Bid change explanations from an AI: ask for the raw numbers or the reasoning?

When an AI assistant justifies a bid or budget change, I'd ask for the exact figures it used (spend, conversions, date range, comparison period) instead of the reasoning. The explanations come out fluent whatever the data says, so "intent has drifted" reads convincingly even when the numbers are flat.

I'm not sure that holds at low volume, though. With thin data the figures are mostly noise, and the model will still write a story around them. So maybe the real check is whether the cited numbers would still support the change if you rounded them. Do people here ask for the raw data, the reasoning, or just check the outcome after the change?

Agreed on the rounding test. One more thing to check before anyone trusts the comparison is the conversion lag window. If a bid change is judged on the last 7 days of conversions, it can look like a collapse when half of those conversions haven't landed yet, so the "drift" story ends up built on incomplete numbers.

Lag is the right thing to look at. One more wrinkle: Microsoft Ads reports conversions by conversion date unless you switch the column, so a 7-day pull can look like a drop just because late conversions haven't landed yet. Pull the same window from 14 to 30 days back and compare against that before you believe any drift story.

Depends on volume. If you're getting 40 or more conversions a week, looking back 14 to 30 days will usually show whether the lag was hiding a real drop. At three or four a week, older windows just add noise, and a slow week looks the same as a real decline. In that case I'd measure how long conversions typically take to arrive for that action first, then only compare windows that sit past that delay. Otherwise you're comparing two incomplete pictures.