The first-line problem
The typical inbound lead flow looks like this:
- Customer fills out a "Contact us" form
- The request lands in email
- The manager sees it 2-6 hours later (if it's not a weekend)
- The manager writes back, asks standard questions: what they need, when, what budget
- After 3-5 exchanges, the manager figures out if it's a real lead
- Only then — a call or a proposal
60-70% of the manager's time goes to initial qualification — work that doesn't require human judgment. Just collecting information along a standard script.
If the customer submitted a request at 10 PM and the reply came the next morning — 30-40% of leads have "cooled off". In that window, the customer has already looked at 3 more competitor sites.
These two pain points are a textbook automation case.
What we built
Instead of a "Contact us" form on haiq.agency, we put up a chat widget. An AI consultant that:
- Answers questions about services and pricing — not from its "own head", but from our up-to-date knowledge base (RAG search across documents describing services, cases, FAQ)
- Runs a natural conversation in English or Ukrainian, in our tone of voice
- Qualifies leads via BANT — Need (what task), Timeline (when), Authority (who decides), Budget (understanding of pricing)
- Asks for GDPR consent before collecting contacts (email, phone) — with a link to the policy
- Creates a lead in the admin panel with a short dialogue summary, BANT data, and a score
- Sends a Telegram alert to the team when a lead is hot (score ≥ 80 or the customer explicitly asks for a manager)
All of this happens in 3-5 seconds per message. The customer doesn't notice they're talking to a bot — they just see that they got a fast answer.
How it works under the hood
The tech stack, if you're curious:
- n8n — orchestrates the whole flow (self-hosted on Railway)
- GPT-4o-mini — the model for dialogue and qualification
- text-embedding-3-small + pgvector — vector search across the knowledge base
- Supabase (PostgreSQL) — sessions, messages, leads
- React widget — the frontend on the site
- Node.js/Express proxy — CORS security and buffering
The flow for processing one message looks like this:
Widget
Proxy
Webhook
+ RAG
-mini
+ Lead
Alert
Inside it's 20+ nodes in n8n: session validation, message storage, vector search for the top-5 relevant chunks from the knowledge base, dynamic prompt assembly (rules + context + history + current BANT state), response parsing (text + BANT extract + contact + GDPR state), session update in the DB, lead creation via admin API, hot-lead alert. Executes in seconds.
Instead of "baking" all the knowledge about services into the system prompt (which makes it heavy and slow), we store documents as vector embeddings. On each request we search only the relevant chunks and add them to the context. It's faster, cheaper, and easier to update — you just edit a document in Supabase.
Now to the numbers: SDR vs AI bot
This is the most important part. Let's break it down honestly.
What a junior SDR/manager costs in Ukraine
Based on 2026 market data (Djinni, DOU, LinkedIn Salary Report):
- Salary net: $700-900/month
- Salary gross with taxes: $900-1,100/month
- Baseline: 40 hours/week, ~22 working days per month = 176 hours
- Processing a single inbound lead: 20-30 minutes
- Throughput: 320-528 leads per month, if the manager only handles first-line work
What an AI bot costs
- Custom development from scratch: from $1,000 one-time (basic scenario — 2-3 weeks)
- n8n on Railway: ~$20/mo
- Supabase: free tier or ~$25/mo
- OpenAI API: ~$0.01 per conversation (5-6 messages)
- 1,000 leads/month: ~$10 in API costs
- Total run cost: $50-70/mo regardless of load up to several thousand leads
Year-over-year comparison
| Metric | Junior SDR | AI consultant | Difference |
|---|---|---|---|
| Startup cost | Onboarding, equipment, ~1 week | $1,000 one-time (2-3 weeks) | — |
| Monthly cost | $900-1,100 (with taxes) | $50-70 (infra + API) | −94% |
| Working hours | 176 hrs/mo (22% of time) | 24/7 = 720 hrs/mo | 4× more |
| Throughput | 320-528 leads/mo | Unlimited parallel | ∞ |
| First-response time | 2-6 hours | 3-5 seconds | ×2000 |
| Vacations, sick days | +8-14% per year | None | — |
| Total per year | ~$12,000 | ~$1,720 | −$10,280 |
The bot recoups its investment in 1-2 months against a junior manager's salary on first-line work. Plus: it runs 24/7, handles unlimited parallel conversations, response time is seconds instead of hours. Plus: doesn't quit, doesn't get sick, doesn't come back from vacation with "you know, I need to think it over some more".
