Client engagement SOP — outreach to close, and what recurs

Derived on 2026-07-30 from ../source/JSGlobtech_GEO_Agency_Playbook.pdf §5–§6, plus the recurring obligations implied by §3.

Thin by design. The playbook is a go-to-market document — it contains no delivery SOPs. This file covers the sales loop and names the recurring obligations each sale creates. The actual delivery runbooks (how a citation report is produced, how a client MCP server is provisioned and monitored) do not exist yet and are the main gap to close before the first Tier 2 client.

The loop

1. Research     ──▶  run the prospect's own query on ChatGPT / Perplexity, record who is cited
2. Cold email   ──▶  ../sales/outreach-templates.md   (ask: 5–10 min demo)
3. Call         ──▶  ../sales/pitch-script.md          (frame: visibility + instant lead capture)
4. Demo         ──▶  ../sales/demo-framework.md        (2 steps, live, on jsglobtech.com)
5. Proposal     ──▶  ../sales/pricing-tiers.md  ⚠ read the pricing warning first
6. Close        ──▶  setup fee + retainer
7. Recurring    ──▶  see below

1. Research (before any email)

Both outreach templates open with a factual claim about the prospect's absence from AI answers. That claim must be true.

• Run the vertical's example query from ../strategy/target-markets.md, substituting the prospect's location.
• Record which competitor is cited first — template 1 needs [Competitor Name].
• Screenshot the result. It is the strongest asset on the call and protects against the prospect saying "we do appear".
• Skip the prospect if they already rank well; the email's premise collapses.

2. Cold email

• Fill every merge field. [Your Phone Number] in template 1 is unfilled in the source.
• Only two of the four target verticals have templates. Real estate and dental need writing.
• Get a citation for the "over 40%" claim or soften it — see ../strategy/kpis.md.

3–4. Call and demo

Pre-flight, the day before (full list: ../sales/tenant-demo-checklist.md):

• MCP endpoint or OpenAPI action connected in the host client and tested end to end
• CRM_WEBHOOK_URL (and optional CLIENT_WEBHOOK_JSGLOBTECH) set in production, or nothing will arrive to show
• Lead destination (Sheet) open; Apps Script includes client_id column
• Demo uses client_id: jsglobtech so tenancy is visible on the row
• Step 1 citation query rehearsed — LLM output is non-deterministic
• Fallback screenshots + curl fallback ready for both steps
• Prefer www.jsglobtech.com URLs (apex 308-redirects)

Post-demo: delete the demo lead row so it does not pollute real leads.

5. Proposal

⚠ Read the contradiction warning in ../sales/pricing-tiers.md before writing a number. The published llms.txt says GEO/MCP is scoped per engagement and sets a $5,000 floor on the cheapest published service; the playbook tiers start at $3,000. Both are visible to prospects and to ChatGPT.

Scope Tier 3 items (multi-endpoint tools, live inventory/property sync) as custom engineering — they are unbuilt.

Recurring obligations after a close

These are the promises embedded in the tiers. Every sale increases this load permanently.

Obligation	Cadence	Tiers	Tooling today
Citation report — where the client appears across ChatGPT / Gemini / Perplexity	Monthly	1, 2, 3	None. Manual, and no documented methodology
MCP server hosting & monitoring	Continuous (sold as 24/7)	2, 3	None. No monitoring, alerting, or uptime checks in the repo
Competitor displacement strategy	Bi-weekly	3	None. Manual analyst work, undefined process
Keep the client's GEO artifacts current	As their business changes	1, 2, 3	Manual edits to their llms.txt and schema

Two structural risks worth restating: Tier 3 carries roughly 3× the reporting cadence of Tier 1, and "24/7 monitoring" is an on-call commitment with no alerting behind it. Both are margin and reputation exposure — see ../strategy/kpis.md.

Citation reporting needs a written method

Because LLM answers are non-deterministic, a report is only defensible with a frozen prompt set per client, named engines and modes, a fixed number of runs per prompt, and citation frequency rather than a yes/no. Without that, month-over-month movement is noise. Writing this methodology is the highest-value missing artifact in the whole engagement model.

Inbound lead obligation

Independent of the sales loop: LEAD_SUCCESS_MESSAGE in lib/leads.ts tells every agent-submitted lead that "we will reach out on WhatsApp/Email within 1 hour". An AI agent makes that promise unattended, at any hour.

• Nothing in the repo notifies the owner when a lead lands (../integrations/lead-capture.md).
• The only record is the Google Sheet; there is no retry if the webhook fails.
• Check the Sheet at least daily, or add a notification, or soften the message. Currently the promise is stronger than the tooling.

Troubleshooting: the "leads are not arriving" runbook is in ./README.md.

Gaps to close

• Written citation-report methodology and a template
• Monitoring/alerting before selling a Tier 2 "24/7" commitment
• Notification on lead arrival, to make the 1-hour promise real
• Provisioning checklist for deploying a client's MCP server
• Outreach templates for real estate and dental
• Resolve the pricing contradiction (../decisions/0005-productized-tiers-vs-custom-scoping.md)