What Changed
When we first scoped this project, we were planning for one doctor's retirement - about 1,200 patients to notify. That's still the pilot. But Stephen, you told us something that changes the entire picture:
This is a core, ongoing business operation - not a single campaign. Each retiring doctor brings 500-2,000 patients. The system needs to handle growing volume as DOCUdavit signs more doctors.
The exact volume per month depends on how quickly DOCUdavit onboards retiring doctors. We've built this analysis around several scenarios so we can plan for the right scale.
The CPSO Pipeline
Here's the business reality that drives the volume:
| Scenario | Doctors/Year | Patients/Year | Avg Calls/Month |
|---|---|---|---|
| Steady Growth | 10-20 | 10,000-40,000 | ~1,000-3,500 |
| Moderate Scale | 30-50 | 30,000-100,000 | ~2,500-8,500 |
| Full Pipeline | 50-100+ | 50,000-200,000+ | ~4,000-17,000+ |
The calls don't all have to go out in one month - campaigns can be spread across the year as doctors are signed. The pilot will reveal the actual pace.
Our Approach: Pilot First, Then Scale
We're not guessing at volume. Here's the smart path:
The Voicemail Message
The message must be generic:
"Hello, this is Sandra calling from DOCUdavit Medical Records on behalf of Dr. Smith's office. We have an important message for you. Please call us back at 416-781-9083. Thank you."
The Three-Phase Strategy
We don't pick the final platform on day one. We earn the right to make that decision by running a real campaign first.
Phase 1: PILOT
1,200 patients on Retell AI pay-as-you-go | ~$180 in calling costs | 1-2 weeks to launch
Goal: Prove the concept works before committing to a long-term platform.
- Test AI voice quality with Sandra's cloned voice on real calls
- Measure delivery rate: what percentage of patients actually hear the message?
- Measure callback rate: how many patients call back after getting the voicemail?
- Test voicemail detection accuracy: does the system correctly identify voicemail vs. live answer?
- Test different calling times: morning vs. afternoon vs. evening
- Verify Canadian number delivery with Bell, Rogers, Telus
- Identify any compliance issues before scaling
Phase 2: EVALUATE
Get enterprise quotes with real pilot data in hand | 1-2 weeks
Goal: With actual pilot results, negotiate enterprise pricing and verify compliance claims.
- Take the pilot data (delivery rates, call durations, callback rates) to Retell AI and Bland AI
- Get real enterprise quotes from both platforms based on projected volume
- Verify Bland AI's PHIPA compliance claim - ask for documentation, certification, or legal opinion
- Determine: does DOCUdavit need live AI conversations, or is voicemail-only sufficient at scale?
- Compare total cost of ownership across all three options
Phase 3: SCALE
Choose the long-term platform and build for volume | 2-4 weeks
Goal: Commit to the right platform and operationalize it for DOCUdavit's ongoing business.
The final choice depends on what we learn in Phases 1 and 2:
- Retell AI Enterprise - if their Canadian compliance story improves and pricing is competitive
- Bland AI Enterprise - if their PHIPA claim checks out and they can match on pricing
- Custom Twilio Build - if volume is high enough that the $6,000-8,000/month savings over platforms justifies the upfront build cost and ongoing maintenance
The Coverage Cascade (Still Applies at Scale)
No single method reaches every patient. At any volume, the cascade strategy applies:
Platform A: Retell AI Recommended for Pilot
What is it? Retell AI is an all-in-one AI calling platform. It handles making calls, detecting voicemail vs. live answer, playing messages, having AI conversations, cloning voices, and tracking results - all from one dashboard.
How It Works
What Happens When Someone Picks Up?
This is where Retell shines. If a patient answers, the AI can have a basic conversation:
- "Hello, this is Sandra calling from DOCUdavit Medical Records. I'm reaching out about Dr. Smith's patient records. Do you have a moment?"
