How many qualification questions should the bot ask?

A useful range is 3–5. More and clients often leave mid-way. After answers the manager should know service, place, urgency and the next step. The rest is clarified on the call.

Do I need AI if buttons in the bot are enough?

For booking, catalogues and a short price list buttons often suffice. AI helps when clients write free text (Ads, Instagram) that must become CRM fields. At the start usually: buttons + fields, AI as a second layer.

What should we do with cold leads?

Usually a price or FAQ template without an urgent call. Hot and warm go to managers first with a clear SLA. Tune hot/warm/cold to team load: if few “hot” leads, loosen the threshold slightly; if the queue is on fire — tighten it.

When is it better not to use auto-replies?

Complaints, refunds, complex B2B, custom quotes, continuing a manager thread, night emergencies with a live duty person. Also any topic where a wrong price or promise hurts trust more than silence.

How do we link qualification to ads?

Pass source (UTM or bot start param) into a CRM field and weekly review not only lead count but hot share per channel. That shows where to cut budget and where to scale. See also the funnel and Telegram attribution articles.

How long is the pilot and what counts as success?

Two weeks on one channel is enough to see a shift. Success — fewer empty chats, faster first meaningful manager reply, a clear hot/warm queue. “Neural net failure” more often means weak questions, no CRM handoff, or cold sitting in the same queue as hot.

Can the bot name prices?

Only if prices are fixed and live in your base (price list / FAQ) and the bot quotes them verbatim. “From–to”, bargaining, custom quotes and discounts go to a human. A prompt ban “do not invent prices” is mandatory.

AI & Neural Networks18 minJuly 20, 2026

AI in CRM and bots: how to qualify leads before the manager calls

Robert MirovUpdated Sep 5, 2026

This article is for owners of clinics, salons, cleaning services, delivery, construction or training businesses in Tashkent and the regions of Uzbekistan — where leads pour into Telegram, Instagram Direct, WhatsApp and the website, while managers spend half the day answering “hello, how much?” without district, volume, urgency or budget. AI in a bot and CRM (a system that tracks clients and deals) does not replace sales and does not “close deals by itself”. It asks scoring questions (a short set of clarifiers that gauge readiness), normalises answers into fields, tags hot / warm / cold and hands a human the collected context. The manager calls already knowing what the chat is about — instead of starting every dialogue from zero. Below: why qualify early, which questions work in Uzbekistan, how to set up handoff into the card, when auto-replies hurt, how to link qualification to ads, and how to run a 2-week pilot without an “AI department”. A calm overall start with neural nets — AI for small business.

01Why qualify before the manager — not after

A typical day without qualification: an ad or post → the client writes “hello” → the manager spends 5–10 minutes on service and district → some people leave on price or “just looking” → the team feels “lots of leads, few sales”. Early qualification is a filter and structure, not a robot instead of a seller. The goal is simple: before the first meaningful human reply, the card already has service, district (or branch), urgency, source and a clear next step.

Before	After
Manager opens 20 “hello” chats every morning with no context	CRM deals with fields; hot at the top of the queue
“Just checking the price” eats prime team time	Cold gets a price / FAQ template, no urgent call
Empty card or lead only in personal WhatsApp	Deal with source, score and link to the thread
Unclear which channel brings ready-to-book leads	You can compare hot share from Google, Instagram, Telegram

What early qualification gives in practice: • the manager sees service, district, urgency, budget band (if you have one) before the call • “just browsing” does not block people ready to book today • a CRM deal with fields, not an empty row • honest channel comparison: where “ready to book” comes from vs “just prices” Channel link-up into one funnel — CRM with Telegram, Instagram and site.

02Scoring in plain words: what we measure and why

Scoring (lead scoring) is not a “magic neural-net score”. It is a pre-agreed set of readiness signals. You decide what “hot” means for your business; the bot (or buttons) collects those signals. For services in Uzbekistan four axes usually matter most: • what is needed — a concrete service / product, not “interested” • where — Tashkent district, city, branch or “online / on-site visit” • when — today, this week, “just checking” • how to contact — Telegram, call, convenient time Add a budget band only if you have clear segments (“basic / standard / premium”) or a real minimum check. Otherwise an early money question scares people off and yields fuzzy answers.

