Service AI Engineering

A support chatbot that closes the easy tickets and admits what it does not know.

Customers get an answer at any hour, and your team keeps only the questions that need a person. The bot answers from your own documents, passes the whole conversation to a human when it is unsure, and shows your support lead which gaps in your help content to fix next.

Free discovery call. NDA before the first call if you need one.
Support bot · last 7 days Example
Answered from your docs 1,284 of 1,583 conversations 81%
Passed to a person with the full conversation 19%
Questions it could not answerAskedNext step
Do you ship to Norway? 14 Not in the docs, added
Can I pause my plan? 11 Handed to a person
Invoice for last quarter 9 Handed to a person

The bot says when it does not know and passes the whole conversation to a person. The list below is what your team fixes next.

  • Your docs Answers come from your own content
  • Honest It says so when it does not know
  • Clean handoff A person gets the full conversation
  • Tested Against real questions before launch

What does chatbot development involve?

The gain for your business is fewer repeat tickets, faster answers for customers and a support team that spends its day on the cases that need judgement. We build a bot that answers from your own content, hands over when it cannot, and shows what it gets wrong. What it is allowed to answer decides most of the work.

  • Customers get answers from your docs

    Retrieval over your docs and knowledge base, with sources shown and an honest admission when nothing matches. We need your docs, help articles and product specs, and the questions customers ask most.

    Common questions close without a ticket.

  • Customers get account answers safely

    A bot that authenticates the customer and reads only that customer's data. We need access to the systems holding orders, accounts or bookings, and a list of what it must never get wrong.

    Order and booking questions answered without exposing anyone else.

  • Your agents start with the full story

    An escalation passes the whole conversation, the intent and the sources to a human agent. We need your help desk and the rules for when a person takes over.

    Nobody makes the customer repeat themselves.

  • One bot on every channel you use

    The same conversation logic on the website widget, your help desk, Slack, WhatsApp or SMS. We need the channels you use and who owns each one.

    One thing to maintain, however customers reach you.

Should you use a hosted chatbot tool or build your own?

Use a hosted tool when your help content is simple and it fits the help desk you already pay for, and build your own when answers depend on your systems or your data must stay with you.

Hosted chatbot tool or custom build, when each fits and the trade-off
OptionBest whenThe trade-off
A custom bot from usAnswers depend on orders, accounts or bookings, or data must stay in your infrastructureA build to pay for and content to keep current
A hosted tool such as Intercom FinYour help content is simple, you already use that help desk and per-resolution pricing works at your volumeIt reaches your own systems only as far as the tool allows, and the bill grows with volume

The choice is yours, and we will tell you if a hosted tool is enough. If the bot is one part of a larger flow, see AI workflow automation.

How do you vet a chatbot development company?

Ask what the bot does when it does not know the answer. The right answer is that it says so and hands over to a person with the full conversation, not that it guesses.

Ask how wrong answers are measured before launch, who keeps the content current afterwards and what your support lead can see in the analytics. A bot with no way to see its failures will not improve.

Why this matters

Chatbots that confidently make things up cost you trust faster than they save you tickets.

A bot that answers wrong with confidence is worse than no bot at all. Customers who learn your bot lies stop trusting your brand. We build chatbots that retrieve from your real documentation, admit when they do not have a confident answer, and hand off to humans with the full context. Honest bots earn trust. Trusted bots get used.

Try the bot

Type a message. See where it routes.

A small intent classifier running in your browser, with five example intents and a fallback. Type your own message or pick one of the example chips. You will see which intent matched, the confidence scores for the top three candidates, and the response that branch would send.

Classification 0ms
Top intent order_status 0.94
  • order_status0.94
  • escalate0.12
  • general_help0.08
Routed response branch: orders.lookup
Support Bot

I can look up your order in seconds. Share your order number or the email you used at checkout and I will pull it up.

  • 5 intents shown · production bots run 30 to 80 intents per language
  • Simulated · this demo matches keywords in your browser, it does not call a model
  • Eval-gated · intent set updates roll only when evals pass

What we build

A bot your customers will actually choose to use.