- If they say yes, deliver the notification
- If they have questions, give basic info or say "Let me transfer you to our team"
- If they want to opt out, note it and end politely
Pilot Budget (1,200 Calls)
| Item | Cost | Notes |
|---|---|---|
| Ben's setup & configuration | $900 - $1,350 | 12-18 hours @ $75/hr |
| Voice cloning consent docs | Included above | Written licence for Sandra |
| Retell AI calls (1,200 x 1.5 min avg) | ~$180 | $0.07/min base + $0.03/min Canada telephony |
| Canadian phone number | $5/month | Cancel anytime |
| Total for pilot campaign | $1,085 - $1,535 + HST |
Scaled Budget Scenarios
| Monthly Volume | Pay-As-You-Go | Enterprise Rate (est.) | Notes |
|---|---|---|---|
| 5,000 calls | ~$750 | N/A (stay on PAYG) | A few doctors/month |
| 10,000 calls | ~$1,500 | ~$1,200 | Enterprise may make sense here |
| 50,000 calls | ~$7,500 | ~$6,000 | Enterprise commitment likely required |
| 100,000 calls | ~$15,000 | ~$12,000 | Negotiated rates, $3K/mo minimum |
Note: Real all-in cost with premium voice/LLM options can range from $0.13-$0.31/min, not the advertised $0.07. Enterprise estimates assume negotiated rates. Actual volume depends on how many doctors DOCUdavit signs per year.
Scale Considerations
Pros and Cons
Pay-as-you-go = perfect for pilot (no commitment)
Handles BOTH voicemail AND live AI conversations
Built-in dashboard - Stephen's team can run campaigns independently
HIPAA/BAA included on Enterprise tier (no extra charge)
Voice cloning built into the platform
Automatic voicemail detection
Canadian phone numbers available directly
Works with all phones - cell, landline, VoIP
Concurrency: 20 concurrent on pay-as-you-go, unlimited on Enterprise
Canadian compliance: working towards PHIPA best practices but NOT fully certified yet
Canadian telephony surcharge ($0.03/min) is 2x the US rate - adds up at scale
System prompts over 3,500 tokens incur additional surcharges
AI live conversations can occasionally sound unnatural - needs testing
Before any voice cloning can begin, Sandra (or whichever employee is chosen) must sign a Voice and Likeness Licence. This covers: what the voice will be used for, how long the licence lasts, what happens if she leaves, and her right to revoke consent. A cloned voice is classified as biometric personal information under Canadian privacy law (PIPEDA). No licence = no voice cloning. Ben provides the template.
Platform B: Bland AI New Addition
What is it? Bland AI is a self-hosted, end-to-end AI calling platform. Unlike Retell (which connects multiple third-party APIs behind the scenes), Bland runs their own voice AI, their own telephony, and their own infrastructure. No third-party API hops means your data stays within their system - a significant advantage for healthcare privacy.
How It Works
The workflow is similar to Retell AI - create an account, clone a voice, write scripts, upload patient lists, and launch campaigns. Bland has a visual "pathway builder" for designing call flows. The key difference is what happens under the hood: all processing stays within Bland's infrastructure.
Current Pricing (Restructured December 2025)
| Tier | Monthly Fee | Per Minute | Daily Call Limit | Concurrent Calls |
|---|---|---|---|---|
| Start (Free) | $0 | $0.14/min | 100 | 10 |
| Build | $299 | $0.12/min | 2,000 | 50 |
| Scale | $499 | $0.11/min | 5,000 | 100 |
| Enterprise | Custom | ~$0.08-0.09 (est.) | Custom | Custom |
Hidden Costs to Know About
Pilot Budget (1,200 Calls)
| Item | Cost | Notes |
|---|---|---|
| Ben's setup & configuration | $900 - $1,350 | 12-18 hours @ $75/hr |
| Monthly subscription (Build tier) | $299 - $499 | Minimum to get reasonable call limits |
| Call costs (1,200 x 1.5 min avg) | ~$198 - $216 | $0.11-0.12/min depending on tier |
| Failed call fees (est. 30% fail rate) | ~$5 | $0.015 x ~360 failed attempts |
| Total for pilot campaign | $1,300 - $1,850 + HST |
Scaled Budget Scenarios
| Monthly Volume | Est. Monthly Cost | Notes |
|---|---|---|
| 5,000 calls | ~$1,100 - $1,400 | Build/Scale tier + per-min + failed call fees |
| 10,000 calls | ~$2,150 - $2,500 | Scale tier likely needed |
| 50,000 calls | ~$6,500 - $8,000 | Enterprise tier required (5K/day limit on Scale) |
| 100,000 calls | ~$12,000 - $14,000 | Enterprise negotiated rates - needs real quote |
Scale Considerations
Pros and Cons
Only platform explicitly claiming PHIPA compliance (Ontario healthcare law)
Self-hosted architecture - no third-party API hops = strongest data privacy story
Sub-400ms response latency (fast, natural conversations)
Visual pathway builder for designing call flows
Voice cloning: 5-15 clones depending on tier
Handles both voicemail and live conversations
Monthly subscription required even during slow months ($299-499 minimum)
Failed call fees add up with volume ($0.015/attempt regardless of success)
Pricing was restructured recently - current rates are higher than older reviews suggest
Scale tier maxes at 5,000 calls/day - Enterprise needed for higher volumes
Enterprise pricing requires negotiation - no public rates
PHIPA compliance claim needs verification (ask for documentation)
Same as Platform A - Sandra must sign a Voice and Likeness Licence before any voice cloning begins. No licence = no voice cloning. Ben provides the template.
