Welcome back.
We’re putting together the AI Ready Network for leaders, figuring out how to make AI useful in their organizations. I want it to be a place where you can bring questions, compare notes, and learn from what others are trying.
It’s free for leaders who are accepted.
Today, I’m sharing an exercise from my recent DHNY masterclass to help you find where an AI workflow breaks down.
What we worked through at DHNY
Earlier this month, I led an AI masterclass at the DHNY Summit in New York.
I worked through a question I hear often in AI adoption convos nowadays:
How do you know whether an AI initiative is improving the result your business cares about?
Discussing how to put AI to work at DHNY Summit.
Where the request gets stuck
We used a made-up patient-scheduling example to look at the whole workflow.
Here’s how it goes: Imagine a patient calling a specialty practice for her first orthopedic appointment. She wants a morning appointment at the downtown office. This practice requires a referral from another doctor before booking, but it hasn’t arrived.
The AI phone assistant checks for the referral, explains the rule, records her preferences, and says, “Our team will follow up.”
If you were to review the call alone, you’d probably score it well. The assistant understood the request and gave the right answer.
Following the request, however, reads a different story.
It lands in a shared list of requests marked “awaiting referral.” No one is assigned to follow up or given a deadline to check on it.
On day three, the referral arrives in a separate inbox, but nothing connects it to the waiting request.
On day six, the patient calls again, and she still doesn’t have an appointment.
The AI handled the conversation correctly. The handoff to the staff arranging the appointment is where I can see that the process broke down.
How to repair the handoff
First, I’d check what is actually holding up appointments.
Are the appropriate slots full? Are cancelled slots going unused? Or are referrals getting stuck?
Because these answers all lead to different projects. For this exercise, assume slots are available, and the referral wait is the problem.
I’d use a six-part checklist to map how the request should move from that first call to a confirmed appointment. Here’s what I’d write for this example:
The goal: An appropriate appointment is confirmed.
Where the request waits: The request and the patient’s preferences stay together in a shared list while the referral is missing.
Who follows up, and when: A named referral coordinator owns the request, with a follow-up deadline set by the practice.
What gets things moving again: The incoming referral is matched to the request and verified. Scheduling resumes with the patient’s original preferences.
How we confirm it’s done: The scheduling system confirms the booking before the patient is told it’s done.
What happens if something goes wrong: A missing referral prompts follow-up at the deadline. Conflicting records go to a person. If a booking attempt returns no confirmation, staff check whether it went through before trying again.
The fix might be as simple as assigning someone to follow up and setting a rule in the practice’s existing software to flag overdue requests.
But in the case that the software can’t connect an incoming referral to the waiting request, I’d ask a vendor to demonstrate the full process, including when a person needs to take over.
What to measure
A correctly handled call, a confirmed booking, and a patient attending an appointment are three different results. And this is how I’d measure each one:
The interaction: Was the call handled correctly?
The workflow: Were waiting requests resolved and appointments confirmed without repeat calls?
The business result: Did more patients attend appropriate appointments, at a cost that made sense?
If call quality improves but patients still have to call back to get booked, we haven’t fixed the workflow.
Try this with one AI-assisted request in your own organization that needed follow-up, such as a customer support question or a sales inquiry. Work through the checklist with the people handling it, then compare your answers with what actually happened. The goal is simply to look for a step where nobody knew what to do next.
Free for accepted leaders. We review every application.
As always, thanks for reading.
Haroon
Founder, AI Ready