How to do exactly what we do. Yourself.
This is a complete operational guide to running a human-first review generation operation. Every step, every vendor, every gotcha — written because the honest version of “do it yourself” converts better than any sales pitch. If you're serious enough to read all of it, you deserve the full picture.
It's genuinely achievable. It's also a 4–6 month build with a real team behind it. Start at Step 1 if you're in. Skip to the bottom if you've already decided.
The Compliance Layer
The operational foundation every calling campaign runs on. Before the first call goes out — or the first SMS sends — each piece below needs to be in place. Miss any one and you're either non-compliant, unreachable, or flying blind on conversion.
TCPA + 10DLC SMS registration
You cannot legally send a contact an SMS containing a review link without prior written consent. The Telephone Consumer Protection Act (TCPA) and carrier-level 10DLC registration are both required. Without 10DLC approval, carriers simply filter your messages before they ever arrive — your conversion rate goes to zero before you start.
- Register your brand with The Campaign Registry (TCR) — the industry body carriers use to vet SMS campaigns before approving your sending.
- Define and document your use-case. Review-request SMS sits in a grey zone between transactional and marketing. Document your consent flow carefully — carriers apply different throughput limits per use-case.
- Submit for carrier vetting. AT&T, Verizon, and T-Mobile each run independent review. Total approval time: 4–8 weeks. You cannot expedite this.
- Build your opt-in flow: written consent must be captured before the first SMS sends. A verbal 'yes, send it' on a call is not legally sufficient.
- Build your STOP-handling: any contact who replies STOP must be removed from all future sending within 10 seconds (automated) and recorded for audit (manual). This is a live legal obligation, not a courtesy.
- Enforce quiet hours: TCPA prohibits SMS outside 8am–9pm in your recipient's local timezone — wire this into your sending infrastructure, not your application layer.
EHR / CRM integration + live data pipeline
Calling contacts within hours of their visit — the window that drives 20–40% conversion — requires a real-time or near-real-time feed of completed appointments or interactions. A weekly CSV export won't get you there. You need an automated pipeline from your own system to your calling queue.
- Identify your system. Every EHR, CRM, and practice-management platform has different API access tiers, different integration pathways, and different levels of cooperation with third parties. Some have no API at all and require HL7 FHIR, direct database access, or manual exports.
- Obtain your API credentials. What's available sets the ceiling on how fast your pipeline can move — and how much custom development you need.
- Build your pipeline: appointment-completed event → extract name + phone + visit type → enqueue for calling. Every step must be encrypted in transit and at rest, with access logging.
- Handle your edge cases before launch: contacts with no phone on file, people who previously opted out (you must maintain this list), duplicate records, cancelled or rescheduled appointments.
- Build in a 2–3 week integration testing period before going live. Data quality issues are extremely common and will corrupt your first calls if not caught early.
Local presence dialing infrastructure
Answer rates drop 60% or more when a contact's caller ID shows an out-of-area number or “Unknown.” Local presence dialing — calling from a number that shares the contact's area code — is not optional if you want sustainable conversion. This is why contacts answer applaud numbers at rates that consistently surprise new operators.
- Procure your local phone numbers across every area code you serve. A multi-location business covering 3–5 metro areas needs 20–50 numbers. Source via Twilio, Bandwidth, or Telnyx.
- Build your number rotation: as your call volume scales, individual numbers will get flagged as spam by carrier spam-detection algorithms. Your infrastructure must automatically detect flagged numbers, retire them, and promote clean replacements — without manual intervention.
- Monitor your per-number answer rates in real time. A sudden drop on a specific number signals a spam flag 12–24 hours before it becomes a serious problem. Set threshold alerts.
The Intelligence Layer
Most DIY attempts reach this phase with a generic script and a gut feeling about when to call. That's enough to prove the model. It stops being enough at scale. The difference between a 5–10% conversion rate and a 20–40% one is almost entirely explained by the quality of the intelligence layer — specifically, call timing and script personalisation.
AI-powered call timing optimisation
When you call matters as much as what you say. A customer finishing a medspa appointment on a Thursday afternoon has a completely different answer-probability profile than a dental customer on a Monday morning. Getting timing right consistently — across specialties, demographics, and geographies — requires a model, not intuition.
The variables that actually move conversion rate:
- Time of day and day of week — effects are non-linear. Friday afternoons consistently outperform Monday mornings across every specialty. But the shape differs by age cohort and appointment type.
- Appointment type and specialty: cosmetic procedure completions generate higher emotional engagement and a longer positive-sentiment window. Routine check-ups are shorter. Knowing which is which changes your optimal call delay.
- Age cohort: 18–34 vs 55+ answer patterns differ significantly by hour and by channel. Younger contacts are more reachable in early evening; older contacts in mid-morning.
- Days since visit: conversion rate falls roughly 40% for every 24 hours of delay after appointment completion. The infrastructure you built in Phase 1 exists specifically to call within 2–4 hours.
- Prior contact history: has this contact answered before? At what time of day? What was the outcome? A model that ignores this is leaving significant conversion on the table.
- Device and carrier type: iOS call-screening behaviour differs from Android. Certain carriers apply more aggressive spam filtering in specific hours.