Platform C: Custom Twilio + ElevenLabs Build Most Complex
What is it? Build a custom phone system from scratch using Twilio (phone infrastructure) and ElevenLabs (AI voice). Ben writes all the code: call logic, voicemail detection, campaign management, dashboard, retry logic, audit logging - everything.
Why is it still on the table? One reason: cost at volume. A custom build is significantly cheaper per call than the platform options. The savings grow with volume - at sustained high usage, the upfront build cost pays for itself quickly.
The maintenance reality: While custom builds are cheaper at scale, they require ongoing maintenance and support that platform providers handle automatically. Retell AI and Bland AI guarantee system uptime, handle infrastructure scaling, and provide support teams. With a custom build, Ben becomes your infrastructure team - which means additional labor costs and dependency on a single developer. This is manageable but needs to be factored into the total cost of ownership.
How It Works
Twilio Pricing Details
Volume discounts are surprisingly shallow:
- First 100K minutes: $0.014/min
- Next 900K minutes: $0.0138/min (only 1.43% savings!)
- You'd need 10M+ minutes to see meaningful discounts
Pilot Budget (1,200 Calls)
| Item | Cost | Notes |
|---|---|---|
| Ben's build (full custom system) | $3,000 - $5,000 | 40-67 hours @ $75/hr. Higher than original estimate due to enterprise features needed: multi-script, campaign management, dashboard, audit logging. |
| Voice cloning consent docs | Included above | |
| Twilio calls (1,200 x 1.5 min avg) | ~$25 | $0.014/min |
| ElevenLabs + LLM + hosting | ~$50 | For pilot scale |
| Phone number | $1.15/month | |
| Total for pilot campaign | $3,075 - $5,075 + HST |
Scaled Budget Scenarios
| Monthly Volume | Est. Monthly Cost | Notes |
|---|---|---|
| 5,000 calls | ~$900 - $1,500 | Twilio + ElevenLabs + hosting + maintenance |
| 10,000 calls | ~$1,200 - $2,000 | Twilio scales linearly, maintenance stays flat |
| 50,000 calls | ~$2,500 - $4,000 | Significant savings over platforms at this level |
| 100,000 calls | ~$4,000 - $6,000 | 2-3x cheaper than platforms |
Custom build costs include Ben's ongoing maintenance (~$750-$1,500/month). The per-call cost is lowest of all options, but the upfront build ($3,000-$5,000) and ongoing maintenance dependency must be factored in.
Scale Considerations
Pros and Cons
Cheapest ongoing cost per call - savings grow with volume
Maximum control over every detail of the system
Twilio is rock-solid, industry-standard Canadian telephony
Twilio AMD (Answering Machine Detection) is mature and accurate
HIPAA-eligible (Twilio has HIPAA-eligible products)
Unlimited concurrency after business verification
No platform lock-in - DOCUdavit owns the code
Highest upfront cost ($3,000-5,000 to build)
Ben builds and maintains everything - Stephen can't self-serve without the custom dashboard
Ongoing maintenance costs (Ben's time for bug fixes, updates, new features)
If Ben is unavailable, the system doesn't get updated
Takes weeks to build vs. days to configure a platform
Compliance is your responsibility - Twilio provides the pipes, you handle the rest
Side-by-Side Comparison
Cost Comparison
| Scenario | ⭐ Retell AI | 🛡️ Bland AI | 🔧 Custom Twilio |
|---|---|---|---|
| First campaign (1,200 calls) | $1,085 - $1,535 | $1,300 - $1,850 | $3,075 - $5,075 |
| Monthly at 10K calls | ~$1,500 | ~$2,150 - $2,500 | ~$700 - $1,200 |
| Monthly at 50K calls | ~$6,000 | ~$6,500 - $8,000 | ~$2,500 - $4,000 |
| Monthly at 100K calls | ~$12,000 | ~$12,000 - $14,000 | ~$4,000 - $6,000 |
All prices CAD + HST where applicable. Enterprise rates are estimates - actual quotes may vary.
