New Optimon now works inside Claude and ChatGPT: ask in your own AI, get Optimon's analysis.See how to connect →

Can Claude or ChatGPT tell you why ad revenue dropped?

Yes, when it works from Optimon's analysis rather than an export. A worked example: one question, six Optimon tool calls, and an $11.8K weekend drop found inside a week that looked flat.

Seven small glass jars of violet liquid on an oak shelf in morning light, the last two on the right holding less than the others.

Yes, when it can ask Optimon. Connected to Optimon, Claude took one question about a week that looked flat, called six Optimon tools, and found that an $11.8K weekend drop was hiding under a 1.6% weekly dip. Nearly all of it came from two in-app ad units that stopped sending ad requests.

That is the short version. The rest of this post walks through the run step by step, with what each call returned, so you can judge whether you would trust the answer and copy the method on your own account.

What the AI needs before it can answer

A general assistant with no data can only guess. Paste a CSV export into it and it can do arithmetic on that file, but it does not know which of your rows are ad units, which source is Ad Exchange, or what last week looked like.

The Optimon connector closes that gap. It is a remote MCP server, the open protocol Claude and ChatGPT use to call outside tools, at https://mcp.optimon.io/mcp. You add it once in Claude (Customize, Connectors, Add custom connector) or in ChatGPT (chatgpt.com/plugins, Add custom MCP server), sign in with your Optimon login, and allow it. From then on the assistant can call tools such as kpis, breakdown, movers, alerts and pricing_rules while you chat, and it gets back the same numbers your Optimon dashboard shows.

By the time you ask, Optimon has done the groundwork: your data joined per ad unit, demand source and pricing rule, months of history to compare against, every floor change it logged, and alerts raised against each ad unit's own normal. The tools return those rows in both directions (what rose, what fell, what is new and what vanished) along with Optimon's own checks. Your AI puts them into words in your chat. Optimon runs no model of its own for the connector.

The question we asked

We used Optimon's demo account, the one our sales demos run on. The run happened on 18 November 2026, in Claude, through the connector. We replaced its figures with made-up numbers that add up the same way, so no real account's revenue appears in this post. The calls, their order and the reasoning are the AI's.

The question was plain:

This week looks flat. Did anything fall underneath it?

"This week" meant 11 to 17 November, compared with 4 to 10 November.

Step 1: the totals, then the same totals by day

The first call was kpis for both weeks with a daily series.

4-10 Nov 11-17 Nov Change
Revenue $215.9K $212.4K -1.6%
Ad requests 296.4M 248.9M -16.0%
Fill rate 27.4% 31.6% +4.2 pts
CPM $2.66 $2.70 +1.5%

A 1.6% dip on the week is the kind of number most of us would note and move past. The daily rows told a different story. Saturday 14 and Sunday 15 November brought in $62.8K against $74.6K on the weekend before, a drop of $11.8K. The other five days were up $8.3K. The week was flat because a good week and a bad weekend cancelled out.

Requests falling 16% while fill rate rises is the second clue. Fill rate is impressions divided by ad requests, so when requests fall faster than impressions, fill goes up. The account served fewer impressions, but it was asked for far fewer.

Step 2: which kind of traffic

Next the AI split the two weekends by request type with breakdown.

Request type 7-8 Nov 14-15 Nov Change
App $38.9K $28.5K -$10.4K
Web $33.6K $32.8K -$0.8K
Video $2.1K $1.5K -$0.6K

App traffic carried 88% of the weekend drop. App ad requests fell 37% between the two weekends, from 96.5M to 60.8M. Web barely moved. A device split agreed: smartphones lost $11.8K and desktop was flat.

