The agency snapshot

An agency's real product is a configuration. The same stages, the same cadences, the same guardrails, applied to a new client in a day rather than a fortnight. Publishing the object list means you can build that whether or not you use our tooling.

1. Pipeline stages

A lead-gen agency's client usually needs two pipelines rather than one.

• New business: New lead (10%), Contacted (20%), Qualified (35%), Meeting booked (50%), Proposal sent (65%), Negotiation (80%), Won, Lost.
• Client delivery, if you track it here: Onboarding, Active, At risk, Renewed, Churned.
• Probability on every stage, because the weighted forecast is a sum of the board rather than a separate spreadsheet.

2. Dispositions and outcomes

• Call outcomes: connected, voicemail, wrong number, not interested, callback requested, booked.
• Meeting outcomes: held, cancelled, no-show, rescheduled.
• Loss reasons: price, timing, chose a competitor, no budget, no decision, wrong fit. Six or fewer, or nobody uses them honestly.

3. ICPs and segments

• One ICP record per target the client sells to, with titles, size, industry and geography.
• A segment per campaign, drawn from the data bank against the ICP.
• A do-not-contact segment for the client's existing customers and their named competitors.

4. Cadences

• Cold outreach, four steps over twelve days.
• Inbound reply, three steps over five days.
• Reactivation of a stale list, three steps over fourteen days.
• Referral partner outreach, four steps over thirty days.
• Every one with a reply pausing the sequence for that contact.

5. Suppression

• The client's existing customer list, imported before anything sends.
• Their competitor domains.
• Anyone who unsubscribed from their previous provider, if you can get the list.
• Applied at audience freeze and re-checked at send time.

6. Agent policy

• Every agent in review, which is the seeded default.
• Guardrails: three days minimum between touches, per-cadence cap, per-mailbox and per-domain daily caps, business-hours window.
• The kill switch owner named, and told they own it.
• Auto only for classification, pausing on reply, and suppression on unsubscribe or bounce.

7. Access scopes

• Your team, scoped by channel and area to what they work on.
• The client's own read access, if they get one, scoped to their pipeline and reports.
• MCP tokens issued with the narrowest scope that does the job, and revoked when a project ends.

Build the seven groups on your first client instance, then export them as a snapshot. Applying that snapshot to the next instance creates the same stages, dispositions, cadences, suppression rules, agent policy and access scopes in one action. The MCP twins are snapshot_export and snapshot_apply, so the whole step fits in a script.

If you'd rather keep the setup as code, write it as a script that calls the MCP tools: pipeline_stage_upsert for stages, icp_create and segment_create for audiences, suppression_add for the exclusion lists, agent_policy_set and agent_guardrails_set for the rules, and user_grant_channel and user_grant_area for access. Unlike a habit, a script can be reviewed in a pull request.

Branding is part of the same setup. Your name, logo, theme and subdomain go on the app itself, so the client logs into your agency's CRM rather than ours.