Feature Comparison
| Feature | ⭐ Retell AI | 🛡️ Bland AI | 🔧 Custom Twilio |
|---|---|---|---|
| Canadian compliance (PHIPA) | Working towards it, not certified | Claims PHIPA compliance | Your responsibility |
| HIPAA/BAA | Included on Enterprise | Available on Enterprise | Twilio offers HIPAA-eligible |
| Live AI conversations | Yes - built in | Yes - built in | Possible but complex to build |
| Voicemail detection | Automatic | Automatic | Twilio AMD (mature, accurate) |
| Voice cloning | Built into platform | 5-15 clones per tier | Via ElevenLabs (separate) |
| Dashboard for Stephen's team | Built in | Built in | Ben builds it (custom) |
| Stephen can run it independently | Yes, after training | Yes, after training | With custom dashboard, yes |
| Works on all phones | Cell, landline, VoIP | Cell, landline, VoIP | Cell, landline, VoIP |
| Concurrency | 20 (free) / unlimited (enterprise) | 10-100 by tier / custom (enterprise) | Unlimited after verification |
| Audit trail | Full call logs + transcripts | Full call logs | Whatever Ben builds |
| Setup time | 1-2 weeks | 1-2 weeks | 4-6 weeks |
| Ongoing maintenance | None (platform managed) | None (platform managed) | Ben maintains it |
| Platform lock-in risk | Medium (data portable) | Medium (data portable) | None (DOCUdavit owns code) |
When Would You Pick Each Platform?
Retell AI (Pilot)
- Best for the pilot - pay-as-you-go, no commitment
- Fastest to launch (1-2 weeks)
- Lowest pilot cost ($1,085-$1,535)
- Dashboard included from day one
- Strong candidate for enterprise scale too
Bland AI (If Compliance is King)
- Strongest Canadian compliance story (PHIPA claim)
- Self-hosted architecture = best for data privacy
- Worth getting an enterprise quote during Phase 2
- Must verify PHIPA claim before committing
Custom Twilio (If Volume Justifies It)
- Cheapest per-call cost - savings grow with volume
- Maximum control and no lock-in
- Only makes sense at sustained high volume (thousands of calls/week consistently)
- Highest upfront cost and requires ongoing maintenance
The Hybrid Path (Our Recommendation)
- Pilot on Retell AI (pay-as-you-go)
- Get enterprise quotes from Retell + Bland with pilot data
- Verify Bland's PHIPA claim
- Choose the scale platform based on real data and real quotes
- Consider custom Twilio only if sustained volume justifies the build cost
The Rules (Plain English)
Several Canadian laws and guidelines control what DOCUdavit can and can't do with automated calls. Here's what each one means in practice:
🏛️ CRTC (Canada's Phone Regulator)
✅ Custodian Authorization Confirmed: DOCUdavit has confirmed that, as custodian of the medical records on behalf of the doctor, they have the doctor's explicit permission to contact patients. Solicitation concerns do not apply - these are authorized custodian notifications about medical record access.
What it says: You CAN make automated informational calls. Automated sales calls require prior consent from each person.
What it means for us: With custodian authorization in place, our calls are clearly informational and authorized. That said, best practice is to keep the message purely informational - "Your doctor retired, your records are here, call us." Avoiding any mention of fees or services keeps us well clear of any telemarketing classification.
- Calling hours: weekdays 9am - 9:30pm, weekends 10am - 6pm (patient's local time)
- Must identify DOCUdavit and the purpose immediately
- If the message is over 60 seconds, repeat the identification at the end
- Penalties: up to $15,000 per call for violations
📵 Do Not Call List (DNCL)
Good news: The DNCL does NOT apply to DOCUdavit. DNCL only applies to telemarketing calls. DOCUdavit's calls are authorized custodian notifications - informational, not sales. No need to check the list. This saves significant complexity and cost.
🏥 CPSO (Ontario Medical Privacy Guidelines)
The big one for voicemails: You CANNOT leave personal health information on a voicemail. Why? You can't verify who's listening. Maybe the patient's spouse picks up. Maybe a roommate.
What it means: The voicemail must be GENERIC. No patient name. No "your medical records." Just: "Important message regarding Dr. Smith's practice. Please call us." The details only come when the patient calls back and DOCUdavit verifies it's actually them.
🔒 PHIPA (Ontario Health Privacy Law)
What it says: Patient data (names, phone numbers linked to a doctor) must be encrypted, access-controlled, and logged. DOCUdavit is an "agent of the custodian" (the retired doctor) and must protect this data at every step.
What it means for platform choice: Whatever platform we use needs encrypted data handling, audit logs, and the ability to delete data after campaigns. DOCUdavit's ISO 27001 certification helps a lot here - it demonstrates existing security practices.
