spend or performance moves outside the expected range, early.
Where does AI forecasting fail?
AI forecasting fails when the future does not look like the past, when the data is wrong, or when it lacks business context. Bad conversion tracking is the most common cause: a model trained on inaccurate conversions will forecast inaccurate results with great confidence.
- New businesses or offers with no history
- One-time events, where there is no prior year to learn from
- Tracking that counts the wrong actions
- Missing business data such as close rate, margin and customer value
What is the difference between platform forecasts and predictive advertising?
Ad platform forecasts, such as Google’s Performance Planner, predict clicks and conversions inside that platform. Predictive advertising combines those forecasts with your own business data, including close rates, customer value and margin, to project revenue and return across channels.
That is the approach behind Cheeky Merlin, the predictive marketing platform we built and use for every client projection. Read predictive advertising, explained for the full method.
How do you evaluate an AI forecasting tool?
- Does it show its assumptions? You should be able to see the inputs behind the number.
- Does it give a range? A single precise number is a warning sign.
- Does it use your business data? Revenue forecasts need close rate and customer value.
- Does it check itself? The best tools compare forecasts to actual results and show their accuracy over time.
- Can a person explain it? If no one can explain why the forecast changed, it is hard to act on.
You can test the basic forecasting math yourself with our ROAS calculator and step-by-step forecasting guide.