Most chatbot projects go wrong in the first meeting. Someone says "we should get a chatbot," a tool gets bought, and three months later it's a chat bubble that says "How can I help you today?" and then can't.
The tool usually isn't the problem. The problem is that nobody decided what the bot was for.
If you want to build a chatbot for business use, treat it less like a software install and more like hiring a very narrow, very fast front-desk assistant. You'd decide what the job is, give them the information, check their work and make sure a manager is nearby. Those are the three steps below. If you're still deciding whether you need one, start with our plain-English explainer on what a chatbot is and come back.
Before step one: collect the questions
Skip the technology for an afternoon. Open the last month of emails, contact form messages, texts and voicemails, and tally what people ask.
You'll find a pattern fast. A handful of questions will make up a large share of the traffic: hours, service areas, how the process works, whether you handle a certain kind of job, how to book. Those are the raw material for your bot. They also decide whether you need a bot at all. If every inquiry is different, a quick personal reply will beat any automation.
Keep that list. Everything that follows is built from it.
Step 1: Decide the job and where it lives
Pick one job to start. Not five. A bot that does one thing well earns trust, and you can add more later.
Good first jobs:
- Answer your top ten questions on your website.
- Capture and qualify leads by asking a few questions and passing the details to your team.
- Book or reschedule appointments against a real calendar.
- Route requests to the right person or department.
Then decide where it lives, based on where your customers already are:
- Your website. The most common home, and the one you control completely. Good for visitors researching or ready to inquire.
- A messaging app or social inbox. Useful if customers already message you there.
- Your client portal or app. Best for existing customers who need status updates, documents or support from inside an account.
Start with one place. A bot on three channels means three things to test and keep current. You can always add a second channel once the first one is working well and you've seen how real customers use it.
Finally, define what "working" looks like in numbers you can check: fewer repeat questions in your inbox, more leads captured after hours, more bookings. If you can't say how you'd measure success, you can't tell whether you built the right thing.
Step 2: Choose how you'll build it
Here are the options, from lightest to heaviest.
A ready-made chatbot platform
You subscribe to a product, build the conversation in a visual editor, paste a snippet of code into your site and you're live. This is the right call for most businesses starting out. It's quick, the vendor handles hosting and updates, and you can get a feel for how customers use it before committing more.
The tradeoffs are the usual ones with off-the-shelf software. You work within the product's limits, you pay monthly, and your conversation data lives on someone else's system. Check where that data is stored and how it can be exported before you commit. And if your needs grow, you may outgrow it.
A platform plus integrations
This is the middle ground. You use a ready-made bot but connect it to your other tools so it can do more than talk: drop leads into your customer records, check availability or notify your team. These connections are usually made through an API, and it's where a developer starts to be useful. Our integrations work covers this kind of plumbing.
A custom chatbot
You build the bot around your own systems. It makes sense when the bot has to do real work inside your business: look up a client's project in your portal, apply your scheduling rules, write into your records or handle private information under your own security rules.
It also makes sense when a platform keeps forcing workarounds. If you're weighing this, our guide to build vs buy software gives you the framework we'd use, and we'll tell you honestly when buying is the better answer. When a custom build is the right call, that's the kind of project our application development team builds.
What drives the cost
We don't quote prices in articles, but we can tell you what moves the number: how many systems the bot connects to, how many different things it can do, how much private data it touches and how much testing it needs. A narrow bot on a platform is a small commitment. A bot wired into your scheduling, records and portal is a software project. The Price Guide explains how we think about custom work.
Step 3: Teach it, test it and launch it with a person standing by
This step is where bots are won or lost, and it's mostly writing, not coding.
Give it real content
Use your own words from the question list: how you actually answer, in your actual tone. Keep answers short, because people read chat in glances. If you're using a bot built on a language model, give it your approved information and instruct it to answer only from that. It should say "I'm not sure, let me get someone" rather than guess. A fluent wrong answer about your pricing, policies or availability is the biggest risk with this kind of bot.
Write the conversation, not just the answers
Map how a chat should flow. Greet the visitor and say plainly that it's a bot. Offer a few clear starting options. Ask one question at a time. Confirm what you heard before acting. And write the dead ends on purpose: what the bot says when it doesn't understand, when someone is upset or when a question is outside its job.
Always give a way out
The Nielsen Norman Group's usability research keeps coming back to the same frustration: people hate being stuck with a bot that can't help. Every chat needs an obvious route to a person: a phone number, a callback request or a message that lands with your team. Hand off sooner when the customer sounds frustrated, and pass along the conversation so they don't have to repeat themselves.
Test it like a customer
Before launch, throw real questions at it. Use the list you collected. Then try the awkward ones: typos, vague questions, rude messages, questions about things you don't offer, and requests for private information. Have someone who didn't build it try to break it. Fix what fails.
Launch small
Release it on one page or to a small group first. Watch the first batch of conversations closely. You'll learn more from fifty real chats than from any amount of planning.
Keep it healthy after launch
A chatbot isn't a set-and-forget feature. Plan a regular check, say monthly:
- Read the conversations. Look for questions it couldn't answer and add them.
- Check the hand-offs. Are customers reaching a person quickly, and do they sound relieved or annoyed?
- Update the content. When prices, hours, services or policies change, the bot must change the same day.
- Review what it collects. Make sure you're storing only what you need and that it's protected.
- Look at the results. Compare against the numbers you chose in step one.
If the bot isn't earning its keep after a fair trial, switching it off is a legitimate outcome. Better to know than to keep a bubble on your site that quietly annoys people.
Where it fits in your bigger picture
A chatbot works best as one part of a site that already does the basics well: clear information, fast pages and easy contact options. It can't cover for a confusing website. If your visitors aren't contacting you, check why the website isn't generating leads before adding more tools.
For the businesses where a bot makes sense, it's a quiet win. Questions get answered at 10pm, leads get captured before they go cold and your team spends its day on the work that needs a human. If you'd like help deciding what to build, or building it, we're happy to talk it through on a discovery call. Bring your list of questions.