Every channel counts a conversion it deserves
Ad platforms, analytics, and the CRM all report conversions, and their totals never match. Each measures something slightly different, and each has an incentive that points in the same direction.
The instinct is to pick one source and standardize. That works until the question changes. The platform's view is right for optimizing a campaign but wrong for reporting revenue. Forcing a single answer makes one of those decisions worse.
So the useful output is not a reconciled number. It is a statement of which source answers which question, and where the remaining gap genuinely cannot be resolved.
Admitting the unresolvable part is what makes the rest credible. A reconciliation that closes every gap has almost certainly closed one by assumption.
How the agent is built in Agent Studio
Channel trust is what gets configured. Prompt skills carry each source, what it measures, its attribution window, and its known bias. Self-reported platform conversions are not neutral, and the configuration should say so.
The system prompt requires that the answer be framed per question rather than as a single reconciled figure, and that it explicitly name what cannot be reconciled.
A flow gathers the numbers: HTTP Request nodes per platform, a HubSpot or Salesforce node for the CRM, a PostgreSQL node for internal records, and a Merge node to assemble them.
A Math node computes the gaps before the agent reads anything, keeping arithmetic out of the reasoning step.
Agent Chat hands over the assembled comparison. Notion receives the reconciliation with a per-question recommendation, and Slack posts the headline gap.
Google Sheets logs the gaps over time. A stable gap is a methodological difference. A moving gap is usually a tracking break. Only history distinguishes them.
What this agent is built from
- Prompt skills: each source, what it measures, its window, and its known bias.
- System prompt: answer per question rather than one figure, and name what cannot be reconciled.
- HTTP Request (Util): pulls platform numbers. Salesforce and PostgreSQL cover CRM and internal records.
- Math (Util): computes the gaps before the agent reasons about them.
- Agent Chat (Util): hands over the assembled comparison.
- Google Sheets (Integration): logs gaps over time so that a stable difference can be distinguished from a new break.
- When an ActionFlow is enough: If you only need the numbers side by side, build the workflow. Deciding which source to trust for which question is the agent's job.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.