Automated Onboarding for Doctors: A Practice Manager's AI Checklist

A practice manager can onboard a new doctor with an AI-built checklist by splitting the job into gates (contract, registration and indemnity, payer enrolment, system access, induction, 90-day review), giving each gate an owner, a due date and evidence, and letting AI track and chase while a named person signs off. AI manages the list; humans verify the facts.

The reason to split it into gates is that onboarding a doctor fails in the gaps between departments. The contract is signed but nobody has checked indemnity; the doctor has system logins but no clinic template; the first patients are booked before a payer has confirmed enrolment. A gate list makes each dependency visible, and an AI can keep that list honest every Monday morning.

Follow me on Instagram@sagnikteaches

Why a gate list beats a flat to-do list

A flat checklist with thirty items looks complete and still lets the wrong thing happen first. A gate list adds a rule: nothing after a gate starts until a person has signed it. That rule is the safeguard, and the AI is just the clerk that applies it.

Connect on LinkedInSagnik Bhattacharya

The gates in this tutorial run in this order, and each one earns its place:

Subscribe on YouTube@sagnikteaches

1. Gate 1: contract and identity. Signed terms, verified identity and right to work.
2. Gate 2: registration and indemnity. Licence or registration checked at the issuer, indemnity cover confirmed in writing.
3. Gate 3: payer enrolment. Applications to each insurer or funder, tracked to confirmation.
4. Gate 4: system access. Records system, phones, email, prescribing, imaging, with the least access that does the job.
5. Gate 5: induction. Clinical protocols, safety, local ways of working, shadowing.
6. Gate 6: 90-day review. A fixed conversation, not an ad hoc chat.

If your practice already has an onboarding process for non-clinical staff, planning a new starter's first two weeks with AI covers the general shape. This tutorial is the clinical version, where some gates carry patient-safety consequences.

What AI does at each gate, and what it must never do

Be explicit about the split, because the failure mode is a tidy tracker that nobody has checked against reality.

One rule holds across all six: the AI never sees identity scans, indemnity certificates or registration numbers if you can avoid it. Give it the tracker (document type, expiry date, status, owner), and keep the documents in the system your practice already trusts for them. If a tool is approved for document reading, an approval process for AI tools is how you record that decision.

Build the tracker first: the fields that matter

Ask the AI for a spreadsheet design, then trim it. This prompt works in a ChatGPT or Claude project (a project keeps the instructions across sessions):

I am a practice manager in a small multi-doctor clinic. Design a tracker for onboarding one new doctor.

One row per task. Columns: gate (1-6), task, owner (role, not name), evidence required (what proves it is done), lead time in working days, depends on (task ID), due date (calculated backwards from the first clinic date), status (not started / waiting / done / blocked), signed off by, date signed off.

Create the rows for these gates: contract and identity; registration and indemnity; payer enrolment; system access; induction; 90-day review.

Rules: no row is "done" without evidence; a gate's last row is always "Gate sign-off" owned by the practice manager; flag any task whose due date is already in the past relative to today's date.

Output as a table I can paste into a spreadsheet. Do not invent legal or regulatory requirements; where a requirement varies by regulator or payer, write "CHECK WITH ISSUER".

The "CHECK WITH ISSUER" instruction matters. Without it, AI tools fill gaps with plausible-sounding requirements, and a plausible requirement that is wrong is worse than a blank.

A filled-in extract, to show the shape

Illustrative extract for a doctor starting clinic on the 1st of September, with the tracker built backwards from that date:

Notice what the AI can do with this. Task 3.2 is blocked, and the first clinic is a week away. A weekly prompt ("List every task due within 10 days or blocked, grouped by owner, with one sentence on the risk to the first clinic date") produces the Monday email that the practice manager would otherwise compile by hand.

Gate 1 and 2: contract, identity, registration and indemnity

Treat these as the hard gate. Until both are signed, the doctor does not see patients, even supervised, and a start date is conditional on them.