Signal	Weak	Strong
Service	“Prices?”	“3-room cleaning” / “teeth cleaning”
Place	Not stated	Sergeli, Chilanzar, online
Timing	“Someday”	Saturday / tonight
Contact	Just “ok”	Number or username + convenient time

AI helps when the client writes free text (“need a 3-room clean in Sergeli on Saturday”) and the model normalises the phrase into CRM fields. If clients already tap “service → district → day”, buttons often beat a “smart” chat. More bot scenarios — AI chatbot for site and Instagram.

03Scoring questions and the hot / warm / cold scale

Do not run a 15-question interrogation. In Uzbekistan messengers people leave quickly if the bot “grills” them. 3–5 questions are enough for the manager to know the next step. Base set for services in Tashkent and the regions: • service / product (buttons + “other” with a short text) • district, city or branch (Yunusabad, Chilanzar, Samarkand, online…) • urgency: today / this week / just checking • budget band — only if segments are real • preferred channel and time: Telegram / call / “after 18:00” UZ example: the client writes “3-room cleaning in Sergeli on Saturday”. The bot (or AI) maps: service = 3-room cleaning, district = Sergeli, urgency = Saturday, tag ≈ hot → CRM card + manager notification.

Tag	Typical signs	What the team does
Hot	Concrete service + place + urgency + contact	Call / reply first in the same slot
Warm	Service known, medium timing or unclear budget	Price list + clarification slot within the day
Cold	“Just prices”, no district or timing	Price / FAQ template, no urgent call

Tune thresholds to your funnel and load. If booking is rigid (clinic slots, masters) — buttons are often more reliable than free chat. If clients write long phrases from Instagram — AI for text normalisation pays off faster.

04Buttons or AI: what to choose at the start

A common mistake is buying a “neural net in the bot” when 70% of leads close with three buttons. The opposite mistake is an endless FAQ with no fields — after which the manager still re-asks the same things.

Situation	Prefer buttons	Prefer AI + fields
Short fixed service catalogue	Yes	Rarely needed
Client writes free text from Ads / IG	Partly	Yes — normalise into fields
Rigid slot booking	Yes	Only as help next to buttons
Lots of “other” and non-standard requests	Weak	Yes, with human escalation
Complaints and refunds	No auto	Straight to a human

UZNEO rule of thumb: buttons and required fields first, then AI on free text. AI must not invent prices “on its own” — only read your price list and FAQ. For ad-copy drafts the model fits better separately — AI for ad and landing copy; bot prices only from your document.

05Handoff to the manager: what must land in the card

Qualification fails if the result is another “please call back” chat with no fields. Handoff is the moment the bot (or form) passes the lead to a human: a deal is created, a notification goes out, the manager sees a summary. Minimum on handoff: • 2–4 line dialogue summary in plain language, not a raw log • fields: service, district / branch, urgency, source • score or tag: hot / warm / cold • link to the Telegram / Instagram thread • next step: “call before 18:00” or “send price list” Technically a typical chain: Telegram bot → webhook → CRM deal. Ready scenarios for CRM and bots beat building integrations from scratch and fixing them at night. Handoff checklist: ☐ CRM card created ☐ Source filled (telegram / instagram / website / ads) ☐ Score set ☐ Manager notified (not only “saw it in the feed”) ☐ No diversion into personal WhatsApp around CRM ☐ Cold is not in the same queue as hot

Before	After
“Someone wrote in Direct” — go find it	Notification + link + fields
Lead lives in the manager’s phone	Rule: no card = lead not accepted
After team leave, leads “vanished”	History in CRM, not personal chats

Team rule: no card = lead not accepted. Otherwise in a month you are again optimising chaos.

