Agent Stage 1 · Propose Verification pending

Digital Program Attribution Agent

Answers what a CX digital program did for retention, by reading program engagement and the FP&A revenue base through one attribution model, in NRR and GRR terms.

Request a change

ABOUT THIS AGENT

What it does

CX runs digital programs — in-product guides, lifecycle email, community, education, adoption and renewal campaigns — and is asked what they returned. The answer lives in two places: engagement data on one side, the FP&A revenue base on the other. This agent joins them through the digital program attribution model and reports a program in the retention terms the business already uses, NRR and GRR.

It runs as an MCP over DataHub and Snowflake and is used in Claude. It reads the FP&A metrics as published rather than recomputing them, so a program review and a finance review start from the same number.

How it works

1A program owner asks, in plain language, how a program performed — overall, by segment, or for one cohort.

2Claude calls the MCP.

3The MCP reads program engagement, account, and FP&A retention data from Snowflake, and the definitions behind them from DataHub.

4It applies the attribution model: which accounts the program touched, inside which window, against a matched set it did not touch.

5Claude returns the NRR and GRR read with the window, the comparison set, and the caveats stated. The program owner reviews it before it is used.

AGENT
Digital Program Attribution Agent

TOOLS

DataHub

Snowflake

MCP

Agent details

WHAT IT KNOWS

Data and definitions

Digital program engagement, the FY26 account hierarchies, and the FP&A retention base: NRR and GRR as published. Each input is defined in the semantic layer below.

EXAMPLE REQUEST

How to use it

“What did [PROGRAM] do to NRR in [SEGMENT] last quarter, against accounts it did not touch?”

WHAT NEEDS A HUMAN

Approvals

The model shows association, not proof of cause. The program owner reviews every read, and anything stated as revenue impact outside CX goes past FP&A first.

WHO OWNS IT

Owner

[CX OWNER]. The attribution model is built with [ANALYTICS OWNER]; NRR and GRR stay owned by FP&A.

SUCCESS MEASURE

How we measure impact

Time to answer a program performance question, today [BASELINE], target [TARGET] — and the share of CX program reviews run on this model rather than a hand-built deck.

CONNECTIONS

What it connects to

DataHub · definitions and lineage

Snowflake · program engagement and the FP&A retention base

MCP · where CX teams use it

SEMANTIC LAYER

The definitions behind the agent

The attribution model is only as good as the words under it: what counts as a touch, how long credit lasts, and whose retention number is the retention number. Defined once, these keep CX reviews, FP&A reporting, and other agents on the same figures.

Net revenue retentionThe FP&A definition, read as published. The agent reports it and never recalculates it, so a CX read and a finance read cannot drift apart.

Gross revenue retentionAlso FP&A's, and read the same way. GRR isolates churn and downgrade, which is what most CX programs are actually aimed at.

Account hierarchy joinThe FY26 hierarchies, so a program touch at a subsidiary and the revenue at its parent roll up to the same account. Shared with the Account End of Year Planning Agent.

Program touchWhat makes an account touched by a program, per channel: the action that counts, who has to take it, and the minimum engagement below which it is noise.

Attribution windowHow long after a touch a retention outcome is still credited to the program, and how credit is split when several programs touch the same account before one renewal.

Comparison setThe untouched accounts a program is read against, matched on segment, size, vertical, and renewal timing. Without it the model reports a correlation and calls it a result.

WHO OWNS IT

Definitions and upkeep

FP&A owns NRR and GRR. CX owns touch, window, and comparison set. Maintenance and request intake sit with the GTM AI team.

WHERE IT GOES NEXT

Reuse across agents

Once a touch and a window mean one thing, the same model can read renewal plays, adoption campaigns, and partner-led motions — and give planning agents the retention side of an account.

TO CERTIFY

Before verification

[RESOLVE: touch and window signed off with CX ops, comparison set agreed with FP&A, and one program back-tested against a known quarter]

Where it is in the process

YOU ARE HERE

1 Propose

Scored on value, feasibility, and risk; data checked.

2 Stage

Built and tested on sample data; analytics and tech sign off.

THEN

3 Pilot

Small group; a person approves every output.

THEN

4 Deploy

Wider rollout, with permissions and logging on.

Guardrails on every agent

Governed

Data comes through DataHub and Snowflake, not ad-hoc copies. Retention metrics are read from FP&A, never restated.

Reviewed

Every read states its window, its comparison set, and what it cannot claim. The program owner approves before it travels.

Permissioned and logged

[CONFIRM: access controls and logging for the MCP, including who may see account-level revenue]

Have a request for this agent?

Bring it to the GTM AI team. We'll point you to the right door.

Submit a request