A relationship follow-through system for partner-led teams

Consider a partner-led advisory practice with 25 people and a portfolio of long-running client relationships.

Partners bring in work, lead sensitive conversations, and hold context that does not fit neatly into a CRM. Associates and operators keep projects moving. Clients expect continuity: if the firm promised an introduction, a note, a proposal, or a follow-up after a board meeting, someone needs to remember why it matters.

The practice does not need AI to manage relationships on its behalf. It needs infrastructure that protects follow-through.

This is where Proximity could be tailored as a relationship follow-through system.

The system's job is to preserve shared memory, prepare context, and surface commitments for human review. It should not automate relationship judgment, decide tone, assess trust, choose strategy, or send sensitive client communication without approval.

The Workflow

The workflow is weekly relationship follow-through review.

The team wants to know:

• Which commitments were made to clients or partners.
• Who owns each commitment.
• What context makes the follow-up sensitive or important.
• Which prior conversations, documents, or decisions are relevant.
• Which promises are approaching or overdue.
• Which follow-ups need partner review before action.

CRMs are useful, but relationship-led work often depends on context that is more subtle than a sales stage. A partner remembers that a client dislikes rushed proposals. An associate knows the last draft was delayed because of missing data. An operator knows the promised introduction depends on a third party. None of that is safe to leave only in memory.

Proximity would turn commitments and context into a reviewable operating layer.

What Proximity Models

For this deployment, Proximity would model the commitment as the core object.

Approved sources might include:

• Meeting notes.
• Email threads.
• CRM records.
• Project trackers.
• Calendars.
• Proposal drafts.
• Partner notes.
• Client history summaries.
• Internal handover notes.

The system would structure those sources into:

• Person, organisation, relationship owner, and internal team.
• Commitment, source, owner, and due date.
• Related project or opportunity.
• Sensitivity or review boundary.
• Prior context and relevant decisions.
• Draft follow-up status.
• Review owner.
• Completion evidence.

Research on transactive memory systems is useful here. Ren and Argote describe how groups encode, store, retrieve, and communicate knowledge across members, including knowledge of who knows what. Lewis's work on transactive memory measures specialisation, credibility, and coordination. Partner-led practices already operate this way socially. Proximity makes parts of that memory more durable and reviewable.

What The System Prepares

Before the weekly review, Proximity could prepare a relationship queue.

Each item might include:

• The commitment.
• Where it was captured.
• Why it matters.
• The owner and next date.
• Relevant prior context.
• Draft follow-up language for review.
• Review boundary, if partner approval is needed.
• Missing context or uncertainty.

The system could also prepare partner call briefs:

• Recent interactions.
• Open commitments.
• Sensitive context.
• Project status.
• Documents or decisions likely to come up.
• Suggested internal questions before the call.

These briefs should not claim to know what the partner should decide. They should make it easier for the partner to enter the conversation prepared.

Anthropic's guidance on effective agents distinguishes workflows from more autonomous agents. Relationship-led work should start with narrow workflows: extract commitments, prepare briefs, surface stale promises, and route sensitive drafts for review. Broad autonomy is the wrong default because the value of the work is human trust.

What Remains Human

Relationship judgment remains human.

Proximity must not decide relationship strategy, rank the value of a client, infer sentiment as fact, send sensitive messages, choose tone, make promises, or decide what a partner should say. It can prepare context and draft options for review, but people own the relationship.

Humans remain responsible for:

• Relationship judgment.
• Client strategy.
• Trust and tone.
• Final messages.
• Commercial decisions.
• Promises and commitments.
• Escalation decisions.
• Professional accountability.

The system should make this boundary visible. A draft is a draft. A suggested follow-up is not an instruction. Sentiment, if used at all, should be treated as a weak signal for human interpretation, not a fact.

Pilot Shape

A practical pilot would start with one partner group and one review cadence.

The first phase would map where commitments are currently made and lost. The second phase would create a commitment queue from approved meeting notes and emails. The third phase would prepare weekly partner briefs and track whether follow-ups were completed.

Success signals include:

• Fewer missed promises.
• Faster preparation for client calls.
• Better handover when team members change.
• Clearer ownership of follow-up.
• More sensitive items routed to partner review.
• Partners reporting that briefs capture useful context without pretending to decide for them.