As volume grows, this matters even more: Processing thousands of patient phone numbers through a third-party platform is a significant data flow. The platform's compliance posture becomes a critical decision factor.
🗣️ Voice Cloning (No Specific Law Yet, But...)
Canada has no voice cloning law yet. But we still need:
- Written consent from the employee whose voice is cloned (a "Voice and Likeness Licence")
- Clear terms about what happens if the employee leaves the company
- The employee creates the voice clone in THEIR OWN account and shares it (platform requirement)
- AI disclosure recommended (not legally required yet): "This is an automated message using a digital version of Sandra's voice"
Voice cloning becomes even more practical with volume. The same 10-20 audio files per doctor get reused across all calls for that doctor. The clone is generated once and used indefinitely (within the licence terms).
📞 Ringless Voicemail (Special Case)
Ringless voicemail is the least regulated option. CRTC Decision 2004-65 approved it. It's not classified as an ADAD (automated dialing device). The phone never rings, so most "nuisance call" rules don't apply. Still need to identify DOCUdavit in the message and follow CPSO voicemail privacy rules.
Ringless VM remains valuable as Round 2 of the cascade at any volume - catching patients who screen unknown callers.
The Problem: No Single Method Reaches Everyone
Whether it's 1,200 patients or 50,000 patients, the reality is the same: some have cell phones, some have old landlines, some never answer unknown numbers, some have disconnected numbers. No single calling method will reach 100% of them.
The solution is a cascade - try the best method first, then work down the list for anyone who wasn't reached.
The Coverage Cascade
Call all patients. The system automatically detects voicemail vs. human, leaves the right message, and logs the result. Retries patients who don't answer (up to 3 attempts at different times of day).
Expected reach: 70-85% of patients either get a voicemail or pick up the phone.
Cost at pilot: ~$180 for 1,200 calls
Cost scales linearly: ~$0.07-$0.15/min depending on platform and volume tier
For patients who never answered after 3 call attempts - drop a voicemail directly into their inbox without ringing the phone. Catches the "screeners" who ignore unknown callers but do check their voicemail.
Expected additional reach: 40-60% of the remaining patients.
Cost: ~$0.06-0.10 per drop (only for the unreached group)
For anyone who couldn't be reached by phone at all - disconnected numbers, no voicemail, never responded. Send them a physical letter. DOCUdavit already does this today, so this isn't new work - it's just a smaller, targeted mailing instead of blasting the entire list.
Expected additional reach: 80-90% of remaining patients (some addresses may be outdated too).
Cost: DOCUdavit's existing mailing process. Much cheaper when it's 200 letters instead of 1,200.
For cell phone numbers that never answered calls or checked voicemail - a simple text: "Important notice from DOCUdavit Medical Records regarding Dr. Smith's practice. Please call 416-781-9083."
Note: SMS may fall under Canada's Anti-Spam Law (CASL) and could require different compliance. Needs further research before implementing.
Cost: ~$0.01-0.03 per text
Coverage Estimate at Pilot Scale (1,200 Patients)
| Round | Method | Patients Reached | Running Total | Cost |
|---|---|---|---|---|
| 1 | AI Phone Calls | ~840-1,020 | 70-85% | ~$180 |
| 2 | Ringless Voicemail | ~70-215 more | 85-95% | ~$15-25 |
| 3 | Canada Post | ~50-150 more | 95-99% | Existing process |
| 4 | SMS (optional) | Varies | ~99%+ | ~$5 |
| Total digital cost (Rounds 1-2) | ~$195-205 | |||
Coverage Estimate at Scale (Example: 10,000 Patients/Month)
| Round | Method | Patients Reached | Unreached | Monthly Cost |
|---|---|---|---|---|
| 1 | AI Phone Calls | ~7,000-8,500 | 1,500-3,000 | $700-1,200 |
| 2 | Ringless Voicemail | ~600-1,800 more | 500-1,500 | ~$90-180 |
| 3 | Canada Post | ~400-1,350 more | 50-250 | Existing process (much smaller) |
| Total digital cost (Rounds 1-2) | $800-1,400/month | |||
Costs scale linearly. At 50,000 patients/month: multiply by ~5. At 100,000: multiply by ~10. Actual volume depends on how many doctors DOCUdavit signs - see volume scenarios above.
