AI Token Pools and Assistant Governance
Pool sizes by plan (reference)
| Plan | Monthly pool | Pool type |
|---|---|---|
| Free Forever | 500K | Org-wide |
| Starter (standalone) | 500K | Per person |
| Team (standalone) | 2M | Per person |
| Growth (standalone) | 5M | Per person |
| One Team | 30M | Org-wide |
| One Growth | 100M | Org-wide |
| One Scale | 300M | Org-wide |
Exact tables and pack SKUs are on app.salestrics.com/pricing and in Changelog #17. Tier upgrades increase pools — token packs bridge a heavy month without permanent tier inflation.
One-time token packs (when to buy)
| Pack | Tokens | One-time price | Typical use |
|---|---|---|---|
| Micro | 1K | $0.50 | Single heavy eval day |
| Starter | 10K | $0.99 | Team pilot week |
| Plus | 100K | $1.99 | Quarter-end read summaries |
| Pro | 1M | $5.99 | Migration assist burst |
| Boost+ | 5M–120M | $29–$299 | Sustained org-wide programs |
Pack rules: bonus tokens roll over until used; packs are not sold during free trial. RevOps should approve pack purchases — not every rep self-serving Boost during a demo prep crunch.
Read-first governance (MCP and Assistant)
Agentic CRM value is context on the account graph — mail, pipeline, cases, docs — not autonomous writes on day one. Roll out in order:
- Read prompts only — summaries, hygiene lists, forecast narratives
- Named accounts — top 10 logos before org-wide automation
- Org MCP keys — rotated, scoped, never in public channels
- Write tools — after mail-on-account and stage policy exist
- Audit monthly — who ran write actions and on which records
Architecture primer: agentic CRM guide. MCP setup: Salestrics MCP and Cursor MCP playbook.
What burns pools fast (and fixes)
| Burn pattern | Why it happens | Fix |
|---|---|---|
| Whole-book summaries daily | No account filter | Top-account Monday ritual only |
| Rewrite every email draft | Mail not on account — thin context loops | Mail-on-account policy first |
| Write tools on bad data | Agents fix hygiene repeatedly | Archive stale records; fix stages |
| Duplicate prompts in Slack | No shared prompt library | RevOps publishes 5 approved prompts |
| Eval sandbox on production | Testing MCP on every opp | Dedicated eval accounts |
Pool exhaustion with clean data usually means tier mismatch — see Starter vs Team vs Growth before buying the largest pack every month.
Free Forever pool discipline
The 500K org pool on Free Forever is enough to prove Assistant and read-only MCP — not to run org-wide automation:
- Two teammates max — align on shared prompts; no duplicate account summaries
- Read-first only — no write tools until paid tier and policy
- Upgrade signal — pool empty mid-month with real work left → paid tier or pack after checkout
- Not a substitute for CRM hygiene — agents cannot fix 10K stale contacts cheaply
Upgrade path: Free Forever to paid.
RevOps monthly ritual (20 minutes)
- Admin usage view — pool % consumed vs days left in cycle
- Top consumers — legitimate power users vs runaway prompts
- Pack inventory — rolled-over balance; avoid duplicate purchases
- MCP key roster — active keys, last rotation, offboarding gaps
- Prompt library — retire prompts that burn tokens without pipeline impact
Pair with billing review: in-app billing RevOps playbook.
Dashboard AI hub (Changelog #17)
The AI dashboard hub connects Assistant to live product actions — not a sidebar disconnected from mail and pipeline. Operators should:
- Start from dashboards you already trust — pipeline, desk queue, mail home
- Run read actions first — explain variance, list stale opps, summarize threads
- Document wins — one prompt that saved a manager 30 minutes becomes a standard
- Escalate writes slowly — hub actions that mutate records need the same policy as MCP write
Product overview: Salestrics Assistant.
Security baseline for AI and MCP
- Org-owned keys — not personal API keys on shared laptops
- Offboarding same day — revoke MCP access when someone leaves
- No customer PII in public prompts — use account context inside the workspace
- Vendor review — security questionnaire covers MCP scope (see evaluate guide)
- Read default in Cursor/Claude — write tools explicit opt-in per environment
Evaluation checklist: evaluate agentic CRM and MCP API key governance.
When to upgrade tier vs buy a pack
| Signal | Buy pack | Upgrade tier |
|---|---|---|
| One heavy month (QBR, migration) | Pro or Boost | — |
| Pool empty 3 months running | — | Team → Growth or One |
| One power user on Starter | Plus for burst | User to Team long-term |
| Org-wide program on One Team | Standard/Growth pack | One Growth if sustained |
Anti-patterns
- Pack as permanent tier — recurring Boost every month → upgrade math
- Write MCP before mail policy — agents mutate records nobody trusts
- No usage owner — finance sees packs; RevOps sees nothing
- Prompt free-for-all — 50 custom prompts nobody maintains
- Ignore tier pools — five people on Starter with 500K each still need discipline
Approved prompt starters (RevOps library)
Monday pipeline read: Summarize open opps over $X with no mail in 14 days and recommend next actions — read only.
Account prep: From mail and cases on [Account], list open risks before tomorrow's Connect call — read only.
Desk queue: Group P1 cases by account with renewal dates in the next 90 days — read only.
Publish three to five prompts org-wide before enabling write tools. More prompts live in MCP workflow prompts.