Chatbot building is the best business in this whole category for a specific reason: it recurs. A one-off design job ends when you deliver. A chatbot needs its content updated, its knowledge base refreshed, its failures fixed and its answers checked every month — and the client knows they cannot do that themselves. That recurring need is what turns a $400 project into a $400 project plus $120 a month, forever.
This is how to build one worth paying for, what to charge, and where beginners get it wrong.
What a business will actually pay for
Not “a chatbot.” One of these four, in descending order of value:
1. A knowledge-base assistant. Trained on the company’s own documents — price lists, policies, manuals, FAQs. It answers accurately because it’s reading their material, not guessing. The most common first sale.
2. A lead-capture and qualification bot. Asks qualifying questions, tags the lead, pushes it to a CRM or sheet, and alerts a human on WhatsApp or email. Directly tied to money, which makes it easy to price.
3. A support deflection bot. Handles the repetitive questions — order status, opening hours, return policy, menu, price — so staff stop answering the same five things. The client’s saving is hours of staff time, which you quantify in the proposal.
4. An appointment or booking bot. Books into their calendar, sends confirmations, handles reschedules. Highest value per conversation because it completes a transaction.
Skip: anything requiring the bot to give medical, legal or financial advice, or to make a decision on a person’s behalf. That is a liability, not a feature.
The platforms, and what they cost in 2026
| Platform | Free tier | Paid from | Best for |
|---|---|---|---|
| Chatbase | 50 message credits/month | $40/month (Hobby) | Knowledge-base bots on the client’s own documents. White-label removal $99/month. |
| Botpress | 100 conversations/month, 3 agents | $150/month annual (Plus) | Technical control, multi-channel agents, WhatsApp. AI usage billed separately. |
| Tidio / Lyro | Basic live chat, one-time 50 AI conversations | $29/month (Starter), Lyro from $32.50/month | Small e-commerce — live chat plus AI in one widget. |
| ManyChat | Free tier available | ~$15/month | Instagram, Facebook and WhatsApp DMs. |
| Botsonic | — | $19/month ($16 annual) | Simple no-code website knowledge bots. |
| Voiceflow | Sandbox free | ~$50/month editor seat | Complex chat and voice agent design. |
The matching rule: build on whatever the client already pays for. Migrating a business off a tool it already uses is a fight you win only once, and the first sale is not the fight.
Model costs are real. Botpress, for example, bills LLM usage separately from the plan — roughly $0.10 of AI usage per conversation, with $25 included at the Plus tier. Budget for it, and either bundle it into the retainer or set an auto-recharge cap with the client. Unbounded AI spend inside a client account is a real business risk.
For comparison, a per-resolution model like Intercom’s Fin charges $0.99 per resolved conversation with a 50-resolution monthly minimum. Different metering, same lesson: the cost model is your problem, not the client’s.
Pricing that works
Setup fee, one-off:
- FAQ knowledge bot: $200–500
- Lead capture with CRM integration: $400–900
- Booking or multi-channel bot: $600–1,200
- Nepali local business: NPR 5,000–20,000 setup
Monthly maintenance:
- $30–80 for a simple bot (content updates, uptime, monthly tuning)
- $100–200 for a bot integrated into their CRM or booking system
- Add tool costs as a clearly stated line, not a surprise
The proposal structure that closes: their current cost in hours, what the bot handles, what it can’t, the setup fee, the monthly fee, and the escalation rule. Send it as a one-page document within 24 hours of the call. Full method in the automation business guide.
The work that actually matters
The platform is an afternoon. These five things are the deliverable:
1. Gather their real content. Price lists, policies, warranty terms, service descriptions, past tickets. If they don’t have this written down, that gap is your first paid project — writing it is billable.
2. Write the scope boundary explicitly. A document listing what the bot answers and what it escalates. This prevents every future argument about a wrong answer.
3. Set human escalation. Anything outside the trained scope routes to a person, with the conversation summary attached. A bot that guesses is worse than no bot.
4. Test with real questions and real bad questions. Blanks, typos, Nepali and Hindi mixed in, angry customers, questions you didn’t train for. Then fix what breaks.
5. Train the team. The client’s staff need to know what it says, what it can’t say, and how to override it. Businesses pay for the bot and then quietly stop using it because nobody showed the staff.
That fifth point is the difference between a project and a retainer. Bots fail from adoption, not from accuracy.
Niche it, and the business compounds
Don’t build “chatbots for businesses.” Build chatbots for restaurants, clinics, real-estate agents, tuition centres, or e-commerce stores in one market. The second client in a niche takes a third of the time, you speak their language, and you can name a specific problem in a specific industry. That is how automation service income becomes a real business — see selling AI services internationally and AI services for small businesses.
The failure modes
- Generic FAQ bot. It answers what was already on the website. No value, no renewal.
- No human handoff. Every frustrated customer becomes an angry review.
- Unbounded AI spend. One client’s bot quietly burning tokens is your loss. Set caps and recharge thresholds.
- Ignoring language. Nepali-speaking customers get an English-only bot and quietly stop using it. Bilingual training is a real differentiator.
- Building and disappearing. The client has questions in week three. That’s the retainer — take it in the first conversation.
- Overpromising accuracy. Say “handles your common questions” not “answers anything.” The second promise creates a legal problem for the client.
Getting paid in Nepal
Local businesses pay by eSewa, Khalti, Fonepay or bank transfer, usually same day. International clients pay through Upwork’s Direct to Local Bank at $0.99 per transfer, or Payoneer. Foreign-currency receipts through a PAN-linked bank account attract 5% final withholding at the bank, final up to NPR 40 lakh a year. Rails: international payments in Nepal. Quote in USD — at roughly NPR 151.8 per USD, a $60 monthly retainer is about NPR 9,100, and ten of them is a real income.
Next steps
- Start an AI automation business in Nepal — chatbots and automation are the same business
- Best AI services to sell to small businesses — the wider local catalogue
- Sell AI services to international clients — Europe overlaps with your evening
- Make money with ChatGPT as a freelancer — the prompts behind good knowledge bases
This week: build one knowledge-base bot for a real business you can reach in person, using their actual price list and FAQ. Show it to them with the scope boundary written down. Ask for the setup fee and the monthly fee in the same conversation.
General information, fact-checked September 2026 — platform pricing and per-conversation AI costs change frequently; verify current plans before quoting. Not financial or tax advice.
