Post-Delivery Relationship Health Monitor

The watch starts when the last invoice is issued, not when the steering pack is filed. Until then you still have an engagement manager, a weekly, and a reason to be in the client's mailbox. After that the team demobs, the channel goes quiet on purpose, and the account partner keeps the relationship without a project to hide behind.

This is not the client satisfaction sentiment monitor. That tool sits on a live job, reading emails and transcripts for cooling replies and escalation. After close you are not scoring how they feel about last week's workshop. You are watching whether the named people still exist, and whether anyone on either side still talks.

A dashboard that says the account is at risk is not the product. The product is a flag the partner can act on this week: who went quiet, who left, and what you are allowed to know. Watch only accounts you intend to return to, with a named partner who still owns them. An orphaned close with no partner is a CRM hygiene problem, not a model problem.

Two signals earn a flag. Tone does not.

Contact frequency. Count consented, work-related touches after close: mail in the account mailbox the partner agreed to watch, meetings on the shared calendar, logged calls in Salesforce. Compare to a baseline the partner set for this account. A quarterly check-in is not the same as a retained advisor who talks weekly. A drop against that baseline is a flag. A drop against a firm-wide healthy-account curve is noise.

Named-stakeholder change. Freeze who mattered at close: economic buyer, day-to-day sponsor, procurement, the operator who lived with the deliverable. Watch those people, not the whole company. Job-change tools in the LinkedIn Sales Navigator class can tell you a title moved. If Salesforce still lists last year's CPO as the sponsor, the model will miss the only event that matters.

Do not score happiness. Sentiment models on after-close mail will grade politeness and brevity as cooling. A quietly unhappy sponsor still writes thanks. A happy sponsor in a blackout writes nothing. Frequency and named people are blunt, and they are the signals you can defend if the client asks what you are watching. If you feed post-close mail into a sentiment dashboard, you have rebuilt the live-job monitor on a thinner, more private corpus. Stop.

A green CRM tile can hide a sponsor who already left. The following is a worked example with made-up firms, not a measured case.

Harbor & Lane closed a 12-week inventory diagnostic for Northwind Retail. Last invoice: April. Named sponsor: Priya Chen, VP of supply chain. The partner logged a warm close in Salesforce.

By July the consented mailbox and calendar showed one thanks note in week one, then nothing. Sales Navigator flagged Chen as Head of Supply Chain at a different retailer. Salesforce still had her as Northwind's sponsor, relationship health green.

A happiness score would have left the tile green. The close email had also said Northwind was in a systems freeze until September. Nobody encoded it, so frequency flags would have fired all summer on a planned pause. What mattered: Chen left. The CFO who signed the SOW and the operator who ran the pilot stores were still there and had not been contacted since the close dinner.

The right flag is two lines to the account partner, not an account-health number:

• Named sponsor Priya Chen is no longer at Northwind (Sales Navigator job change, as-of July). Salesforce still lists her as primary. Confirm before anyone emails Northwind.
• No consented contact with the CFO or the store-ops lead since the last invoice. Quiet period ran through September. Do not treat that silence as churn. After September, the partner owes those two people a call.

The partner called the CFO in October, named the diagnostic, and asked who owned supply chain now. That is the job. An auto-generated checking-in mail to Chen's old address, or a sentiment alert forwarded to the client, would have been the failure.

Honor mailbox consent. The monitor may see mail the account partner (or a named coverage owner) has agreed to include: the client domain on threads they were on, the project mailbox, meetings they accepted. It may not crawl the firm's entire Outlook tenant looking for the client's name. It may not read a partner's private thread with a coach, a legal-hold folder, or a personal mailbox that happens to contain the client.

• In: the project mailbox, the partner's consented client folder, calendar meetings with the named contacts, Salesforce activities those people logged, public job-change events for the named list.
• Out: side conversations the partner did not enroll, staff-level chatter, anything the client marked confidential at close, recordings the client did not agree to keep after the job, and any thread whose participants are only internal.

If you would not tell the client you are counting this mail, do not count it. Gong-class archives follow the same rule: only calls the client knew were recorded, and only while retention still allows them. Do not reopen a recorded steering session to score the vibe six months later.

Salesforce records who is named and whether anyone logged a touch. It is not a sentiment engine. LinkedIn Sales Navigator is a job-change radar for the named list, not permission to message the former sponsor from a company account. Microsoft Outlook is frequency if and only if the mailbox owner consented. None of these tools should mail the client.

Route every flag to the account partner. The model does not send mail, sequence a cadence, or open a Salesforce campaign. The empty check-in is how clients learn you automated them.

The partner gets a short brief: what changed (frequency versus their baseline, or a named person who left), evidence they already own (last consented touch, last meeting, job-change source and date), what not to do (do not email the departed person's old address; do not forward this brief to the client; do not treat a freeze in the close notes as churn), and a suggested next human step (call the remaining economic buyer, or wait until the freeze lifts).

A follow-on opportunity signal detector can give a real reason to talk, an unresolved problem from the close pack. A reusable case study drafter can give a consented, anonymized proof point when the client actually wants a write-up. Neither is a reason to auto-mail. If delivery was rocky, wait.

If the next conversation is a pitch, the partner still needs a pre-pitch client context brief that does not treat last year's sponsor as current. Job-change flags belong there as dated facts, not as a health score.

Silence after close is often the plan. Implementation blackouts, year-end, and procurement freezes are written into close emails more often than they are written into Salesforce. If the monitor treats every quiet quarter as decay, the partner will mute it in a month.

Encode the quiet period at close: start date, end date, who asked for it, and which named people it covers. Suppress frequency flags until that date. Still watch named-stakeholder turnover. A sponsor can leave during a freeze, and that is still a call to someone who remains, not a checking-in to the old alias.

Kill these in the first trial: a 1-to-100 happiness tile (delete the score, keep the two signals), an allowlist that includes private threads the partner would not show the client, a watchlist of every closed invoice, and any auto-cadence. If Salesforce still names a departed sponsor, the flag is that CRM is wrong, and also do not email that person.

Judge the trial on whether the partner made a better call, not on alerts sent. A useful month is one sponsor-leave caught before the next RFP, one freeze respected, and zero client-facing mail from the system.

When a closed account comes back to bid, win/loss pattern synthesis is a different corpus: why you won or lost, not whether anyone emailed last quarter. Do not mix a health flag into a close-reason theme.