How Tracking Works
After Round 1, every patient ends up in one of these categories:
| Status | What Happened | Next Action |
|---|---|---|
| Voicemail Delivered | AI left a message on their voicemail | Wait for callback. Done. |
| Live Conversation | Patient answered, AI delivered the message | Done. |
| Called Back | Patient called the number back | Staff handled it. Done. |
| No Answer (retrying) | Rang but nobody picked up | System retries automatically. |
| No Answer (max retries) | Tried 3 times, never reached | Move to Round 2 (Ringless VM) |
| Disconnected | Number is no longer in service | Move to Round 3 (Canada Post) |
| Fax / Invalid | Hit a fax machine or invalid number | Move to Round 3 (Canada Post) |
After each round, export the "unreached" list and feed it into the next method. As volume grows, this handoff can be automated so the cascade runs itself.
Project Timeline - Phased Approach
This timeline reflects the three-phase strategy: pilot first, evaluate, then scale.
Weeks 1-2: Pilot Setup & Test
Platform: Retell AI (pay-as-you-go)
| When | Task | Who |
|---|---|---|
| Week 1, Day 1-2 | Create Retell AI account, buy Canadian number, complete KYC business verification | Ben |
| Week 1, Day 2-3 | Sandra records voice samples, clones voice in the platform | Sandra + Ben (guiding) |
| Week 1, Day 3-4 | Write voicemail and live-answer scripts, send to Stephen for approval | Ben writes, Stephen approves |
| Week 1, Day 4-5 | Configure AI agent: call flow, scripts, calling hours, retry logic | Ben |
| Week 1, Day 5 | Prepare and sign Voice and Likeness Licence | Ben provides, Sandra signs |
| Week 2, Day 1-2 | Test calls to our own phones - check voice quality, voicemail detection, script delivery | Ben + Stephen |
| Week 2, Day 2-3 | Small pilot: call 10-20 real patients as a test batch. Monitor results. | Ben runs, Stephen monitors |
Weeks 2-3: Run Pilot Campaign
1,200 patients - the real test
| When | Task | Who |
|---|---|---|
| Week 2, Day 4 | Upload full patient list (1,200 numbers). Launch campaign. | Ben (first time), Stephen watches |
| Week 2-3 | Monitor results. System auto-retries failed calls. Handle any issues. | Ben monitors, Stephen's team handles callbacks |
| Week 3 | Round 1 complete. Run Round 2 (Ringless VM) and Round 3 (Canada Post) for unreached patients. | Ben + Stephen decides |
Weeks 3-4: Analyze Results & Get Enterprise Quotes
Turn pilot data into leverage
| When | Task | Who |
|---|---|---|
| Week 3 | Compile pilot results: delivery rate, callback rate, average call duration, voicemail detection accuracy | Ben |
| Week 3-4 | Contact Retell AI for enterprise pricing (with pilot data as leverage) | Ben |
| Week 3-4 | Contact Bland AI for enterprise pricing + request PHIPA compliance documentation | Ben |
| Week 4 | Present enterprise options to Stephen with real data and real quotes | Ben + Stephen |
Weeks 4-6: Decision on Scale Platform
Choose the long-term solution
| When | Task | Who |
|---|---|---|
| Week 4-5 | Stephen decides on scale platform based on quotes, compliance, and budget | Stephen (Ben advises) |
| Week 5-6 | Sign enterprise agreement with chosen platform (or begin custom Twilio build) | Stephen + Ben |
Weeks 6-8: Enterprise Setup & First Scaled Campaign
Go big
| When | Task | Who |
|---|---|---|
| Week 6-7 | Configure enterprise account, set up multi-campaign workflows, create script templates for different scenarios | Ben |
| Week 7 | Train Stephen's team on running campaigns at scale (upload, launch, monitor, export) | Ben trains, Stephen's team learns |
| Week 7-8 | Launch first scaled campaign | Stephen's team (Ben supports) |
| Week 8 | Deliver all documentation: compliance checklist, script templates, voice consent template, operations guide | Ben |
- Sandra's availability for the voice recording session
- Stephen's turnaround time on script approval
- Platform KYC verification (usually 1-2 business days)
- Patient data - we need the CSV from the mailing house before we can test or launch
- Enterprise quote turnaround from Retell AI and Bland AI (1-2 weeks typical)
- If Custom Twilio is chosen, add 3-4 weeks for the build
What Could Go Wrong (And What We Do About It)
Every project has risks. As volume grows, some risks become more significant. Here's what we've identified and our plan for each.
High Impact Risks
What could happen: Poor call quality, numbers flagged as spam by Bell/Rogers/Telus, low connection rates.
How likely: Low-medium. Both Retell and Bland support Canadian numbers, but we need to test with real campaigns.
Backup plan: The pilot is designed to catch this. If Canadian carrier delivery is poor on one platform, we try the other. If both struggle, Custom Twilio uses Twilio's industry-standard Canadian carrier relationships.
