Auto-prompt Yandex/2GIS reviews + AI replies for auto service
An independent auto service (5 lifts, residential Moscow district) was losing new customers to chains rated 4.7+, because its own rating was stuck at 4.1. An n8n pipeline with a delayed WhatsApp prompt and AI replies raised the rating to 4.7, reviews grew from 23 to 184, traffic from maps +47% MoM.
4.1 rating is a filter on new-customer flow
An independent auto service in a residential Moscow district competes on maps with chains and large service stations. The Yandex.Maps and 2GIS algorithm ranks cards higher with rating 4.5+. The service was stuck at 4.1 — new customers see it lower in results, and if they do click, they go to a 4.7+ competitor.
Substantively the service worked fine: average quality for the market, prices below chains. But satisfied customers didn't leave reviews — everyone goes about their business after the car is fixed. But unhappy ones leave reviews, and the 1-star hangs without a response for weeks until the manager "gets around to it".
The result was double asymmetry: positive experience doesn't become social proof, negative gets entrenched and scales. 23 reviews over a year and a half.
Delayed prompt + split flow 5★/≤3★
Closing a work order in Alpha-Auto (1C) triggers a webhook to n8n. 18 hours after car pickup the customer gets a WhatsApp message asking for a rating. 5★ — link to Yandex.Maps and 2GIS, ≤3★ — feedback form + discount on next visit. In parallel GigaChat monitors new public reviews and generates warm replies.
Order close → POST to n8n with customer ID and services
Sweet spot: car already in use, impression fresh, doesn't feel like spam
Message with inline 1-5 star buttons, personalized text
5★ → Yandex/2GIS link; ≤3★ → internal form + discount flow
Monitor new reviews, generate replies with manager review
18 hours after pickup — sweet spot
Asking immediately at pickup — the customer is in a hurry, says "thanks" and leaves. After a week — already forgotten. 18 hours works: the car has already been in use (you can evaluate repair quality), but the impression is still fresh, and there's no feeling of "loaded with spam right after payment".
Separate flows for 5★ and ≤3★
This is the critical part of the architecture. Asking an unhappy customer to write a review on Yandex is suicide. If the customer rates 1-3, they're offered a feedback form (what specifically didn't please) + 15% off next visit. The manager gets a notification and reaches out personally. Only 5★ customers get a link to the public map. This is not "review fraud" — all reviews are left by real customers in their own words; we just don't publicly address those who were dissatisfied.
GigaChat generates replies to new reviews
n8n polls Yandex.Maps and 2GIS every 4 hours for new reviews. For each new review GigaChat writes 2-3 reply variants considering tone and specifics. The manager picks one, edits if needed, and publishes. For negative reviews a separate prompt template — no excuses, with acknowledgment, solution proposal. Time to reply publication: from weeks to an hour.
Metabase dashboard for the owner
Rating dynamics by day, NPS breakdown by work type (diagnostics, tire change, suspension, engine department), source tracking "where new customers come from". The owner sees on one screen that the funnel works and where the remaining growth opportunities are.
Self-hosted n8n + ready Russian services
Orchestrator of all nodes: webhook, delay, split, AI call, monitoring
Trigger source: order close → webhook with details
WhatsApp Business API: prompt-message delivery with inline buttons
Warm-reply generator for reviews, NPS classifier
Monitor new reviews and publish replies
Alternative reputation channel, parallel pull of new reviews
Owner dashboard: rating dynamics, NPS by service type
Storage of reviews, replies, NPS tags for analytics
1C integration, n8n configuration, testing, manager training
hosting + WAPI + GigaChat tokens + change support
What changed in 4 months
weighted across Yandex.Maps and 2GIS
+700% in 4 months, average 4.7
every 1-3★ gets a reply within 4 hours
MoM by UTM tags for Yandex.Maps and 2GIS
The headline number isn't the rating, it's +47% new customer flow from maps. That's a direct result of improved ranking after crossing the 4.5+ threshold. On maps it's "see-click-call" — no performance marketing, zero budget.
Bonus we didn't bake into the KPI: NPS breakdown by service type showed "suspension diagnostics" averages 4.3, while "tire change" — 4.9. The owner optimized the diagnostics process (how cars are received, what's explained to the customer) and brought NPS here to 4.7 too.
Where else the same funnel applies
Universal pattern — "deal close → delayed prompt → NPS split → AI replies on public flow". Fits anywhere with CRM that closes deals and public listings on Yandex/2GIS/Avito:
- → Local e-com / showrooms — after delivery/visit prompt in WhatsApp
- → Restaurants and cafes — after payment via iiko/r_keeper, delayed prompt
- → Medical centers and clinics — after the appointment with medically tactful text
- → Local services — cleaning, repair, in-home professionals, tutors
- → Marketplace sellers — Wildberries / Ozon with NPS mechanics ported to own site
- Base n8n scheme with delay node, NPS split, and AI reply — business configuration 2-3 days
- Prompt for warm review replies — tunable by TOV and industry
- Connectors for Yandex.Maps Reviews API and 2GIS API — ready n8n nodes
- Metabase dashboard: rating dynamics, NPS by service breakdowns, source tracking
If your map rating is below 4.5 — it's fixable in 3-4 months
A reputation funnel is the fastest channel for growing traffic without a performance budget. Time to production — 3-4 weeks. Payback typically 2-3 months from organic map-listing traffic growth.
Аудит за 5 000 ₽ — с конкретным отчётом и сметой
Расскажу что внедрить в вашем бизнесе в первую очередь, какая будет окупаемость, и нужен ли вообще AI для вашей задачи (иногда — нет).
Или просто напишите свой вопрос — отвечу в течение 2 часов