Available nowRead-onlyClaude and ChatGPT, with set-up guides
Why it works
We do all the heavy lifting in collecting, preparing and contextualising the data so your agent doesn't have to.
Agents are indeterministic by design. They drift. They make mistakes. You need the right answer every time.
Going through layers of data and systems is a necessity but agents don't have the ability to do it on their own.
Data is like a rubric cube, you need all sides to solve the puzzle.
Enlarge ⤢
Example
Three different poeple ask in three different way, but they all need the same answer
Point two people's agents at raw logs and they can get two different stories from the same question on the same day. Three people ask about the same drop, each in their own words and their own AI. They get the same answer, because the meaning of every field is fixed before any of them asks.
Ad ops, in Claude
“Why did Channel A fall this week?”
Channel A revenue, down 16% on last week.
Revenue, in ChatGPT
“Where did we lose revenue this week?”
Channel A revenue, down 16% on last week.
Data team, in your own agent
“Revenue by channel, this week against last.”
Channel A revenue, down 16% on last week.
Example data. The channel and figures are made up.
Six stages
Think of Watching That as the refinery between the crude oil that is your data lake and the hyper engine that is your agent.
An agent on raw logs burns context on rows, reads IDs it can't resolve and gets less accurate as the data grows. Prevent this nondeterministic drift and risk to your business by preparing and presenting the data in the optimal way for your agent.
Ingestion
01
Raw feeds from your ad servers, SSPs and players, collected as they land.
Enrichment
02
Names in place of IDs, and fuller context than the log files carry, kept current as your systems change.
Unification
03
Events across systems stitched into one ad session, counted once.
Contextualisation
04
Each field carries its ad-ops meaning, so every team counts an ad the same way.
Analysis
05
Checks, alerts and reconciliation on one dataset.
MCP
06
Read-only access for the AI you already run, on each person's own login.
Enlarge ⤢
Across systems
Go beyond the What happened to How it happened with hyper-contextual time series data sessions.
Ask which advertiser, agency, DSP or buyer drove a change to your yields and eCPM, and your AI gets one joined answer.
Advertiser and agency, named as they were set up on the day.
DSP and buyer from your FreeWheel bid-level logs, by name.
Google Ad Manager line items matched to SpringServe, and FreeWheel to your partner reports.
Follow an ad down to its bids, and back.
Enlarge ⤢
Checked first
Have confidence and know that your data is right before it's made available to the agent
Monitors watch your feeds and key metrics, so a broken feed raises an alert before it reaches your AI.
A flatline monitor alerts you when a feed stops arriving.
Anomaly monitors flag sudden changes in requests, errors, fill rate and views.
When an ad server changes its log files, we update the connector.
One US TV network group went from outages running six weeks unnoticed to catching them within a day.
Enlarge ⤢
Everything in one place
Everything from your ad server logs, bid level logs, SSAI logs, SSP reports to your playout apps and partners.
You need to collect the logs files, turn the IDs into meaning, unify over 100 events into sessions, add all the context that drives insights and provide real time analysis to knock out anomalies and issues. All with the necessary security and compliance guardrails to keep your Agents in check.
Enlarge ⤢
Security and governance
Your security team can say yes.
Your settings are safe from your AI
The MCP server can't create or change reports, monitors or settings.
No API keys to hand out
Your AI connects on each person's own Watching That login. There are no API keys to manage.
Your agent gets the answer, never the data
Your AI receives only the answer it asked for. The data stays in Watching That.
Customer result
A+E Networks cut ad errors 70% in the first quarter.
“Normally video ad errors are only accessible in our ad server log files and before Watching That we couldn't monitor their patterns in near real time nor run effective root cause analysis, leaving alot of our revenue at risk. ”
Claude and ChatGPT, leveraging the FreeWheel, Springserve and Google Ad Manager log data like never before.
Your AI
Claude and ChatGPT, with a set-up guide for each. Microsoft Copilot and other agents that support remote MCP servers with sign-in should connect through the standard, and Snowflake Cowork should connect via custom MCP. We haven't tested these yet.
Ad servers
FreeWheel, Google Ad Manager, SpringServe
SSPs and exchanges
Beachfront, Google AdX, Index Exchange, Magnite, Microsoft Advertising, Nexxen, OpenX, PubMatic, Sharethrough, SpringServe Marketplace, TripleLift, Yahoo
How do I get FreeWheel or Google Ad Manager data into Claude or ChatGPT?
Connect the Watching That MCP server to your assistant and sign in with your Watching That login. Your assistant can then query the FreeWheel, Google Ad Manager, SpringServe, SSP and player data connected to your account, already joined and with names attached.
Does it replace our data lake?
No. Keep the data your team already models in your lake. We add the sources it doesn't cover yet and the joins between them, and your AI can call both.
Our team already models our FreeWheel logs. What do you add?
We add bid-level logs, SpringServe, SSP reports and the joins across them, the sources most lakes still hold raw. Book a walkthrough on your own data and judge the difference yourself.
Which AI agents does it work with?
Claude (on the web, in Claude Desktop and in Claude Code) and ChatGPT on the web, with a set-up guide for each. Microsoft Copilot and other assistants that support remote MCP servers with sign-in should work, but we haven't tested them.
Does it work with Snowflake Cowork?
Cowork connects outside tools through custom MCP, so it should reach the Watching That MCP server. We haven't tested it yet.
Is it read-only?
Yes. Your assistant can list your data sources, metrics and dimensions and run queries. It can't create or change reports, monitors or settings.
How far back can it look?
Detailed data covers about the last seven days. For older periods your assistant uses the daily rollups, which support fewer dimension combinations.
What can't it do?
It can't change anything, and it only sees the data sources your account can access. We don't deliver data into your own warehouse today.
Ask your ad data any question.
Connect the agent you already use and sign in with your Watching That login.
