Accounts with sentiment shift in the last 30 days
Customers whose tone moved meaningfully: before the renewal call gets weird.
- Best for
- Monday CS standup
- Primitives
- AccountsSentimentTrends
The prompt
You are scanning the customer base for early warnings. Using Sentra: 1. For every Account with an active contract, compute the average sentiment of Interactions in the last 30 days vs. the prior 90 days. 2. Flag any account where the shift is more negative than one standard deviation from its own baseline. 3. For each flagged account, return the three Interactions that drove the drop — with the quoted moment Sentra extracted, the speaker, and the surface. Output a table sorted by magnitude of negative shift: | Account | Δ sentiment | Most-cited concern | Source date | For the top three, add a one-paragraph "what to do this week" — referencing the open Commitments, the next renewal date, and the right internal owner.
See it work
Run it from
Claude
Cursor
ChatGPTAny agent · MCPHow it runs
- 01Copy the prompt into any agent connected to Sentra over MCP.
- 02The agent reads the company memory, scoped to what you are allowed to see.
- 03One grounded answer comes back, with every line traceable to its source.
Or automate it with Actions
The same prompt runs as a Sentra Action: on a schedule, or on a semantic trigger. Output lands in Slack or email, and every send waits for your approval.
Trigger idea: monday cs standup
Sentraover MCPSentiment deltas · last 30 days4 accounts moved this month
Meridian · negative
threads · Aug 18 to Sep 1Three clipped one-line replies from their platform lead since the August outage. Renewal in 45 days; a service credit memo is still owed.
Harbor · gone quiet
email · Aug 14Weekly touchpoints dropped to none for three weeks. Last message asked about CSV exports, which remain unscheduled.
Solara · wary
calls · AugTone shifted from warm to procedural when procurement changed hands in August.
Brightline · positive
call · Aug 29 · 00:31:04"This is finally clicking" after the enablement session; expansion mentioned unprompted.
The prompt above becomes this. Fictional data, real mechanics: every line cites the meeting, thread, or record it came from.
Without Sentra
Not reachable without persistent memory: the context this recipe needs was never written down anywhere an agent could retrieve it.
With Sentra
- 01For every active `Account`, query sentiment trend: last 30 days vs. prior 90.
- 02Flag accounts where the shift is > 1σ negative against the account's own baseline.
- 03For each flagged account, return the three `Interactions` that drove the drop — with quoted excerpts, the speaker, and the surface.
- 04Pull open `Commitments`, next renewal date, and the internal `Actor` owner for each.
- 05Compose the sorted table with 'what to do this week' for the top three.
~55,000 tokens