The Shared Record

If you read The Mahindra Call, you know the diagnosis: the relationship with your biggest customer lives in one person's head. This is the fix — what a shared record actually is, what goes in it, and how to stand one up this week.

---

What a shared record is

One place that holds what's true about a customer relationship. What's been committed. What's open. What each side believes. That's it.

It is not a CRM. A CRM tracks deals — pipeline stages, contact info, close dates. Useful, but it answers "where is this sale," not "what have we promised these people and what do they think is happening." Plenty of shops have a CRM and still keep the actual relationship in somebody's head.

---

What goes in it

Four things, minimum:

Commitments. What you've promised them and what they've promised you, each with a date. "Samples by the 20th" is a commitment. So is "they're sending the revised print Friday."

Open questions. Anything asked and not yet answered, in either direction. The question their quality team asked three weeks ago that nobody wrote down is exactly the one that turns into a problem.

Decisions. What got decided, when, and by whom. Not the meeting notes — the outcomes. The decal moved half an inch; here's who approved it and the date.

State. Where things actually stand: open orders, current revisions, the schedule both sides are working from.

---

Where it lives

Matters less than you think. A shared spreadsheet works at first. A Notion page works. A database works when you outgrow those. The rules matter more than the tool: one place, not three. Everyone who touches the account can read it. It gets updated when things happen — not reconstructed Friday afternoon from memory.

---

How AI changes this

The reason this never got done before isn't discipline. It's capture cost. The information arrives as email, phone calls, and hallway conversations, and somebody has to convert all of that into the record. That conversion work is a job, and nobody at a 20-person shop has it.

That conversion is now exactly what AI does cheaply. My own setup runs this way: a pipeline reads the inbox, classifies what matters, extracts the commitments and open questions, and a human approves each one before it writes to the record. The approval step is the point. The machine does the converting. The judgment — what matters, what's really a warning — stays human.

---

Start this week

1. Pick one account. Your biggest.
2. Make one page with the four headings above.
3. Backfill it from the last 30 days of email. Budget an hour.
4. Set one rule: anything promised, in either direction, gets written down the same day.
5. Test it in two weeks: can someone other than the account owner answer "what do we owe them right now" without asking anyone?

If the answer to a customer's question lives in one person's head, you don't have a record. You have a hostage. Start with one account and four headings.