Agent Stage 1 · Propose

Conversation Insights Agent

Reads sales conversations in Snowflake, scores each one for sentiment and competitor mentions, and turns them into a report for leadership and enablement.

Request a change

ABOUT THIS AGENT

What it does

Today, what buyers say in calls and chats stays buried in transcripts. This agent reads them in Snowflake, finds the DataHub definitions for each field, and reports how buyers feel and which competitors come up. It is read-only: it writes nothing back to Salesforce, and a person reviews every report before it is shared. This is a proposal at Stage 1.

How it works

1Someone asks for a report, naming a period and a segment such as a region or a team.

2The agent uses DataHub, through MCP, to confirm the right Snowflake views and what each field means.

3It pulls conversations from Snowflake, through MCP, and masks names and emails before the model reads them.

4It rates each conversation as positive, neutral, negative or mixed, with a reason and a short quote, and notes any competitor mentioned.

5It rolls the results up into a report where every claim links to its conversation ID.

6A person on the enablement or GTM AI side reviews the report before it goes to leadership.

AGENT
Conversation Insights Agent

TOOLS

Snowflake

DataHub

MCP

THE OUTPUT

What the report looks like

One report with two views. Both roll up to team, region and segment, never to an individual rep, and small groups are hidden so no quote can point to one person.

FOR LEADERSHIP

Direction and risk in a minute

A three-line headline, the sentiment trend against the prior period, competitor share, and a watch list of open opportunities with negative or competitive conversations. Monthly, or on request.

FOR ENABLEMENT

Material to act on

The objections that keep coming up, how buyers talk about competitors, evidence quotes, and recommended actions tied to that evidence. Weekly digest.

HeadlineThree lines: sentiment movement, the competitor in focus, and how many conversations need follow-up.

SentimentPositive, neutral, negative and mixed shares, the change against the prior period, the top drivers in each direction, and where the agent disagrees with the recorded sentiment.

CompetitivePer competitor: conversations, share, sentiment when mentioned, and the usual context. Plus the tools prospects use today, and short buyer quotes.

Watch listOpen opportunities with negative or competitive conversations, with stage, owner and next step.

Recommended actionsThree to five, each tied to the evidence behind it.

Data notesCoverage, missing transcripts, excluded records, confidence, and links to the definitions used.

See a sample report

Agent details

WHAT IT KNOWS

Data and definitions

One Mind conversation records, live chat transcripts, and Outreach-attributed tasks in Salesforce, plus account and opportunity context. Each input is defined in the semantic layer below.

EXAMPLE REQUEST

How to use it

“Give me conversation insights for EMEA for the last 30 days: sentiment trend, which competitors came up, and what enablement should act on.”

WHAT NEEDS A HUMAN

Approvals

A person reviews every report before it goes to leadership. Sentiment and competitor tags are model estimates, so each one shows its evidence and confidence.

WHO OWNS IT

Owner

[OWNER TBD]

SUCCESS MEASURE

How we measure impact

Baseline: [BASELINE TBD]. Target: [TARGET TBD]. We plan to check agreement with human-labelled sentiment and competitor tags on a test set.

CONNECTIONS

What it connects to

Snowflake · where the conversations are queried

DataHub · definitions, owners, certification

MCP · how the agent reaches both

Outreach · sequences and conversations

SEMANTIC LAYER

The definitions behind the agent

What each input means and where it comes from. The views below are in the SALESFORCE_RAW schema in Snowflake and are described in DataHub.

One Mind conversationsConversation intelligence records: full transcript, summary, recorded sentiment, key topics, current tools in use, next steps, outcome, and BANT, SPICED and MEDDPICC fields. Linked to the account, contact and lead.

Live chat transcriptsChat text with start and end times, response times, and lead, contact and account IDs.

Outreach tasksSalesforce tasks carrying the attributed Outreach sequence name, call disposition and duration, and call sentiment where recorded.

Accounts and opportunitiesContext for the watch list: account, stage, amount and owner.

SentimentThe agent's own rating per conversation (positive, neutral, negative, mixed), shown next to the recorded sentiment where one exists. [CONFIRM: how sentiment is recorded and which values it holds]

Competitor mentionA named competitor or a tool the prospect uses today, with the context it was mentioned in. [COMPETITOR LIST TBD]

WHO OWNS IT

Definitions and upkeep

[OWNER TBD]. The source views sit in the Sales Ops domain in DataHub.

OPEN QUESTIONS

To confirm

[CONFIRM: whether Outreach message and chat text is stored in Snowflake beyond the Salesforce tasks] [CONFIRM: whether the model may read transcript text] [CONFIRM: read-only reporting, or also write back to Salesforce]

TO CERTIFY

Open definitions

[FIELD AND METRIC DEFINITIONS to confirm with Sales Ops and Data Science]

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, permissions and logging on.

Guardrails on every agent

Transparent

Every insight shows its source conversation IDs and a confidence level.

Reviewed

A person approves each report before it is shared with leadership.

Permissioned and logged

Read-only access limited to the views above, with names and emails masked and every query logged. [CONFIRM: access role and logging]

Have a request for this agent?

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

Submit a request