06When to turn auto-replies and AI off

Not everything should be automated. In Uzbekistan medicine, legal topics, custom quotes, refunds and any “you promised something else” conflict are especially sensitive. Better straight to a human: • complaints, refunds, “I was cheated”, negativity in public comments • complex B2B: tender, contract, post-pay, several decision-makers • wide “from–to” prices with many variables (repair estimate, custom course) • client already talking to a live manager and expects continuity • night emergencies — if you have a duty doctor / master, not an FAQ bot AI must not invent discounts, service availability, warranty terms or medical claims. Working setup: knowledge base (price list + FAQ) + prompt bans (“do not name prices outside the list”) + escalation on disputed topics.

Topic	Auto-reply	Human
Hours, address, service area	Yes, from the base	If the area is disputed
Fixed price from the list	Yes, verbatim	On bargaining and exceptions
Complaint / refund	No	Always
Custom estimate	No	Always
“Best in the city” / % guarantees	Ban in the prompt	Only your approved text

One wrong price from the model hits trust harder than twenty minutes of honest “we will pass this to a manager”.

07Link to ads and the full funnel

Qualification is stronger when the lead source is clear. Otherwise you never learn which creative and which channel bring hot leads vs only “prices?”. Minimum attribution: • UTM on the site / in ads or a bot deep-link start param → CRM “source” field • one funnel: site / Ads → bot or form → qualify → manager • weekly: how many hot from Google / Instagram / Telegram / Direct — not only “how many leads”

Without source	With source in CRM
“Ads don’t work” on emotion	Clear: channel A brings cold, channel B — hot
Budget cut where it hurts to hear	Budget moves toward channels with hot share
Bot and Ads live separately	One chain through to the deal

Full chain — site + Ads + bot + CRM funnel. When clients write on Telegram without a site form — how to count Telegram leads (attribution).

08Typical qualification mistakes in small business

The same mistakes repeat from cleaning to clinics. Better to see them in a table before the pilot than “fix the neural net” a month later.

Mistake	Why it hurts	What to do
10+ questions in a row	Client leaves on question 3	3–5 questions; the rest to the manager
Score with no action	Tag exists, queue is the same	Hot → separate priority and SLA
AI without a price list	Invented prices at night	Prices only from Docs / CRM
Handoff into a personal chat	History disappears	Rule “no card = not accepted”
Cold in the same queue	Hot wait while you answer “prices”	Template for cold; live focus on hot/warm
All channels at once	Setup chaos	Pilot on one channel for 2 weeks
Expect the bot to close the deal	Disappointment and shutdown	Bot gathers context; a human sells

If numbers did not move after the pilot — fix questions, hot/warm/cold thresholds and handoff. Do not buy “another neural net” on top of a broken process.

092-week pilot without an “AI department”

Run a narrow scenario. The goal of two weeks is not to “implement AI”, but to answer: are there fewer empty chats, and is the manager’s first meaningful reply faster? Pilot steps: 1. List 10 frequent client questions and 4–5 CRM fields without which the manager will not call 2. Put FAQ and price list in one document — the base for bot and AI 3. Turn on qualification on one channel only (often Telegram) 4. Handoff only hot + warm; cold gets a price / FAQ template 5. Every 3 days fix the script and thresholds from real bot mistakes 6. Assign one person who reviews disputed dialogues

Metric	How to measure in 2 weeks
Time to first meaningful manager reply	Before / after on a sample of chats
Share of empty “hello” with no fields	% of chats without service/district
Share of hot among all channel leads	Tag in CRM
Bot errors on price / district	List of 5–10 cases → fix the base

After two weeks compare the numbers. If they did not move — questions and handoff, not a new subscription. If they moved — expand to a second channel and harden the rule “no card = lead not accepted”.

Summary

AI in CRM and bots pays off where managers drown in “hello” with no context and leads live in five personal chats. Short scoring (3–5 questions), a hot / warm / cold scale tuned to your funnel, summary and fields into the card, ad source, no automation of complaints, complex quotes or disputed prices. Need a stack for your lead volume — CRM and Telegram bot from UZNEO. If you are only choosing the first neural-net task — start with AI for small business.