How we automated test-drive scheduling with AI: real AutoFlow 360 case (+35% show-up)

AutoFlow 360 is a LatAm multi-brand distributor with 3 branches, focused on mid-range Asian and European brands. They receive about 650 digital leads/month from Meta Ads, Google Ads, MercadoLibre and web forms. Sales team has 18 salespeople and 4 SDRs who handled leads via manual WhatsApp.

Baseline measured over 2 weeks before implementation: average first-response time on a digital lead 4 hours 12 minutes, test-drive show-up 42%, reschedule rate (appointment moved before happening) 31%, cost per attributed sale USD 412.

The sales GM intuitively knew they were losing sales in the middle funnel but couldn't quantify exactly where. Initial question wasn't 'let's implement AI' — it was 'let's measure first, then decide'. That diagnostic week was what enabled a successful project later.

First diagnostic finding changed project direction. Initial hypothesis: 'leads go cold because WhatsApp doesn't scale'. Measured reality: bottleneck was in 3 specific spots manual operations couldn't solve — initial response latency (4h+), absence of automatic pre-appointment reminders and rescheduling difficulty (customer had to talk to an SDR again).

73% of leads that did NOT show for test drive had one of 3 stories: (1) SDR responded 5 hours later and customer already contacted another dealer; (2) appointment was set but no reminder arrived and customer forgot; (3) customer wanted to reschedule the day before, couldn't reach the SDR in time and the appointment simply didn't happen.

Critical observation for any dealer evaluating AI: the problem is rarely missing channel — WhatsApp was already in use. The problem is that human operations can't scale to the response speed customers expect in 2026 (sub-2 minutes per automotive industry benchmarks). AI doesn't replace the SDR; it takes off tasks humans can't do fast enough.

The solution designed for AutoFlow 360 has 4 layers running in parallel on WhatsApp Business API. Design rule: every interaction not requiring human judgment is handled by the agent; every interaction that does is handed off to SDR/salesperson with full context.

• Layer 1 — Capture and initial qualification: lead enters (Meta Ads, web, MercadoLibre), agent opens WhatsApp in <30 seconds, greets, identifies vehicle of interest, captures name + area + purchase intent (days/weeks/months).
• Layer 2 — KYC and scoring: if lead shows short intent (<30 days), agent requests ID photo via WhatsApp, extracts data with OCR, cross-references credit bureau and assigns close-probability score.
• Layer 3 — Real calendar scheduling: agent queries vehicle and salesperson availability in live calendar, offers 3 concrete slots, confirms with customer, blocks the vehicle and creates the appointment in CRM.
• Layer 4 — Reminders and rescheduling: agent sends T-24h and T-2h reminders via WhatsApp with location + salesperson name + vehicle. If customer responds with change, agent reschedules in 1 message without SDR involvement.

Key technical decision: use mature components, don't build from scratch. Final stack was 80% existing service integration and 20% custom logic for AutoFlow 360-specific flows.

• Channel: WhatsApp Business API via Meta Cloud API (no intermediary BSP to optimize template cost).
• Conversational agent: OpenAI GPT-4o with automotive-LatAm-specific prompt engineering + function calling for actions (query calendar, schedule, reschedule).
• Document OCR: AWS Textract for ID/license data extraction.
• Calendar: self-hosted Cal.com with vehicles and salespeople as 'resources' so the agent sees real-time availability.
• CRM: bi-directional integration with HubSpot (existing) via API; every conversation logged with scoring and status.
• Orchestration: n8n for T-24h/T-2h reminder workflows, escalation to human SDR when agent detects confusion or complaint.
• Reporting: Looker Studio with HubSpot data + agent metrics; GM sees show-up rate, latency and conversion per branch in real time.

Metrics measured with same methodology before and after: 4-week sample, events tracked in CRM, cross-validated manually. No make-up.

• First-response time: 4h 12min → 47 seconds (-99.7%).
• Test-drive show-up rate: 42% → 77% (+35 absolute points).
• 'Lost reschedule' rate (moved appointment that never happens): 31% → 6% (-25 points).
• Sales attributed to test drive: +28% in 90 days (controlling for seasonality).
• Cost per attributed sale: USD 412 → USD 256 (-38%).
• Weekly hours freed for SDR team for higher-value tasks: ~52 hours/week across the 4 SDRs.
• NPS of customers through the flow: 71 (vs auto industry benchmark 38).

No project goes perfect. These are the 3 mistakes we made at AutoFlow 360 that would change the implementation playbook for future cases.

• Mistake 1: not involving SDRs from day 1. Initially designed the flow with the GM and presented it to the team afterwards — found resistance that cost 2 weeks to resolve. Lesson: involve 2 SDRs as co-designers from week 1.
• Mistake 2: underestimating WhatsApp template cost in the model. Project had estimated margin of USD 1,200/month in templates; actual cost was USD 1,850/month. Lesson: simulate worst case x 1.5 before closing pricing with client.
• Mistake 3: launching all 3 branches in parallel. The lowest-volume branch had bugs that would have been caught earlier with sequential piloting. Lesson: 4-week pilot in 1 branch (highest volume for feedback speed), then parallel rollout to the rest.

If your dealer receives more than 300 digital leads/month and your test-drive show-up rate is below 60%, this case is almost 1:1 replicable. Realistic LatAm 2026 budget sits in the mid tier: USD 3,500-5,500 setup and USD 800-1,200/month operation (see cost guide for detail).

Recommended roadmap: week 1-2 diagnosis and baseline; week 3-4 technical setup (WhatsApp API, agent, calendar, CRM); week 5-6 pilot in 1 branch; week 7-8 rollout to remaining branches. Total: 8 weeks to full operation across 3 branches.

The KPI to defend in the boardroom: show-up rate. Most visible, auditable and best correlated with incremental margin. If you raise show-up by 30 points, additional sales pay back the project in 90-120 days — the rest are operational savings and CX improvement materializing afterwards.