Step 3: which app, then which ad units

movers by asset pointed at one app. Its weekend revenue fell from $19.4K to $10.3K, a loss of $9.1K. Its ad requests fell 38%, its CPM went from $1.84 to $1.22, and its fill rate rose from 24% to 31%.

movers by ad unit inside that app found the two units that explain it:

Ad unit Revenue change Ad requests Fill rate CPM
In-app unit A -$4.9K 1.62M to 0.23M (-86%) 48.2% to 84.6% $10.05 to $15.10
In-app unit B -$3.5K 0.86M to 0.25M (-71%) 55.4% to 79.8% $16.40 to $21.60

Together they account for $8.4K of the app's $9.1K drop. Both lost most of their ad requests. On the requests that were left, both filled better and earned a higher CPM. That is the shape of supply disappearing, not demand.

These two are high-CPM units, which is also why the app's blended CPM fell while CPM inside each unit rose. Lose most of your expensive inventory and the average drops even if every remaining impression sells for more.

Step 4: rule out the usual suspects

Before writing an answer, the AI checked the two explanations a yield manager would raise first.

Floors. pricing_rules with view=rules_for_ad_unit showed which rules served unit A, and view=events, the floor history Optimon keeps for each rule, showed no change on the main one. The account did log floor changes on other rules on 12 November, so it was right to look.

Alerts. alerts for 13 to 16 November returned four alerts: unfilled impressions, ad units without programmatic demand, ad units with no pricing rule, and direct line items at zero CPM. None of them was about a fall in ad requests.

What the AI concluded, and what it could not see

Its answer, in short: the week was flat because the weekend fell $11.8K while weekdays grew. Most of the weekend drop sits in one app, and most of that in two ad units that received far fewer ad requests. Fill and CPM rose on what remained, and no floor changed on the main rule, so the likely cause is on the app side: a release, a placement turned off, an SDK or mediation setting, or a change in how many screens users opened that weekend.

It also said what it could not check. Optimon sees what Google Ad Manager reports. It does not see the app's release history or session counts, so the AI could not tell an app change from a drop in usage. The next step it gave was to compare the app's sessions for those two weekends in your app analytics. If sessions held steady while ad requests fell, look at what shipped before 14 November.

The run took one question. Doing the same by hand means a weekly report, a daily report, three breakdowns, a floor history and an alert list, and the last two exist only because Optimon kept them. A GAM report export has neither.

Run the same check on your account

If you are an Optimon customer, connect your assistant on the connector page and paste any of these:

  1. "This week looks flat. Go day by day and tell me if one day or one request type fell while the rest grew."
  2. "Revenue was down yesterday. Split the change into impressions and CPM, then show me the ad units, demand sources and countries behind it."
  3. "For the ad units that fell most, show ad requests, fill rate and CPM for both periods, and tell me if it looks like less supply or weaker demand."

The connector also ships a named prompt, "Why did revenue move", that runs this whole sequence for a date range you give it.

Optimon is the platform that reads your Google Ad Manager and Prebid data and explains what moved and why, in the dashboard, in the morning email, and now in your own Claude or ChatGPT.

FAQ

Does the AI see other publishers' data?

No. The connector takes the account from your Optimon login on every call, so the AI reads only the publishers you see in your dashboard. An agency login reads the publishers in its group and has to say which one it means.

Can I run this in ChatGPT instead of Claude?

Yes. ChatGPT adds the same connector on the web as a custom MCP server with OAuth sign-in, and the tools and prompts are the same. The answer is written by whichever assistant you use, so the wording will differ, but the numbers come from the same Optimon tools.

What if my AI gets a number wrong?

Every finding in the answer comes with the rows it was built from, and the numbers are the ones your Optimon dashboard shows, so you can check any line there. Ask the AI to show the tool output for a step when something looks off, or ask it to break one row down further.

Sources

Contact

Let's talk yield.

Whether you want a demo, a free audit, or just want to ask us something - we're a real team that reads every email.

Response time

We reply to every message within one business day. Enterprise enquiries typically get a call scheduled same-day.

Direct line

Prefer email? hello@optimon.io

Not sure where to start?

Drop your domain into the homepage scanner. We'll show you what we find before you commit to anything.

Free scan →