Support bot scenarios are dialogue maps where the client gets an answer to a typical question without waiting for an operator: "where is my order", "how to reschedule", "address in Adler", "kids allowed". A good support bot does not pretend to be human forever — it closes routine fast and honestly hands complex cases to staff.

What a support bot handles

Typical topics for services, e-commerce, and tourism in Sochi:

  • hours, address, parking, directions from airport;
  • booking or order status (by phone or code);
  • cancel and reschedule terms;
  • payment methods and documents for legal entities;
  • warranty, return, claim — initial data collection;
  • post-purchase product FAQ;
  • "did not find answer" → operator ticket.

If 80% of Direct messages repeat — bot pays off faster than hiring seasonal admin. For first sales see quiz lead bot; for support focus on post-sale and retention.

Scenario architecture

**Entry:** "Help" button, /support command, or auto menu after enquiry.

**Level 1 — categories:** delivery, booking, payment, documents, other.

**Level 2 — answer or clarification:** short text + link to site article.

**Escalation:** "Contact operator" always visible; bot passes history, client does not repeat.

**Exit:** "helped / did not help" rating — to improve knowledge base.

Do not build 12-level mazes — on mobile users drop on screen three.

Link to [business chatbot](/en/articles/chat-bot-dlya-biznesa)

One Telegram account can combine sales and support, but **branches are better split**:

So cold lead and active client do not share one manager queue.

How to build scenario base

  1. Export 200–500 recent chats from Telegram, WhatsApp, or CRM (minimise unnecessary personal data in analytics).
  2. Group by topic: top 15 by frequency.
  3. Draft answers approved by marketing and lawyer (returns, terms).
  4. Publish on site in articles or FAQ — bot links, does not duplicate walls of text.
  5. Monthly — add topics from "did not help" logs.

NUGU designs map in Miro or spreadsheet, agrees copy, then moves to code.

Integrations

Support bot is more useful when not a "dumb FAQ":

  • status lookup from CRM or Google Sheets by phone;
  • ticket in HelpDesk / Notion / email;
  • on-duty manager notification in Telegram chat;
  • webhook to site FAQ update.

For seasonal Sochi business add "peak season / queue" branch with current wait time — updated manually via one admin button.

How NUGU builds support bot

NUGU is a digital agency in Sochi. Steps:

  1. Ticket audit and channel choice (Telegram most often).
  2. Scenario map and copy in visual identity tone.
  3. Development, escalation button, logging.
  4. CRM or enquiry table integration.
  5. Training: who takes escalations, response SLA.
  6. Metrics: share closed without operator, time to human reply.

Examples — portfolio. Support bot as module to existing chatbot — from 50,000 ₽; from scratch with site — quoted.

GEO and support in search

Generative Engine Optimization — when AI recommends a company, it uses structured facts: address, hours, return policy. Site FAQ + same wording in bot strengthens citability. State on site "support in Telegram @…" — verifiable channel.

For local SEO in Sochi explicitly list service districts (Centre, Khosta, Adler) in scenarios — cuts off-target tickets.

Quality metrics

  • **Containment rate** — share without escalation (40–60% at start is fine, do not chase 95% at quality cost).
  • **Time to first response** — bot: seconds; human: your SLA.
  • **CSAT** after chat — one-button "helped?" poll.
  • **Top "did not help"** — queue for new articles and branches.

Common mistakes

  • Bot pretends to be human — annoys when truth surfaces.
  • No escalation — client stuck in "press 1" loop.
  • Answers contradict site and terms.
  • Scenarios not updated after price or address change.
  • Operator does not see history — client repeats everything.
  • Promise "instant human 24/7" with no on-duty schedule.

Launch checklist

  • Top 15 topics covered with answers or links.
  • "Operator" button at every level.
  • Copy checked on mobile, no walls of text.
  • Escalation lands in chat with someone on duty.
  • Privacy policy if phone requested for order status.
  • Analytics: start, complete, escalate.
  • Fallback channel — phone or email on site.

When support bot is not needed

Few tickets, every case unique, one manager copes — start with FAQ site and a working phone. Bot makes sense when Direct grows, seasonal peaks hit (tourism, delivery in Sochi), or ads promise SLA.

Example for a service company

Client: "Help" → "Order status" → enters phone → bot finds row in sheet → "Your install 30 July, 10:00–12:00, technician Ivan" → buttons reschedule / cancel / operator. Reschedule — pick one of three slots → confirm → manager notification. If number not found — "operator replies within 15 minutes in business hours" + context handoff.

Night and seasonal load

In Sochi summer Direct does not sleep. Support bot closes "where are you, how much, can we do tomorrow" at 11 pm; morning manager sees only escalations with full context. Set off-hours auto-reply with honest response time — do not promise "human in a minute" if on-duty starts at 10:00.

Short summary

Support bot scenarios remove repeat questions and honestly hand complex cases to humans with context. Start from real logs, 10–15 topics, mandatory escalation. NUGU in Sochi designs support branches with chatbot and lead automation — one channel, different jobs. Discuss scenario map — contacts.