Real retrieval over your real docs. Clean escalation. Honest about its limits. Multi-channel from one core. The kind of bot that improves CSAT, not the kind that tanks it.

01

Retrieval over your real documentation

The bot answers from your actual docs, knowledge base, and product specs. Not from generic training data, not from vibes. If your docs say something, the bot says it. If they do not, the bot does not invent.

Answers grounded in your sources, and an honest admission when they are not.

02

Escalation that does not lose context

When the bot hits its limits, the handoff includes the full conversation, the intent classification, and the retrieved sources. Your human agent picks up where the bot left off, not from zero.

The customer does not have to repeat the question.

03

Admits what it does not know

No confident wrong answers. When retrieval fails or confidence drops, the bot says so and escalates. Your customers learn the bot is honest, which makes them trust it.

Customers learn the bot is honest about its limits.

04

Multi-channel from one core

Same bot ships to website widget, Intercom, Crisp, Slack, WhatsApp, and SMS. One conversation logic, one source of truth, every channel. Your team maintains one thing.

A new channel reuses the same conversation logic.

05

Cost ceiling per conversation

Per-conversation cost cap, per-tenant rate limits, prompt caching. You see the spend before the bill arrives.

A per-conversation cost cap you set before launch.

06

Analytics that explain bot behavior

Per-intent resolution rate, escalation reasons, top unanswered questions, retrieval source attribution. Your support lead can see what the bot does and what to fix.

Your support lead sees what to fix without reading transcripts.

Tested

against a held-out evaluation set before launch, not judged by feel

The evaluation set ships with the build, so you can re-run it yourself after any prompt change.

The retrieval layer

Answers grounded in your real documentation.

Embedding-based search across your knowledge base, with confidence thresholds and tenant filtering. The bot never answers from training data when your docs do not back it up.

rag/retrieve.ts ts
    
      
          
          // rag/retrieve.ts
        
          
          import { embed } from './embedder';
        
          
          import { vectordb } from './storage';
        
          
           
        
          
          export async function retrieve(query: string, opts: RetrieveOpts) {
        
          
            const embedding = await embed(query);
        
          
            const results = await vectordb.search({
        
          
              vector: embedding,
        
          
              topK: opts.topK,
        
          
              filter: { tenant: opts.tenant, doc_type: { $in: opts.allowedTypes } },
        
          
            });
        
          
           
        
          
            return results
        
          
              .filter((r) => r.score >= opts.threshold)
        
          
              .map((r) => ({
        
          
                content: r.metadata.content,
        
          
                source: r.metadata.url,
        
          
                title: r.metadata.title,
        
          
                lastUpdated: r.metadata.updated_at,
        
          
              }));
        
          
          }
        
    
  

What it costs

Chatbot pricing follows what it is allowed to answer.

A bot over your public documentation and one that answers about a customer's own account are the same sentence and very different projects. Here is what moves the number.

What moves the number

  • What it draws on Public docs are straightforward. Answering from account data means authentication, scoping and a hard guarantee it never crosses users.
  • How it handles not knowing Saying so plainly and handing to a human is the single most important behaviour and the one most often skipped.
  • Where it lives A widget on the site, inside the product, or in a support desk each change the integration and the expectations.
  • How you keep it honest Content goes stale. Keeping answers current is ongoing work, not a launch task.
  • What you measure Deflection rate, escalation rate and which questions it fails on. Without those you cannot tell whether it is helping.

Small, well-defined work is usually better handled as an hourly block than scoped as a project, and we will say so rather than inflate it into one.

Rough idea to delivery

  1. You send the details What you want built, roughly, plus budget range and timing. The form asks for all of it
  2. Same day, Monday to Friday We read it and reply. Nothing is scheduled before we know what it is about
  3. Then the call Free discovery, booked once there is enough on the table to make it worth your hour
  4. Within 48 hours of the call A written fixed-price quote you approve before anything starts
Tell us about your project

How we work

How a chatbot project runs.

Free
first call with the engineers
10+
monthly developer hours
48h
to a fixed-price quote

Ongoing work uses one blended team rate that covers analysis, front end, back end and QA, with every hour logged.