What to verify, and where

• Identity and right to work: follow your own legal process for the check. The AI's job is to list what you collect and when it expires, not to assess documents.
• Registration or licence: look it up on the regulator's public register and record the date and result. A doctor's own word or a certificate copy is not a substitute for the register.
• Indemnity cover: read the certificate for the scope (private work, locum work, procedures you offer), the dates and any exclusions. An AI summary helps a second reader, but the person signing off should read the original.
• References and previous employment: track requested, received and followed up, with the date of any gap that needs a conversation.

A chaser that does not sound like a robot

The AI is good at the slow admin: chasing a referee, a certificate, a form. A sample prompt and a realistic sample output:

Draft a polite chaser to a doctor who has not yet sent their indemnity certificate. Start date is 1 September; I need it by 15 August. Two earlier requests were sent on 22 July and 1 August. Keep it under 90 words, no pressure language, say what happens if it is late (the first clinic may need to move).

Illustrative output: "Hi [first name], a quick reminder that I still need your current indemnity certificate before 15 August so we can complete the checks for your 1 September start. I know it is a busy time. If it would help, send whatever your provider can issue now and we can add the final version later. If the checks cannot be completed in time, your first clinic may need to move, and I would rather avoid that. Thanks, [your name]."

What you would fix: the output offers to accept "whatever your provider can issue now", which you might not allow. Check it against your own policy before sending. Never let a chaser promise something the sign-off rules do not.

Gate 3: payer enrolment without losing track

Enrolment is where onboarding schedules go to die, because the waiting time belongs to someone else and nobody owns the follow-up. The rule: one row per payer, one owner, one next action with a date. The AI's value is keeping the dated next action visible. For the renewals that come after, tracking payer renewals with AI uses the same table and extends it past the first application.

Each payer publishes its own process and turnaround, and they differ. Do not assume a lead time. Ask the payer, write what they say into the tracker with the date you asked, and let the AI work backwards from it. If a payer cannot confirm before the first clinic, decide in advance what the doctor does in the meantime: see other patients, shadow, or start later. Write that decision in the tracker so it is not made in a panic on the day.

A quick sum shows why the order matters. Illustrative figures: a payer tells you on the phone that enrolment takes 45 working days from a complete application. The application goes in on 10 July, so confirmation is due around 11 September, well past a 1 September first clinic. Now the decision is visible seven weeks early: book the doctor's first two weeks with patients from other payers, or move the start to 15 September. Had the form gone in on 19 June instead, the same 45 days would have landed on 21 August, with more than a week to spare. Ask the AI to recalculate every payer row whenever an application date or a quoted turnaround changes, and to show the clinic date each one blocks.

A before and after of the weekly status email

Before: the practice manager reads three inboxes and a spreadsheet and writes, "Waiting on payer A, I think the form went in." After: the AI summarises the tracker into "Payer A: submitted 10 July, no confirmation, last chased 4 August, next chase 11 August, risk: blocks all Payer A patients for the first clinic." Same facts, but now the next action has a date and an owner.

Gate 4: system access with least privilege

Doctors usually need the records system, scheduling, messaging, prescribing and imaging access, and each carries different risk. Ask the AI to draft an access matrix from your role descriptions, then review it with whoever runs IT and with your information-governance lead. The principle is least access that lets the doctor do the job from day one, with extras added on request.

Here is the role description for a salaried GP [paste]. List every system access request that role normally needs in a small clinic, grouped as: needed on day one, needed within 30 days, only on request. For each, say what could go wrong if access is too broad. Do not assume which software we use; use generic names.

Illustrative output (trimmed): Day one: records (read and write, own patients), scheduling, secure messaging, prescribing. Within 30 days: reports and audit views. On request: billing and payroll data, admin settings. The things to fix: the AI listed "audit views" as 30-day access; you may want it on day one if your induction includes an audit task.

Then test. A login that exists but has the wrong template, the wrong default clinic or no printer is not "ready". Ask the IT contact to run the doctor's first morning in the test environment: log in, open a patient record, write a note, send a message, print a prescription. Record the result as evidence for Gate 4.

If the doctor will use an AI scribe or dictation tool, treat that as its own access decision, with the same approval record as any other system. Dictation or ambient notes helps you decide which style to approve before day one.

