87% of failed AI projects in clinics make the mistake in the first 15 days, before they’ve turned anything on.
That sentence sums up what we see again and again with chiropractic, physio and osteopathy clinics trying to roll out an AI assistant. The tech works. Vendors are reasonably solid. What breaks the project is the calendar: clinics expecting something live in a week, or assuming the system tunes itself without three rounds of iteration.
Retell AI (2025) reports that a well-designed voice-AI pilot in healthcare needs a shadow mode period of 4 to 8 weeks to verify production accuracy, identify workflow friction, and let the clinical team calibrate trust. Clinics that skip it take 13.5 months to reach positive ROI. Clinics that respect it, 7.5 months. Almost half.
This is the 60-day plan we use when rolling out HeyCAi, split into four phases with their milestones, common mistakes and the KPIs worth watching before each jump.
Days 1 to 15: discovery and setup
The first two weeks go into understanding what you have and what you want the system to do. Turning anything on comes later.
Start by auditing the current call and WhatsApp flow. How many calls come in daily, how many are missed, what kind of request is most common (booking, cancellation, availability check, pricing question), what share arrives after hours. Without this baseline you can’t measure improvement later.
Then comes documentation. A complete list of services with their duration and price. Working hours per practitioner. Cancellation policy. The FAQs your receptionist answers ten times a day. Edge cases (urgencies, elderly patients, calls in another language). This usually takes three to five days if the clinic has never done it.
In parallel, your technical lead signs the data processing agreement with the provider, configures the integration with your clinic software and decides phone routing. If you use QuiroHiro, our own clinic management software, this part is essentially done out of the box. For PracticeHub, Cliniko, Jane App and others, the connection takes 1 to 2 weeks.
The most common mistake in this phase is skipping FAQ documentation because it seems obvious. It isn’t. The assistant will run into patterns your receptionist solves by context without thinking, and each of those contexts has to be taught.
Days 16 to 30: training and shadow mode
This is where the nerves kick in and where the biggest impact on the final result lives.
Shadow mode means HeyCAi listens to real calls and reads real WhatsApp messages but doesn’t respond. It generates transcripts, predictions of what it would have said, what it would have booked, what it would have transferred. The reception team reviews those predictions and flags them as correct, incorrect, or “the assistant would have handled this better than I did”.
Three to five days of shadow mode produce more information than two months of sandbox testing. Clinics that skip this phase have 70% more complaints in the first month of production (Retell AI 2025).
In parallel, the provider trains the model with the documentation from phase 1. Specific scenarios get tested: patient who changes their mind mid-call, patient who asks for an appointment with a specific physiotherapist, patient with acute pain who needs urgent escalation. Each edge case documented, each tested.
By the end of week 4, the clinic team should be able to confidently answer these questions:
- What types of appointments can the assistant close without supervision?
- What cases does it transfer to a human and how?
- What does it do if the patient mentions something medically sensitive?
If any of those answers isn’t clear, the project doesn’t move to the next phase.
Days 31 to 45: limited production (after-hours and overflow)
The transition to production works like a tap opening gradually, not like a binary switch.
The first phase of production covers only two types of calls. After-hours calls, where the cost of a mistake is low (worst case, the patient calls back tomorrow). And overflow calls, when reception is busy with another patient and the call would have rolled to voicemail anyway.
During these two weeks, someone reviews transcripts every day. Not all of them, a 20% sample. What you’re looking for is failure patterns, not individual failures. If the assistant confuses “physiotherapist” with “physio” in 15% of calls, retraining is needed. If it gets a local street name wrong, also. If it transfers too quickly, the confidence threshold needs adjustment.
At the end of week 6, there’s a KPI review with the team:
- Resolution rate without transfer: should be above 60%
- Average call time: ideally under 3 minutes
- Call-to-confirmed-appointment conversion: target 70% to 80% after hours
If the numbers look good, you move forward. If not, you stay another week in limited production.
