Skip to content
Expedify
Start free
Expedify in an Afternoon
Course contents

Module · The nineteen workflows

AI on the phone: qualifying, and writing up the call

Lesson 19 of 40 · 6 min

Two workflows, and they are deliberately not one. The first has a conversation. The second turns a conversation into a record. Keeping them apart is what makes the second one worth ten times more than it looks.

The call: an agent with a script it is allowed to depart from

3.2 · Let AI call the lead and qualify them

Look up who to ring, place the call, raise a follow-up task for a human. The voice agent hangs off the call node — with the rubric and the write-back as its own tools.

Scroll for all 7 steps →

Read the business-hours setting before anything else. A loan call at nine at night does not annoy a lead, it loses the case — so the node carries a window, a working-week and a timezone, and holds the call rather than dropping it. That field is the difference between an automation a client will keep and one they switch off in week two.

What the caller is told it is

qualifier — ## Role
## Role
You are calling on behalf of Anvaya Finserv, a loan and protection advisory in Delhi
NCR. You are not an advisor and you are not a salesperson — you are the first
conversation, and your job is to understand the case.
The voice agent's real Role section.

“You are not an advisor and you are not a salesperson.” The agent is scoped to the first conversation — understand the case, hand it on. That single sentence is doing regulatory work: an advisory that lets a model advise has a problem, and this one does not.

The write-up, and the reason it is a separate workflow

3.3 · Summarise the call and move the deal on

It does not watch the AI caller. It watches for a finished transcript — and then writes the note and the next step.

Scroll for all 4 steps →

Look at what it triggers on: a call record whose transcription has finished. Not “the AI caller has run”. So it writes up every call that produces a transcript — including the ones your advisors made themselves. The AI calling is the optional half; the writing-up is the half that pays for itself on day one.

That is the sentence to use in a demo. Most prospects are not ready to let a machine phone their customers. Almost all of them have advisors who do not write up calls properly, and this fixes that without changing who makes the call.

Two agents, two jobs

Runs

The caller
during the call, out loud
The writer-up
after it, on the transcript

Reads

The caller
the scoring rubric, as a tool
The writer-up
the transcript, from the trigger

Writes

The caller
the answers onto the enquiry
The writer-up
a note, and a task with the next step

If it is wrong

The caller
a person hears it happen
The writer-up
a person reads it on the record

Neither of them decides anything about the customer. One gathers, one records. Every judgement in this pipeline is still made by the rubric in Scoring a lead against a written rubric or by an advisor.

Why this one is a draft, and what to say about it

The calling workflow cannot run in your demo org, and that is not a gap in the build. It waits on one thing only: the client's own telephony number. Everything else — the agent, the rubric, the window, the write-back, the follow-up task — is finished and readable. A draft here means “configured and waiting on something only you can supply”, and saying that plainly is more convincing than a demo that dodges it.

Change one thing

  1. Open the caller's prompt and read its Objective. Then narrow it — take one question out of the call.
  2. Open the business-hours field and shift the window. Note that the setting is a configuration, not code, and a client can hold their own opinion about it.
  3. Open the write-up workflow's trigger and read its filter. Say out loud what would happen if the filter were removed.

Next: Answering questions on three channels — the same voice machinery pointed at people who enquired months ago.

Sign in to track progress