How Yes Madam replaced recruiter phone screens and cart-recovery agents with one voice AI stack
Yes Madam runs at-home salon services across 50+ Indian cities — a two-sided marketplace that needs to hire qualified beauticians at speed and convert millions of high-intent app sessions into actual bookings. They consolidated two separate human teams (talent screening + cart recovery) into Caller Digital's voice AI, running in Hindi, Hinglish, and 6 regional languages.
Two completely different funnels — supply-side hiring and demand-side bookings — both bottlenecked at the phone call
Yes Madam's model only works if two flywheels stay turning. On the supply side, the company must continuously vet, train and onboard beauticians across 50+ cities — most of whom apply via WhatsApp/Google forms in regional languages. On the demand side, customers add high-AOV salon services to their cart but drop off before locking a slot, especially during peak weekend hours.
Both problems were eating headcount. Recruiter teams were spending 60-70% of their day on first-round phone screens that mostly weeded out unqualified or unreachable applicants. Customer-success agents were chasing abandoned carts manually, but couldn't keep up with volume — and most reached customers hours after intent had cooled off.
Vernacular language coverage was the deeper issue. A Bengaluru applicant prefers Kannada, a Patna applicant prefers Hindi with local diction, and a Mumbai applicant might switch between Marathi, Hindi and English in the same call. Hiring a multilingual phone-screening team that could match this was both expensive and slow.
Hiring funnel choked at first-round screening
Thousands of beautician applications a month, but recruiters could only screen a fraction — and most calls hit voicemail or wrong-number. Throughput, not interest, was the constraint.
Manual screening was inconsistent
Different recruiters asked different questions, scored differently, and missed key disqualifiers (no kit, wrong area, language mismatch). Quality of supply varied widely city-to-city.
Cart recovery couldn't run 24/7
Most abandoned-cart events happen evenings and weekends — exactly when human callers aren't on shift. Calls placed the next morning recovered far fewer bookings than calls placed within the hour.
Booking confirmation was an afterthought
Confirmed bookings still saw last-minute cancellations and beautician no-shows because no one was making proactive day-of confirmation calls in the customer's preferred language.
One voice AI stack, two production workflows, three integrations into the existing tech stack
Caller Digital deployed a single voice AI platform that runs two distinct conversational workflows on the same backbone — beautician screening and consumer cart-recovery + booking — each with its own prompt graph, language model behaviour, and integration into Yes Madam's tooling.
The screening agent calls every new applicant within minutes of form submission, runs a structured 4-minute interview (experience, certifications, services offered, area serviceable, kit and equipment, availability), captures answers as structured data, and writes a fitment score back into Yes Madam's ATS. Recruiters now only see pre-qualified applicants.
The customer agent triggers within 20 minutes of a cart-abandon event, talks the customer through their pending booking in their preferred language, resolves objections (price, slot availability, beautician preference), and books the slot via Yes Madam's booking API in real time — including handling rebooking and cancellations. The same agent handles day-of confirmation calls.
Structured beautician screening
4-minute multilingual interview captures 14 data points — experience, certifications, services, kit, area, languages, availability — and writes a fitment score into the ATS for recruiter review.
Sub-30-minute cart recovery
Outbound trigger fires within 20 minutes of cart abandonment when intent is hottest. Customer never has to repeat what was in their cart — the agent has full context.
Live booking via API
Agent doesn't just remind — it actually books. Slot availability, beautician assignment, and payment mode all confirmed in-call. Booking writes back to the production engine before the call ends.
Mid-call language switching
Hindi, Hinglish, Marathi, Bengali, Tamil, Telugu, Kannada, Gujarati. Detects user's preferred language from first response and switches mid-conversation without re-prompting.
Day-of confirmation + rescheduling
Same agent calls 2 hours before the appointment, confirms address and beautician name, and handles last-minute reschedules — cutting last-minute no-shows.
Built-in DPDP & DND compliance
Consent capture on every outbound call, opt-out routing into Yes Madam's CRM, DND scrubbing before dial, all calls logged for the regulatory audit trail.
Two human teams collapsed into one workflow, faster speed-to-lead on both sides of the marketplace
Across the first deployment quarter, Yes Madam ran the voice AI in parallel with the human teams to A/B the outcomes. The screening agent processes the entire applicant queue daily with consistent rubric scoring, and cart-recovery calls now fire while intent is still warm.
| Dimension | Before | After |
|---|---|---|
| First-round applicant screening | Manual, business-hours only | Automated, 24×7, < 5 min latency |
| Screening rubric consistency | Recruiter-by-recruiter variation | Identical 14-point rubric, scored programmatically |
| Cart-recovery coverage | Business-hours, batched next-day | 24×7 within 20 min of abandon |
| Booking inside the recovery call | Customer told to re-open the app | Slot booked via API in-call |
| Language coverage | English/Hindi only | Hindi, Hinglish, +6 regional |
Frequently Asked Questions
The voice agent detects the applicant's preferred language from their first response and switches between Hindi, Hinglish, Marathi, Bengali, Tamil, Telugu, Kannada and Gujarati mid-conversation. Code-switching (e.g. Hinglish to Marathi inside one sentence) is handled natively without forcing the applicant to restart in a new language.
Each call produces a structured record: 14 data points (experience, certifications, services, kit, area, languages, availability), a fitment score, the call recording, and a transcript with sentiment markers. Recruiters open the ATS and see only pre-qualified applicants ranked by fitment.
Caller Digital triggers via webhook from Yes Madam's commerce stack within seconds of a cart-abandon event, and the outbound call typically connects within 15–20 minutes — well inside the window when purchase intent is still warm.
It books. The agent confirms slot availability, beautician assignment, address, and payment mode in-conversation, then writes the booking back to Yes Madam's booking engine via API before the call ends. The customer doesn't need to reopen the app.
Every outbound call captures explicit consent for recording and follow-up. Opt-out responses route directly into Yes Madam's CRM and suppress the number from future campaigns. The DND scrub runs before dial and all calls are archived for the regulatory audit trail.
Yes. The same agent handles inbound calls from customers asking about bookings, beautician availability, or service area — using the same booking and ticketing APIs as the outbound flow.
Run a similar voice AI rollout for your team
We work with Indian enterprises across BFSI, D2C, retail SaaS, and consumer services. Book a 30-minute consult to see what voice AI can do for your call queue.
© 2025 Caller Digital | All Rights Reserved