What could happen: A patient complains to the CRTC about receiving an automated call. CRTC investigates. Higher call volumes mean more exposure surface.
How likely: Very low - DOCUdavit has confirmed custodian authorization from the doctor to contact patients, our calls are informational (not telemarketing), and we follow all ADAD rules.
Backup plan: Immediately pause the campaign. Review the complaint. DOCUdavit's custodian authorization, compliance documentation, and ISO certifications provide strong legal standing. If needed, switch to Canada Post mailings for that batch.
What could happen: None of the platforms are fully PHIPA-certified (Bland claims it, but we haven't verified). Processing patient phone numbers through a non-compliant platform could be a regulatory issue as volume grows.
How likely: Medium. This is the biggest open question in the project.
Backup plan: Phase 2 specifically addresses this. We verify Bland AI's PHIPA claim and check Retell's progress. If neither passes, Custom Twilio with Canadian-hosted servers gives us full control over data handling. DOCUdavit's ISO 27001 certification provides additional coverage.
What could happen: After committing to a platform with a monthly minimum, DOCUdavit discovers they need to switch. Migration costs, re-configuration, and downtime.
How likely: Medium if enterprise contracts are signed. Typically 6-12 month commitments.
Backup plan: The phased approach mitigates this. We pilot before committing. Enterprise contracts should include exit clauses. Patient data (phone numbers, call logs) is always exportable. The scripts and voice clones can be recreated on another platform.
Medium Impact Risks
What could happen: Enterprise quotes come in higher than our estimates in this document.
How likely: Medium. Enterprise pricing is always negotiable, but the numbers here are estimates until we get real quotes.
Backup plan: Get quotes from both platforms and play them against each other. If platform pricing is too high, Custom Twilio offers the cheapest per-call rate. Having three options gives us negotiating leverage.
What could happen: Sandra's cloned voice doesn't sound convincing. Patients hang up or feel uncomfortable.
How likely: Medium. Quality varies based on recording quality and the platform.
Backup plan: Sandra records the messages herself - real voice, not AI. We lose the "AI-generated at scale" angle but the messages still work. Pre-record 10-20 versions (one per doctor) and upload as audio files. This actually simplifies compliance (no voice cloning consent needed).
What could happen: Old phone numbers, disconnected lines, elderly patients. Only 50-60% reached in Round 1.
How likely: Medium-high. Some records may be years old. Older patient lists will have more disconnected numbers.
Backup plan: This is exactly why the cascade exists. The phone campaign doesn't need to reach everyone - it just needs to reduce the mailing list to a manageable size.
What could happen: Patients hear the voicemail but don't call back. Generic voicemails have lower engagement than personalized ones.
How likely: Medium.
Backup plan: Adjust the script for warmth/urgency. Try different calling times. Follow up with Round 2 (ringless VM) as a reminder. The goal isn't 100% callbacks - it's proving "reasonable efforts" to notify.
Lower Impact Risks
What could happen: Patient phone numbers exposed, call logs leaked, unauthorized access.
How likely: Low. All platforms are SOC 2 certified with encryption.
Backup plan: DOCUdavit's ISO 27001 certification + platform security + our compliance setup minimize this risk. If it happens: follow PHIPA breach notification rules (notify the custodian physician, they notify patients + IPC).
What could happen: The voice clone is tied to Sandra's account. If she leaves, she can revoke access.
How likely: Low short-term, inevitable long-term.
Backup plan: The Voice Likeness Licence defines what happens at termination. For future campaigns, clone a different employee's voice. Or switch to a stock AI voice (less personal but no dependency).
What could happen: Stephen needs French voicemails sooner than expected.
How likely: Low - we'll confirm on the Zoom call.
Backup plan: All platforms support French. A bilingual employee records French samples. We create a second set of audio files. Adds a few hours of work but doesn't change the architecture.
Platforms & Approaches We Evaluated
Before landing on the three platforms presented, we researched and evaluated a wide range of options. Here's what we looked at and why each was set aside.
AI Calling Platforms
No-code drag-and-drop call builder with strong stated compliance (HIPAA, SOC 2, ISO 27001, BAA included).
Why not: Zero Canadian compliance evidence - no PHIPA or PIPEDA mentions anywhere. Zero pricing transparency beyond $375/month Pro plan. Cannot confirm Canadian phone number support. No way to estimate costs at volume without a sales call. Too many unknowns for a healthcare operation.
Developer-focused voice AI orchestration layer.
