I wrote ACA business as a personal producer for years and pushed my own book past a million a year in production. I can tell you exactly where the money leaked, because it leaked the same way every single season: leads I paid for went cold while I was on the phone with someone else.
Not because the leads were bad. Not because a competitor had a better plan. Because follow-up is a mechanical job and I was trying to do it with willpower and a notepad.
This is the architecture that fixed it.
Key takeaways
- Follow-up fails for mechanical reasons — timing and attempt count — not for lack of sales skill.
- Split the job: the AI handles contact, qualification, and scheduling. The licensed agent handles advice and the sale.
- Build it in four stages over a week. Building all four at once is why most agents quit.
- Consent, disclosure, and opt-out are not optional, and the licensed agent is still liable for the outcome.
The four jobs hiding inside “follow-up”
When agents say follow-up is broken, they usually mean one specific stage is broken. Naming the four separately is what makes the problem fixable.
| Stage | What it involves | Owner |
|---|---|---|
| Contact | Reaching a human at all — speed and persistence | AI |
| Qualification | Confirming basic facts and real interest | AI |
| Scheduling | Booking against actual availability | AI |
| Advice and enrollment | Plan comparison, recommendation, application | Licensed agent |
The line you do not cross
The AI never quotes a plan, compares benefits, or recommends coverage. Those are licensed activities. Using AI does not transfer liability — if the tool produces something wrong and you act on it, you own it. Keep a human in the loop for anything that constitutes advice or a sale.
What changes when you split the job
One person doing all four jobs
- First touch happens whenever you are free
- After-hours and weekend leads go untouched
- Attempts stop at two, then leads age out
- Notes live in your head, not in the CRM
- Growth means hiring another caller
40 dialsto get four real conversations
AI contacts, agent advises
- Every lead touched within a minute of arriving
- Coverage runs nights and weekends
- Full cadence runs whether or not anyone remembers
- Transcript and disposition sync automatically
- Growth is a settings change, not a hire
4 conversationswith the dialing already done
The build, in the order that actually works
Stage 1 — Connect the source and automate the first touch
Do nothing else until a new lead triggers an outbound attempt within sixty seconds. Leave your existing sequence completely alone. You are adding a first touch in front of it, not replacing anything.
Measure before and after on one source
Pick a single lead source. Record contact rate for two weeks before, then two weeks after. Change one thing at a time or you will never know what worked — and you will not be able to defend the spend.
Stage 2 — Extend the cadence
Once the first touch is automatic, the marginal cost of attempts three through eight is nearly zero. Space them across mornings, midday, and evenings, and mix call with text.
Stage 3 — Add qualification
Now let the AI confirm the basics before it hands off: who it is for, rough household size, whether they are currently covered, and whether they want to talk to a licensed agent. This is where agent time stops being wasted on unqualified conversations.
Stage 4 — Add booking and warm transfer
If a licensed agent is available, transfer live. If not, book against real availability. Never promise a callback that lands in a queue nobody works.
Instant trigger
The dial fires off the form post itself, not a batch job on a fifteen-minute timer.
Identity and disclosure
Every contact opens by identifying the agency and the reason for the call.
Consent trail
Consent, disclosure, and opt-out are logged on the contact record, not remembered.
Round-robin transfer
Ring the available team, not one person, so a busy agent never kills the handoff.
Calendar fallback
No pickup means a booked appointment, not a task in someone's queue.
CRM sync
Transcript, tags, and disposition land on the record automatically.
Where the effort goes, before and after
Share of an agent's follow-up hours, illustrative.
Directional. The point is the reallocation of hours, not the exact split. Results may vary.
Governance is becoming table stakes
The NAIC model bulletin on insurer use of AI has been adopted in a majority of states, and its expectations — written policies, human oversight, documentation, vendor accountability — flow downstream to agencies. Write down what your AI does, who reviews it, and how you handle errors. You will need it eventually; having it early costs almost nothing.
Follow-up is a mechanical job. Stop trying to beat it with willpower.
Mike MooreBefore you switch anything on
Tech Savvy Insurance is a training and software community — not an insurance company, agency, or law firm — and does not provide insurance, legal, tax, or compliance advice. You are responsible for your own licensure and for complying with all applicable CMS, HIPAA, state, and carrier regulations. AI-generated outputs may contain errors — always verify before use. Results may vary.
How Ambrose builds this specific system
Everything above works with any tools you assemble yourself. Inside Ambrose, the platform included with a Tech Savvy membership, the four stages map onto specific parts of the same tenant instead of four separate subscriptions. The channel-bridge spoke dispatches outbound email, SMS, and Telegram from one place, and reply-bot drafts replies and books appointments directly — that’s stages 1 and 4 without stitching a dialer to a separate scheduling tool (Ambrose docs, Spokes). The lead-memory spoke keeps a per-lead dossier so qualification notes from the first contact are already there when the licensed agent picks up the transfer. For the actual calling and texting infrastructure, Ambrose integrates with VAPI and Retell for voice AI and Twilio for SMS, or with GoHighLevel’s own Voice AI if that’s already your CRM (Ambrose docs, Integrations).
Stage 2 — extending the cadence to eight-plus touches over ten days — is where a routine does the part a human would otherwise have to remember. A routine is a scheduled prompt attached to an agent or team, and one of the documented example patterns is a weekly renewal or outreach scan that runs automatically and posts the result to Slack (Ambrose docs, Routines). You write the cadence once; it runs whether or not anyone remembers to check.
None of this changes who does the advising. The AI still never quotes a plan or recommends coverage — that discipline is the same whether you build the four stages by hand or inside Ambrose.
Start Monday
Pick one lead source. Put an instant first touch in front of your existing sequence. Measure contact rate for two weeks. Then extend the cadence, add qualification, and add booking — in that order.
That is the whole system. The part that requires a licensed human stays with the licensed human. Everything else was always mechanical, and mechanical work is exactly what machines are for. If you’d rather build it inside one tenant instead of stitching together a dialer, a texting tool, and a scheduler, one Ambrose seat is included with a Tech Savvy membership, and the weekly build-with-you calls are where agents actually wire this up on their own book: https://techsavvyinsurance.com/.
Frequently asked questions
Sources
- FCC — Telemarketing and robocall rules — fcc.gov
- CMS — Medicare communications and marketing guidelines — cms.gov
- NAIC — Model bulletin on the use of artificial intelligence systems by insurers — content.naic.org
- Ambrose docs — Spokes — app.hiambrose.com
- Ambrose docs — Integrations — app.hiambrose.com
- Ambrose docs — Routines — app.hiambrose.com
Ready to put this into practice?
Join a private community of Health & Life insurance professionals using AI, Meta Ads, and automation to grow — without draining their bank account.
Join Tech Savvy — $97/month