AI Customer Success

What is this? A concrete recipe for wiring an AI customer-success agent against Salty’s CRM. The agent maintains an account health score, logs every touchpoint, surfaces at-risk accounts, and drives QBR prep — all over MCP or REST.

The scenario

You sell SaaS to 50–500 customers. You can’t afford a CSM headcount per account. An LLM agent watches product usage signals, conversation transcripts, and ticket data; logs touchpoints; and when health drops, alerts you with the receipt of why. Salty is the system of record. Native objects map cleanly to the CS workflow.

Data model

Salty object	Your concept
`company`	Customer account
`person` (with `primary_company_id`)	Stakeholder at that account (champion, exec sponsor, end user)
`deal`	Renewal opportunity (one per account per renewal cycle)
`activity`	Touchpoint — call, QBR, email, in-product nudge
`note`	Free-form context (call notes, exec change, competitive intel)
`task`	Follow-up items (REST/CLI only — no MCP write tool in v1)
`custom_attributes` on company	`health_score`, `arr_cents`, `tier`, `last_qbr_at`

Set up the schema

Extend the company object once at start-of-life:

curl -X POST $SALTY_API/schema/company/attributes \
  -H "Authorization: Bearer $SALTY_API_KEY" \
  -d '{"attribute_key":"health_score","display_name":"Health Score",
       "data_type":"number"}'

curl -X POST $SALTY_API/schema/company/attributes \
  -H "Authorization: Bearer $SALTY_API_KEY" \
  -d '{"attribute_key":"tier","display_name":"Tier","data_type":"enum",
       "enum_values":["starter","growth","enterprise"]}'

curl -X POST $SALTY_API/schema/company/attributes \
  -H "Authorization: Bearer $SALTY_API_KEY" \
  -d '{"attribute_key":"arr_cents","display_name":"ARR (cents)","data_type":"number"}'

What the agent does on a typical week

Task	Tool calls	Why
Pull the at-risk list	`search_companies` with filter `custom_attributes.health_score < 60`	Daily standup feed
Update health after weekly review	`update_company` setting `custom_attributes.health_score`	Recalculated from usage + sentiment
Log a CSM call	`log_activity` parent: company, type: `csm_call`, payload includes notes + duration	Audit trail
Add a context note from Slack	`add_note` parent: company	Surface non-meeting signal
Move renewal deal forward	`update_deal` stage transition	Track $$$ at risk

Driving it from Claude Desktop

Conversational prompt:

“Pull every company with health score below 70 and tell me who their primary contact is and when we last logged a touchpoint.”

Claude will:

search_companies with {filter: {custom_attributes: {health_score: {lt: 70}}}} (the filter syntax)
For each company, search_people with {filter: {primary_company_id: {equals: <id>}}}
Reads each company’s last_touchpoint_at custom attribute (returned inline in step 1, kept current by your activity logging) to show recency
Surface a ranked table you can act on

The whole “QBR prep” workflow is one prompt instead of 40 minutes of dashboard clicks.

React to churn signal via webhooks

When your usage-monitoring system pushes a low-engagement event, your agent can update Salty and Salty fires a webhook to your alert channel:

curl -X POST $SALTY_API/webhook-endpoints \
  -H "Authorization: Bearer $SALTY_API_KEY" \
  -d '{
    "url": "https://hooks.slack.com/...",
    "subscribed_events": ["company.updated"]
  }'

Then your Slack receiver checks data.custom_attributes.health_score and pages you if it dropped below threshold. Salty does the persistence + signing + retry; you just react.

Reporting

Get a CSV-ready dump of every company with health, ARR, and last activity:

curl -G $SALTY_API/companies \
  -H "Authorization: Bearer $SALTY_API_KEY" \
  --data-urlencode "limit=100" > companies.json

The agent can run this nightly, transform to CSV, and post to your weekly dashboard.

Volume guidance

A 100-account CS practice with weekly check-ins makes ~10,000 API calls/month (~25 calls per account per week). The Solo ($20/mo, 100k calls) plan has 10× headroom. Pro ($99/mo, 1M calls) if you have hundreds of accounts.

Schema engine — custom_attributes with enum/reference/json types
Webhooks — react to state changes in real time
Concepts → Authentication — sk_live_ vs OAuth for human-and-agent flows
Concepts → Rate limits — per-plan ceilings + burst

Get Started

Use cases

Concepts

AI Customer Success

The scenario

Data model

Set up the schema

What the agent does on a typical week

Driving it from Claude Desktop

React to churn signal via webhooks

Reporting

Volume guidance

​The scenario

​Data model

​Set up the schema

​What the agent does on a typical week

​Driving it from Claude Desktop

​React to churn signal via webhooks

​Reporting

​Volume guidance

​Related

The scenario

Data model

Set up the schema

What the agent does on a typical week

Driving it from Claude Desktop

React to churn signal via webhooks

Reporting

Volume guidance

Related