Why not: HIPAA compliance costs $1,000/month as a separate add-on. No Canadian-specific compliance (no PHIPA, no PIPEDA). API wrapper architecture means patient data passes through multiple third-party services (Vapi + your chosen LLM + your chosen TTS + Twilio) - a data privacy risk for healthcare. No dashboard for non-technical users. Total pilot cost: ~$1,360 for 1,200 calls.
AI calling platform designed primarily for sales conversations.
Why not: Opaque pricing (must talk to sales). Limited voice cloning options. Unverified healthcare compliance. Built for long sales calls, not short medical notifications. Poor architectural fit for DOCUdavit's use case.
Ringless Voicemail Providers
US-based ringless voicemail provider. Only provider with any published pricing: $0.06-0.10/drop for small volume packages.
Status: Viable for Round 2 of cascade. US data residency is a PIPEDA concern for bulk patient data. No Canadian-specific compliance. No bulk pricing disclosed. "Select landlines" only for Canadian support.
Ringless voicemail service with built-in "Mimic AI" voice cloning.
Status: Could not verify Canadian number support, compliance posture, or healthcare suitability. Needs direct vendor contact to evaluate.
Canadian-based providers - data residency advantage.
Status: Promising due to Canadian base, but zero detailed product information available. Pricing, features, compliance certifications all unknown. Worth contacting if ringless VM becomes primary.
Technical Approaches
ElevenLabs' native batch calling feature that auto-dials from a CSV.
Why not: Incompatible with their Zero Retention Mode, meaning patient data would be processed and potentially retained on their servers. For healthcare data under PHIPA, this is a compliance issue on standard plans. Enterprise tier (with BAA) has unknown and likely expensive pricing.
Route live call audio through ElevenLabs for real-time AI voice during calls.
Why not: Adds $0.07/min ConversationRelay fee on top of Twilio's calling costs. More expensive than Retell AI and requires significantly more development work. Overkill when pre-generated audio works fine.
Twilio's drag-and-drop visual call flow builder.
Why not as primary: Still requires significant configuration and some coding for batch outbound campaigns. No built-in voice cloning. More control than Retell but more work for similar results. Kept as backbone of Platform C (Custom Build).
We looked at every viable option before narrowing down. The research covered Canadian telecommunications law (CRTC), healthcare privacy regulations (PHIPA, PIPEDA), voice cloning legality, DNCL exemptions, platform pricing at various volume tiers, and detailed cost modeling for each approach.
Next Steps
Here's how this project moves forward. Each step builds on the previous one.
Read through all the tabs. Pay special attention to the Strategy tab (the three-phase approach), the Side-by-Side tab (cost comparisons at scale), and the Risks tab. Come to the Zoom call with questions and a preference.
We walk through this together and make key decisions:
- Confirm the phased approach - pilot on Retell AI, then decide at scale?
- Live conversations vs. voicemail-only? - Does the AI need to talk to patients who pick up, or just leave messages?
- How many script variations? - Recent retirement vs. years-old files? Different messaging per scenario?
- Who is the voice? - Sandra? Someone else? Bilingual?
- Patient data format - What does the spreadsheet from the mailing house look like?
- Volume projections - How many doctors per year is DOCUdavit signing now? How quickly is that growing?
- Campaign pacing - Do campaigns go out as doctors are signed, or in batches (e.g., quarterly)?
- Budget comfort - What's the monthly budget range Stephen is comfortable with for calling costs?
- Live AI conversations needed, or voicemail-only at scale?
- How many script variations for different scenarios?
- Rough split: cell phones vs. landlines in patient lists?
- French - now or later?
- Who will run campaigns day-to-day on Stephen's team?
Ben sets up the account, Sandra clones her voice, we write the scripts, configure the system, and test with a small batch. Timeline: 1-2 weeks from kickoff.
Full campaign with real patients. Monitor delivery rates, callback rates, voice quality, voicemail detection accuracy. Run the full coverage cascade (AI calls, then ringless VM, then Canada Post for unreached patients).
Ben compiles the pilot data and presents it: what worked, what didn't, what needs adjustment. This data becomes our leverage for enterprise pricing negotiations.
With real pilot data, approach both platforms for enterprise pricing. Request Bland AI's PHIPA compliance documentation. Compare total cost of ownership across all three options.
Based on pilot data, enterprise quotes, compliance verification, and budget - commit to the long-term solution. Sign enterprise agreement or begin custom Twilio build.
Configure the enterprise platform (or build the custom system), set up multi-campaign workflows, create script templates, train Stephen's team, and launch the first scaled campaign.