Gate 5: build an induction plan the doctor can actually follow

An induction that is a folder of policies is a reading assignment, not an induction. Ask the AI to convert your documents into a day-by-day plan, then keep the human parts human.

A one-week plan for a new doctor, illustrative

• Day 1: building tour, safety and emergency procedures, system logins tested, meet the clinical lead, shadow a full clinic.
• Day 2: shadow a second doctor, review referral routes and results handling, read the top ten protocols (the ones the clinic uses daily).
• Day 3: see a short list of patients with a supervisor in the room, review notes together at lunch.
• Day 4: half a normal list with a named buddy on call.
• Day 5: review of the week: what was confusing, which protocols were hard to find, what to add to the plan.

Give the AI your own policy documents, not the open web, and ask for quiz questions that test whether the doctor knows where to find things (the emergency drug location, the escalation phone tree, the results-handling policy). Fix the wrong ones: AI-generated questions often test trivia, not the five things that matter on day one. For a way to turn induction topics into a skills grid, see building a staff training matrix with AI.

A realistic mistake: the induction that skipped the buddy

A clinic generated an excellent induction timetable and gave the new doctor a folder. On day three the doctor could not find where to send an urgent referral, because the AI-generated plan listed policies but did not name a person to ask. The fix is a "Who to ask" column for every induction item, filled in with a role and, on the day, a name. A buddy is part of the checklist, not an extra.

Gate 6: the 90-day review as a fixed agenda

Put the review in the calendar at the offer stage, not at day 80. A fixed agenda keeps it from becoming a chat. Ask the AI to build a one-page pack from facts you provide:

1. Clinical workload and appointment mix against what was agreed.
2. Documentation quality, from a sample of notes reviewed by a senior colleague.
3. Patient feedback and any complaints, with outcomes.
4. System problems logged and resolved.
5. Induction gaps: what the doctor wished they had known earlier.
6. Support and supervision going forward, and whether any restrictions on practice can lift.

The AI should assemble the page from your records; it should not draft the assessment. Keep the judgement human. After the review, add the doctor's answer to "What was missing from induction?" into next time's plan. After three doctors, your onboarding template has been tested by three real starts, and it is better than any generic list.

One checklist, three starts: locum, sessional and returning

One tracker can serve several kinds of start if you keep a core and switch rows on or off. Ask the AI to produce the variants from your master list, then check each against what your policies really require.

The locum who starts on Monday

A locum often has days, not weeks. Keep Gates 1 and 2 in full (identity, registration, indemnity), cut Gate 3 if the locum works under your practice's arrangements and the payer rules allow it (ask the payer, do not assume), shrink Gate 4 to the minimum access for the assignment, and replace Gate 5 with a one-page "first morning" sheet: where the emergency kit is, how to escalate, how results are handled, who to ask. Gate 6 becomes a ten-minute debrief on the last day. The AI can draft the one-page sheet from your protocols in a minute; the practice manager checks every phone number on it.

The sessional doctor with two clinics a week

The risk here is that induction takes a fraction of the time and the doctor meets only half the team. Stretch Gate 5 over the first six clinics instead of five days, and add a "second-week check-in" row. Access should still be least-privilege, but make sure the doctor can reach the same protocols as a full-timer, because the gap shows up on the days nobody senior is in.

The returning doctor after a long absence

Registration and indemnity gates matter more, because both may have lapsed or changed. Add a row for a supervised re-entry plan agreed with the medical director, a refreshed system induction (the software has probably changed) and a shorter first review at 30 days. Keep the plan itself human-written; the AI only formats it.

When the tracker is trusted too much

A tidy tracker invites false confidence. Three failure patterns to watch for:

• Status without evidence. Someone marks "indemnity: done" after a phone call. The tracker needs an evidence column and a rule that no evidence means no tick.
• Stale dates. The AI computed due dates once, from the original start date, and the start moved. Recalculate whenever the date changes, and say so in the weekly prompt.
• Invented requirements. The AI adds a "mandatory training certificate" that your regulator does not require, or omits one that it does. Every requirement row should trace to a policy, a regulator page or a payer instruction you have read yourself.