If you're starting from zero, plan to run on heuristics for your first 3–4 months while the model accumulates enough call outcomes to be reliably better. The model improves with every call — but only if your outcome data is clean and labelled from day one.
AI-customised scripts + real-time sentiment routing
Scripts need to be bespoke — built around your own brand, voice, service type, and the specific emotional context of the visit. A caller who sounds like they're reading a generic template breaks trust immediately. The script layer is where the relationship between caller and customer either forms or doesn't.
What a properly built script and routing layer looks like:
- Build your base script architecture: develop distinct call flows per service type, shaped around your brand voice, the type of visit or interaction, and the emotional context the customer is likely in when the call lands.
- Build your A/B testing framework: generate 8–12 opening variants per service type using an LLM (Claude or GPT-4 works well as a draft engine). Measure call duration, positive-sentiment rate, and review conversion. Use proper statistical significance thresholds — don't call a winner on 30 calls.
- Set up real-time transcription: wire your telephony (Twilio Voice) to a transcription API (AssemblyAI or Deepgram) to capture a live transcript as each call happens. This is the raw input for everything downstream.
- Run sentiment classification on every transcript: pipe it through a fine-tuned classifier or a structured GPT-4 prompt chain to score the call's emotional trajectory — especially the final 30 seconds before the review ask. Calibrate your classifier per service type.
- Wire your dynamic routing: positive sentiment → SMS with review link fires within 60 seconds of call end. Neutral or mixed → hold SMS, flag for manual follow-up. Negative → route to your internal feedback form, never send a public review link. This is what protects your rating from unhappy contacts.
- Wire closed-loop learning: every outcome (review posted vs not, star rating if posted) feeds back as a training label for your classifier. The model compounds with data — but only if your feedback loop is wired from day one.
Latency matters more than accuracy here. If the SMS arrives 15 minutes after the call ends rather than 60 seconds, conversion drops roughly 40%. Your transcription-to-classification-to-send pipeline needs end-to-end latency under 90 seconds. This is a real-time engineering problem, not a batch job.
Everything above is buildable. We know — because we built it.
The failed vendor evaluations, the scripts that didn't land, the compliance gotchas, the models that needed months to train — we ran that whole gauntlet so you don't have to. If you'd rather skip straight to results, that's exactly what applaud is for.
The Calling Operation
Every meaningful increment in this process comes from the quality of the actual conversation — the right moment, the right warmth, every single patient. Doing that by hand across your whole schedule is exactly where DIY collapses. applaud runs warm, human-sounding outreach at the optimal moment and routes anyone unhappy privately to you — tuned and monitored by a real team, so the conversion rate compounds instead of collapsing.
Hire, train, and QA your calling team
Offshore calling reduces your per-call cost. It also reduces your answer rate by 30–40% (contacts aggressively screen international numbers) and your sentiment rate significantly (vocabulary, accent, and cultural familiarity all matter). For sustained conversion above 20%, US-based callers are not optional — they're the rate limiter.
- Source your callers from relevant customer service backgrounds — ideally adjacent to your industry so the vocabulary and tone transfer naturally. Empathy and warmth are non-negotiable; specific knowledge can be trained.
- Run call conduct training: cover what you never say on a call, what's in scope, what isn't, and how to handle anything unexpected. Test it, don't just describe it, and document everything.
- Train your callers on brand vocabulary: they need to sound fluent in the language of your business — generic phrasing breaks trust immediately.
- Require live call shadowing: minimum 20 supervised calls before solo deployment. Use your QA rubric during shadowing — the first call scored on a rubric is a more honest signal than any interview.
- Design your QA rubric: score every call across 8–10 dimensions — opening warmth, brand compliance, handling of hesitation, naturalness of the review ask, close quality. Set a minimum passing score for solo deployment.
- Plan for attrition from day one: expect 40–50% first-year turnover in calling roles. Build a continuous hiring pipeline — treating attrition as a surprise is the most common operational failure in this model.
The real hiring signal: callers with high empathy ratings consistently outperform those optimised for volume or call speed. This is a relationship call, not a sales call. Hire for warmth first — the mechanics can be trained.
Daily call operations + review routing
At steady state, the operations loop looks like this every single working day:
- Morning: pull overnight completions from the CRM or EHR feed, run the timing model to generate per-contact optimal call windows, allocate to caller schedules for the day.
- Day: callers work the queue. Every call is recorded. Twilio webhooks fire call-end events to your pipeline. Transcription runs within 30 seconds. Sentiment scores. SMS fires within 60 seconds — or is held for manual review.
- Review routing: positive sentiment → review platform SMS fires immediately. Mixed → flag for operations coordinator to decide. Negative → route to internal feedback form. Every routing decision is logged for audit.
- Afternoon: QA team reviews a random 10% sample of that day's calls. Low-scoring calls trigger same-day remediation feedback to the caller.
- End of day: confirm all TCPA opt-outs from STOP replies have been processed and recorded. Confirm no numbers triggered spam flags. Update the daily conversion report.
- At 50 new visits per week: expect roughly 3–4 hours of active calling per day. That's 1–2 part-time callers at minimum viable scale.