Tell us about your project
  1. Find the questions worth answeringTwo weeks. We audit your existing support volume, identify the questions worth automating, ingest your docs into the retrieval index, and design the escalation rules. Fixed scope, fixed price.
  2. One channel first, behind a feature flagThree to five weeks. Bot ships behind a feature flag on a single channel first. Retrieval, escalation, and analytics from day one. Your support team shadows real conversations before wider rollout.
  3. Weekly tuning with your support leadTwo weeks. Canary rollout, weekly tuning sessions with your support lead, multi-channel rollout once the bar is hit. Handoff includes the maintenance runbook. Your team owns the bot at the end.

Quick enquiry

Support questions repeating?

Tell us where your answers live today and what the top repeated questions are. If it must answer about a customer's own account rather than general questions, say so.

Prefer the full form? Tell us about your project

No drip sequences, no marketing list. We reply and that is it.

Common questions

Questions before you build a support chatbot

  1. Will the bot replace our human support team?

    No, and we will warn you if you ask us to try. The bot handles the questions humans should not be answering for the hundredth time. Humans handle the questions that need judgment. Done right, your team grows into the higher-value work.

  2. How do you keep the bot from hallucinating?

    Retrieval-first architecture. The bot only answers from your real docs. When confidence drops or retrieval fails, the bot escalates instead of inventing. We measure wrong answers against a test set before launch and again after every prompt change.

  3. Which channels can the bot ship to?

    Website widget, Intercom, Crisp, Zendesk, Help Scout, Freshdesk, Slack, WhatsApp Business, SMS via Twilio, Discord. We can ship to any channel that has a webhook or API. Same bot, same logic, every channel.

  4. How do you handle conversations in multiple languages?

    Claude handles many languages natively. We test in the languages your customers actually speak and tune the retrieval source documents per language. The bot detects the language and responds in the same one.

  5. What about data privacy and PII?

    PII filtering before model calls, configurable data residency, conversation retention limits, no training opt-in. Your security team reviews the data flow before launch, and we document it for a HIPAA or GDPR review.

  6. What do you need from us to build a support chatbot?

    Your help content in whatever form it exists today, a sample of the questions customers really ask, and access to the help desk the bot will hand over to. If it must answer about accounts, we also need access to those systems and a clear list of what it must never say.

  7. How do we know whether the chatbot is helping?

    Watch three things: how many conversations it resolves without a person, how often it escalates and which questions it fails on. The analytics we build show each of them by intent, so your support lead can fix the content behind a failing question instead of guessing.

  8. How much does it cost to build a customer support chatbot?

    Every project gets a fixed price within 48 hours of a free call, and ongoing work uses one blended team rate. A bot over public documentation is a contained project; one that answers about a customer's own account costs more, mostly because of what it must never get wrong. Model usage is billed to your own AI provider account, with a monthly ceiling set before launch.

  9. What are the pros and cons of a customer support chatbot?

    A good bot answers common questions instantly at any hour and frees your team for the hard ones. A bad one invents answers, traps customers in loops and costs trust faster than it saves tickets. The difference is in the build: answers only from your own content, a clear handover to a person, and testing before launch.

  10. Should we use a chatbot tool like Intercom Fin or build our own?

    Use a hosted tool if your help content is simple, you already use that help desk, and per-resolution pricing works at your volume. Build your own when answers depend on your own systems, such as orders, accounts or bookings, when data must stay in your infrastructure, or when the per-conversation bill grows faster than the savings.

What clients say

★★★★★
“Wbcom Designs was methodical, reliable, innovative, and knowledgeable. They communicated with us every step of the way and delivered on time and on budget.”
Upwork review, Forum setup, 2023
★★★★★
“A partner, very trustworthy, very reliable. Gets the job done.”
Upwork review, Site development, 2 years

Rated 4.8 out of 5 across 360+ Upwork jobs since 2010, on our company and founder profiles.See them on Upwork

Our plugin customers rate us 4.8 on Trustpilot (96 reviews).Read them

Ready for a bot your customers will trust?

Tell us what your customers ask most.

Discovery call is free. Fixed-price quote within 48 hours. NDA on request.