But it's not a silver bullet
Honest about the limitations. The bot doesn't — and shouldn't — do:
- Final negotiations and closing the deal. That's human work — reading mood, building trust, negotiating nuances.
- Complex technical architecture discussions. If the customer asks "how do we integrate our legacy system with such-and-such API with 2FA" — the bot can't handle it. Those questions go to the manager.
- Emotional work with a "cold" customer. If the customer is skeptical, annoyed, or just not ready yet — you need a human.
- Personal calls. Discovery calls, presentations, demos — those are live.
The bot handles first-line work. It's like an SDR that never sleeps, handles everyone at once, and passes warm and hot leads to a human with full context. The manager receives the lead together with a short summary:
"Customer runs a cosmetics store, wants Shopify + Rozetka automation, budget $2,000-3,000, ready to start within a month, decides on their own. GDPR consent: yes. Score: 87/100."
The manager starts the conversation already knowing the task, the budget, and the timeline. It's not 15 minutes of data gathering — it's a concrete proposal from minute one.
What this gives the business
1. Nothing gets lost
Customer wrote at 3 AM — got an answer in 5 seconds. Didn't go to a competitor.
2. Managers do real work
Not the grunt work of data collection, but closing deals. That drives conversion up 2-3×.
3. Scale without fear
Launched a Google Ads campaign with 100 clicks a day — the bot handles it. 1,000 clicks — still handles it. No need to hire more SDRs.
4. Legal cleanliness
GDPR consent is captured explicitly, with a timestamp. The privacy policy is shown before data collection. In 2026, working with European clients without this isn't worth the risk.
5. Analytics out of the box
You see which questions customers ask most, where they get "stuck", which case studies land, which ones don't. That's fuel for optimizing the site and your offers.
When it fits, and when it doesn't
Not everyone needs an AI bot. Honestly.
Skip it if:
- You get 5-10 leads a month — do it manually
- Your product is so complex only a CTO handles the first consultation
- Clients come by personal referral and message you straight on Telegram
- You have niche B2B with a months-long deal cycle — different funnel
Worth it if:
- E-commerce, SaaS, digital services
- 50+ inbound requests per month
- Paid traffic (so it pays back quickly)
- Standardized customer questions on the first line
- You want to scale without hiring an army of SDRs
Frequently asked questions
How much does it cost to build a custom AI consultant?
Basic development — from $1,000 one-time, typically 2-3 weeks turnkey. Plus infrastructure: n8n on Railway (~$20/mo), Supabase (free tier or $25/mo), OpenAI API (~$10/mo for 1,000 leads). Total operating cost — $50-70/mo regardless of load up to several thousand leads.
How is an AI consultant different from a regular chatbot?
A regular chatbot follows a script and doesn't understand context. An AI consultant uses RAG search over your knowledge base, holds a natural conversation, qualifies leads via BANT, collects GDPR consent, and creates leads in your CRM with a short dialogue summary. Response time — 3-5 seconds.
How many leads per month can an AI consultant handle?
Technically — unlimited, since the bot processes dialogues in parallel. In practice, for 1,000 leads/month, API costs are around $10. A junior SDR at $1,000/month max closes 320-528 leads (if they only handle first-line work). The bot scales without hiring.
Who is an AI consultant NOT for?
Skip it if: you get 5-10 leads per month (do it manually), your product is so complex only a CTO can do the first consultation, clients come by personal referral, or you have niche B2B with a months-long sales cycle. Worth it if: e-commerce, SaaS, digital services with 50+ inbound requests/month and paid traffic.
What's the payback period compared to hiring a manager?
A junior SDR costs ~$12,000/year (salary + taxes). An AI bot costs $1,720/year ($1,000 development + $60/mo run cost). Difference of ~$10,000/year on first-line lead processing. Payback period — 1-2 months. Plus the bot runs 24/7 and handles unlimited parallel conversations.
Want the same thing?
If you run an e-commerce with 50+ requests a month and you catch yourself thinking "we're losing leads on the first line" — let's talk.
We build AI consultants tailored to specific businesses: with your knowledge base, your tone, your CRM. Development from $1,000, payback 1-2 months.
Discovery call, 15-30 minutes, free — we look at your case and give you an honest answer on whether automation makes sense right now.
📧 hello@haiq.agency
💬 chat on haiq.agency
💼 @haiq_agency on Telegram
This article is based on hands-on experience building an AI consultant for our own haiq.agency site. If you're curious what it looks like in action — just click the chat widget in the corner of this site. That's exactly who you'll be talking to.