Everything in this guide is genuinely learnable.
We publish it because transparency earns more trust than mystique. If at any point you'd rather work with a team that already has the infrastructure, the track record, and the trained models in place — that's what we do. No judgement either way.
The Measurement Stack
The most common failure mode in DIY review generation isn't the calling — it's not knowing what actually worked. Without full-funnel attribution at every step, you'll optimise for the wrong variable for months before realising it.
Full-funnel attribution
You need visibility at every step in the conversion chain. Without it, you won't know whether a drop in reviews is a timing problem, a script problem, a link-click problem, or a Google spam-filter problem. They each require completely different interventions.
- Map your full conversion chain and instrument every step: visit completed → call queued → call attempted → call answered → positive sentiment → SMS sent → link clicked → review platform opened → review submitted → star rating recorded.
- Store a timestamp and outcome code at every step, per contact. Don't rely on Twilio logs — they're not queryable at the depth you need for optimisation.
- Use your CRM or build a custom database to store the complete journey per contact. You need to be able to query: 'Show me all calls from Tuesday afternoons in the 2–4 hour post-visit window, by service type, that converted to reviews' — and get an answer in seconds.
- Connect to the Google Business Profile API (and any other review platforms you're targeting) — poll for new reviews daily, match against recent SMS sends, and close your attribution loop. This is the only way to know which calls actually resulted in posted reviews.
- Build your weekly dashboard: conversion rate by service type, by caller, by script variant, by time-of-call, by days-since-visit. This is the single artefact your whole optimisation loop runs on.
Ongoing Operations
Everything in Phases 1–4 is setup. This is the actual job. Most DIY operations succeed at launch and quietly degrade over 3–6 months because the ongoing operational load is consistently underestimated. The difference between an operation that compounds and one that plateaus is almost entirely in this phase.
The weekly operations loop
This is what running the system actually looks like every week, without exception:
- Run your contact list hygiene: remove all TCPA opt-outs, identify contacts who already left a review in the last 30 days (calling them again is a compliance risk and a trust risk), and merge duplicates from CRM updates.
- Do your caller scheduling: match your team's availability to the timing model's call volume forecast for the week. Under-scheduling means missed windows; over-scheduling is wasted cost.
- Run your QA reviews: manually score a 10% random sample of last week's calls. Flag calls for retraining. Deliver individual caller feedback within 48 hours — delayed feedback doesn't change behaviour.
- Run your number health audit: check per-number answer rates. Any number showing a sudden drop (>15% below baseline) should be retired and replaced within 24 hours.
- Process your TCPA opt-outs: verify every STOP reply from the week was actioned in real time. Document every opt-out in your compliance log with timestamp and method.
- Iterate your scripts: based on A/B data, retire underperforming variants and propose one new test per service type each month. Document your hypothesis before testing — this keeps the experimentation disciplined.
- Pull your weekly report: review count, rating trend, conversion-rate funnel, spend breakdown, and competitor velocity. This is how you track progress and prove ROI — accuracy and clarity matter.
Continuous model improvement
The intelligence layer from Phase 2 is only as good as the data you feed back into it. The loop never closes on its own — you have to wire it deliberately, and maintain it actively.
- Write every call outcome back to your training dataset immediately — it's a training label. Delay labelling and your model quality degrades. Track: review posted (yes/no), star rating if posted, SMS link clicked (yes/no).
- Retrain your timing model monthly once you have 3,000+ labelled outcomes. Below that threshold, a well-structured heuristic will outperform a model trained on insufficient data. Don't over-trust early model outputs.
- Run your script A/B iteration cycle at minimum quarterly. Fine-tune your sentiment classifier on new call data every 6 months — language and customer expectations drift over time.
- Recalibrate for carrier landscape shifts: spam-detection behaviour on major carriers changes every 6–12 months. Run an annual full review of your number rotation strategy and timing assumptions.
- Stay current on regulatory changes: TCPA case law evolves, 10DLC rules change, FCC rulemaking moves. Schedule an annual review of your consent flows and opt-out handling with counsel — what was compliant in 2024 may not be in 2026.
The compounding advantage of a purpose-built operation: applaud has run this loop across 100,000+ calls. Our timing models, scripts, and classifiers have seen enough data to be reliably better than heuristics. A new operation starts cold. For your first 3 months, you will consistently underperform a trained operation — not because you're doing it wrong, but because the model hasn't seen enough yet.
We do all of this for you. Without the investment, the trial and error, or the learning curve.
Every step in this guide is something applaud has already built, tested, and refined — the compliance infrastructure, the EHR integrations, the call timing models, the specialty scripts, the trained calling team. This is specifically what we do, and we've been doing it long enough to have made most of the mistakes already.
You get all of that without the build time, the trial and error, or the learning curve. We handle everything from contact list to posted review — not as one feature among many, but as the whole thing.
Pay only per posted Google review. No setup fee. No monthly retainer. Cancel any month.
20 minutes on Zoom. No deck, no pressure. We'll pull your Google profile and your closest competitor before the call, walk through the gap live, and tell you exactly what it would take — whether that's us or not.