An easy audit: once a quarter, pick one completed onboarding at random and try to reproduce each sign-off from the evidence alone. Whatever you cannot reproduce is a hole in the process, not in the doctor.

Running it weekly: the one-prompt Monday routine

The routine that turns this from a document into a process is a single prompt, run every Monday from the start of the offer period until the first clinic. Paste the current tracker and use:

Today is [date]. Here is the onboarding tracker for [role] starting on [date]. List: (1) overdue tasks, (2) tasks due in the next 10 days, (3) blocked tasks and what each blocks, (4) any gate that cannot be signed off before the first clinic at the current pace. For each, name the owner and suggest one next action. Do not mark anything done.

The last sentence ("do not mark anything done") is the guard against an AI that wants to be helpful. Status changes come from people with evidence in hand.

Quick checks that the whole onboarding worked

• Every gate has a signature and a date from a named person, not "approved" in a chat.
• No task is "done" without evidence you could show an auditor or a regulator.
• The doctor completed a full first morning in the test environment before the real one.
• The first week of clinics ran without a blocked payer, missing login or unknown escalation route. If one happened, it is already a new row in the template.
• You can name what AI did (tracking, chasing, drafting) and what people did (verification, sign-off). If anyone asks, the split is on one page.

For help placing this inside a wider set of clinic admin jobs, ten admin tasks a small clinic can hand to AI shows where onboarding sits.

Finally, keep the template as a living file. After each doctor starts, add the lessons, retire the steps nobody needed and note the lead times you actually saw. The point of automating onboarding is not a clever workflow; it is a process that gets slightly better with each hire, and that a colleague could run while you are on leave.

Questions practice managers ask about onboarding with AI

Can AI do the credentialing itself?

No. AI can build the checklist, track due dates and draft chasers, but verifying a licence or indemnity cover is a named person's job against the issuer's own record. Treat the AI as the project manager and a human as the signatory at every gate.

Should I upload a new doctor's ID and certificates to ChatGPT?

Not into a general chat. Keep scans in your document system and give the AI only the tracker fields: document type, expiry date, status and owner. If you do want AI to read documents, use a tool your data-protection lead has approved for that.

How early should onboarding start?

Start at offer acceptance, not on day one. Payer enrolment and some access requests have waiting periods outside your control, so build the tracker backwards from the planned first clinic and flag any gate whose lead time already exceeds the gap.

What should the 90-day review cover?

Cover clinical workload and patient feedback, documentation quality, system problems, whether induction answered the doctor's real questions, and any gaps in cover or supervision. Use the same page for every doctor so you can compare onboarding quality over time.

Further reads

• How to Onboard New Hires Onto Your AI Tools and Rules — Brief new starters on your AI tools and rules.
• Subcontractor Management: Track Insurance Expiry and Onboarding With AI — Same expiry-tracking idea applied to insurance certificates.
• AI Scribe for GPs: Costs, Free Tiers and the NHS Pilot Limit — Costs to weigh before a new doctor asks for a scribe.
• How a Small Clinic Can Use AI at the Front Desk Safely — Front-desk AI a new doctor will meet on day one.
• Therapist Credentialing: Using AI to Track Applications and Renewals — A parallel tracker design for therapists' applications.
• AI Scribe Security: Audio, Retention and BAA Vendor Questions — Vendor questions before a new doctor gets a scribe login.
• How to Set Up Human Review for AI Work Without Slowing Down — The sign-off habit behind every gate, in general terms.
• Using AI for HR: Probation, Renewal and Right-to-Work Deadline Checklist — Reuse the tracker for probation and renewal deadlines.
• AI in Healthcare Staffing: What a Small Agency Can Automate — Credential expiry chasers, shift-fill messages and timesheet checks a small healthcare staffing agency can automate, and the sign-offs that stay human.
• CQC and GP Practices: Build an Evidence Folder With AI — Organise a GP practice's CQC evidence by key question and quality statement, with an AI-built index, policy review log and mock-inspection summary.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general onboarding practice for clinics; OpenAI's business data page and Anthropic's privacy help page on model training (read 4 November 2026). Verify every credentialing requirement with the issuing body or payer.