Days 46 to 60: full production with human escalation
The final phase opens HeyCAi to all incoming calls and messages, always keeping a human escalation channel for when the AI doesn’t understand or when the patient explicitly asks for one.
The most important shift in this phase is cultural, not technical. Your receptionist moves from answering calls to supervising the assistant and handling the complex cases that get transferred. That’s what generates the real savings: a receptionist who used to answer 80 calls a day now handles 20 and dedicates the other 7 hours to work that creates value (patient follow-up, coordination with practitioners, collections, billing).
This phase also activates advanced features: automated WhatsApp reminders, outbound calls to confirm appointments, no-show prediction. Each one shows up in the numbers within days.
The 60-day success indicator is simple. If patients are happy and the team doesn’t want to go back to the old system, the project is ready. If complaints persist, or the team still answers calls the assistant should resolve, two or three more weeks of tuning are needed.
Common mistakes that stretch the project from 60 days to 4 months
Projects that drag on usually fail because of misaligned organizational decisions, rarely because of technology.
No internal owner. If nobody in the clinic owns the project with dedicated time to review transcripts and tune the system, the provider ends up working alone and the result doesn’t match operational reality.
Skipping shadow mode out of urgency. The pressure of “we’re already paying, let’s turn it on” kills more implementations than any technical limitation. The first bad patient experiences are the most expensive to reverse.
Not measuring from day one. Without a baseline, you can’t show improvement, which hurts internal buy-in and cost justification. A clinic that measures well typically reaches positive return in 7.5 months. One that doesn’t, in 13.5 months. Almost double.
Realistic project cost over 60 days
Specific figures depend on the provider and clinic size, but there’s a reasonable range.
HeyCAi for a clinic with 200 to 500 calls per month runs at €229/month (Clinic plan, all-inclusive). Initial setup is included; no separate fee.
If you use QuiroHiro alongside HeyCAi, setup is same-day. For PracticeHub or Cliniko, it’s 1 to 2 weeks of integration work, also included. Total 60-day cost lands between €450 and €700 depending on plan.
In return, a typical clinic that captures after-hours calls and reduces no-shows recovers €3,000 to €6,000 a month. Break-even crosses during the second month of production.
Frequently asked questions
Can I implement HeyCAi in less than 60 days?
Yes. Some clinics turn it on in 2 to 3 weeks. The difference is they skip shadow mode and iterative tuning, which translates to more patient complaints in the first months and slower return. If minimising risk matters, 60 days is the reasonable minimum. If pure speed matters and you accept iterating in production, 3 weeks is viable.
Do I need to hire someone new to lead the project?
No, but you need someone with authority. Usually the person managing reception or the clinic manager. Realistic time commitment is 2 to 4 hours weekly across the 60 days, split between provider meetings, transcript review and configuration decisions.
What if my team resists the change?
Most common scenario. The approach that works best is involving the team from the discovery phase (days 1 to 15) and positioning the assistant as a tool that takes mechanical tasks off their plate so they can focus on the work they do best. Clinics that frame AI as “we’re going to replace the receptionist” have 60% more internal resistance than clinics that frame it as “we’re going to lift the load of those 80 daily calls”.
How do I know if my clinic is ready for AI?
Three signals. First, the phone is a bottleneck (missed calls, waiting lists to pick up). Second, your clinic software has open APIs or webhooks (or you use QuiroHiro, which is integrated by default). Third, there’s at least one organised person in the clinic who can lead the project. If all three hold, 60 days works. If any fails, fix it first.
What KPIs should I look at on day 60 to decide if the project succeeded?
Four key metrics:
- Resolution rate without transfer: target above 60%
- Call-to-confirmed-appointment conversion: target above 70% after hours
- Reduction in missed calls vs phase 1 baseline: target above 80%
- Patient NPS or satisfaction: shouldn’t drop versus baseline, ideally rises
If three of the four hit target, the project is consolidated.
If you want to hear HeyCAi handling a real call before kicking off the 60-day plan, get a live demo call in 30 seconds. For deeper questions about the voice side specifically, HeyCAi Voice on callcai.ai has more detail.