Two kinds of follow-up: the cadence and the written-for-one message
Everybody goes quiet eventually. One workflow answers that with eighteen nodes of certainty; the other with five nodes and a judgement. Both are right, for different reasons.
A person you spoke to last Tuesday has not replied. Nothing has gone wrong — they are busy, or waiting on a document, or quietly talking to their own bank. This is the largest single category of work in any sales operation, and your org handles it two completely different ways.
| Nodes | 18 | 5 |
| Decides anything? | No. Fixed messages, fixed days. | Yes. Writes each message from that person's file. |
| Costs | One email per send. | One model call per person, every week. |
Nodes
- 18
- 5
Decides anything?
- No. Fixed messages, fixed days.
- Yes. Writes each message from that person's file.
Costs
- One email per send.
- One model call per person, every week.
The cadence: four messages, twenty-one days
6.1 · Send a 4-message follow-up over 21 days
An enquiry reaches Contacted and a three-week rhythm begins — four mails, and a check of the record before every one of them.
Scroll for all 18 steps →
| When | What lands |
|---|---|
Day 1 | What we agreed, and what we need from you |
Day 3 | Where the twelve lenders stand on your case |
Day 7 | Is something in the way? |
Day 21 | One thing worth knowing, whenever you come back to this |
Day 1
- What lands
- What we agreed, and what we need from you
Day 3
- What lands
- Where the twelve lenders stand on your case
Day 7
- What lands
- Is something in the way?
Day 21
- What lands
- One thing worth knowing, whenever you come back to this
Read the four subjects in order. They are not four attempts at the same message — they escalate in honesty rather than in urgency, and the last one says the relationship continues after the sequence stops. Somebody wrote those, and a client will want to rewrite them, which is a text box each.
It asks the record before every send
Between each pair of messages there is a delay, then a query, then a fork. The query re-reads the enquiry's stage; the fork asks whether it is still where it was. If the answer is no, the sequence stops immediately and this happens instead:
A task: They moved — pick the conversation up
All 3 of the exit arms converge on that one box. Whichever message the person was between when they moved, the cadence stops the same way — and it stops by handing the relationship to a human rather than by simply going silent.
This is the Booking the meeting, and chasing the no-show habit again, three times in one workflow. A cadence that does not re-read is the thing everyone has been on the receiving end of: the fourth chasing email about something you already replied to. Two nodes prevent it.
The other kind: one message, written for one person
6.2 · Write a different message for every person
Every Monday morning, everyone who has gone quiet is pulled in one query, and each of them is handed to an agent that writes their message and sends it.
Scroll for all 5 steps →
Look at the send node: subject, to_email, html_content are all {{ai}}. The agent writes the subject, chooses the recipient and writes the body. The workflow only guarantees that an email is what gets sent — the same configured-tool pattern you read in Answering questions on three channels, turned up as far as it goes.
And look at what the query hands it: not just a name and an amount, but the last thing anybody logged on the file. That single column is the difference between a mail-merge and a message.
Two messages it actually wrote
These came out of your org's own agent, run against two real enquiries in it. Same prompt, same week, nothing else in common:
Subject: Loan Against Property Offer Review – Rohit Kaur
Hi Rohit,
I wanted to check in regarding the Loan Against Property offer for
₹79,02,491 we discussed at the Gurugram branch. I understand you
wanted to compare two offers before deciding, so if you'd like, I
can walk you through the details of the offer line by line to help
clarify any questions. Would you prefer a quick call or a visit to
the branch for this?
Best regards,
[Your Advisor's Name]
Anvaya Finserv, Gurugram BranchSubject: Follow-up on your Education Loan documents
Hi Pallavi,
I wanted to check in regarding your Education Loan application for
₹14,37,318. I see we have your PAN and Aadhaar, but are still
waiting on your income proof to proceed further. If it's easier, I
can arrange a home visit to collect the document at your
convenience. Would you prefer a home visit or to drop the document
at our branch?
Best,
[Advisor's Name]
Anvaya Finserv, Delhi NCRNeither of those is a template with a name slotted in. One offers to walk through an offer line by line because he is comparing offers; the other offers a home visit to collect one specific document, because that is the document the file says is missing. This is the demo people remember — and it is the one to show a client who thinks personalisation means putting a first name at the top.
Now read the last line of both
They end in a bracketed placeholder. The prompt tells the agent to sign off from the person's own advisor, and nothing in the context says who that is — so it left a hole rather than inventing a name.
The agent did exactly as it was told, and the output is still not sendable. Its constraints say not to invent anything outside the context, and it obeyed. The fault is upstream: the query does not fetch an owner, so the instruction cannot be met. Read what an agent writes before you switch it on — the failure you are looking for is rarely a refusal, it is a sentence that is almost right.
Which one to reach for
| Reach for the cadence when | Reach for the written one when |
|---|---|
| The message is the same for everybody — documents, next steps, what happens now. | The reason each person stopped is different, and it is written down somewhere. |
| The client needs to approve the exact wording in advance. | The client's files carry notes worth reading. |
| Volume is high and margins are thin. | The cases are large enough that one reply pays for a month of model calls. |
The message is the same for everybody — documents, next steps, what happens now.
- Reach for the written one when
- The reason each person stopped is different, and it is written down somewhere.
The client needs to approve the exact wording in advance.
- Reach for the written one when
- The client's files carry notes worth reading.
Volume is high and margins are thin.
- Reach for the written one when
- The cases are large enough that one reply pays for a month of model calls.
Most orgs end up running both, on different segments. That is a good answer to give a client, and a better one than choosing a side.
Change one thing — give the agent the missing field
- Open
6.2 · Write a different message for every personand find the query that pulls the people who have gone quiet. - Add the enquiry's owner to what it selects. It is one more column.
- Open the agent and add that column to its Available Context, in the same style as the lines already there.
- Save, and leave the workflow as a draft. It emails real customers when it is on.
- Run the agent node on its own and read the sign-off. That is the whole loop: find the hole, feed it, look again.
Try it
- Find the record in your org with the most interesting last note, and predict what the agent would write for them.
- Work out what the cadence does if somebody replies to the first mail but nobody updates the stage. Then decide whose fault that is.
Next: Working the old list, and the year after the sale — the same question asked of people who went quiet a year ago, and of the ones who actually signed.
Related lessons
Base rates — what a piece of evidence is actually worth
A face-recognition system that is 99.9% accurate and almost entirely wrong, and a number that sent an innocent woman to prison. Both are the same arithmetic, and it is the arithmetic that decides what any piece of evidence is worth.
ReadConfirmation and survivorship — what you never looked for
Two questions about evidence you did not go looking for. One is a rule you have to discover, and one is a pattern in five famous people — and in both, the thing that would have told you the truth is the thing nobody checks.
ReadLoss aversion, sunk cost and regression — what it costs you
Four questions you answer about yourself rather than about a scenario, and your own answers are the finding. Then the pattern that makes praise look useless and criticism look like it works, whatever you actually do.
Read
