Example of a Chatbot, Built Start to Finish: One Small Bakery’s Website Assistant, Piece by Piece

Example of a chatbot articles usually show a single impressive conversation. This one shows everything behind it: a small, made-up neighbourhood bakery builds its website assistant from nothing, and you see each piece, from the facts it learns and the first message it sends to the mistakes it makes in week one and how they get fixed.

πŸ‘€ What the visitor seesCrumb & Co Β· website chat
πŸ€– Crumb & CoHi! πŸ‘‹ Ordering a cake? Custom cakes need 2 days’ notice. Ask me anything.
can i get eggless chocolate cake for sunday
πŸ€– AI Agent Β· from β€œCake menu”Yes! Eggless chocolate is one of our most popular. A 1 kg cake is β‚Ή1,100 and serves about 10. Order by Friday 6 pm for Sunday. Shall I take your details?

The finished bot, two weeks in.

⚑ In short

  • The business: Crumb & Co, a fictional two-person bakery that takes custom cake orders online and by phone.
  • The goal: stop answering the same cake questions on WhatsApp all evening, and capture more orders.
  • The build: one page of facts, one welcome line, two Flows, three handover rules. About three hours.
  • The lesson: most of the work was writing down facts the owners already knew.
Example of a chatbot in LiveChatBot: AI Agent knowledge sources and a sample chat answered from them
How it looks in LiveChatBot: AI Agent knowledge sources and a sample chat answered from them.

The business

Meet Crumb & Co (a made-up bakery with very real problems)

Crumb & Co is an imaginary neighbourhood bakery in Bengaluru, run by two partners, Anita and Joseph. They sell bread and pastries over the counter, and their real money comes from custom cakes for birthdays and anniversaries. Most cake orders start with a question: on the phone, on WhatsApp, or through the contact form on their simple website.

The problem is volume and timing. Questions arrive all evening, when Anita and Joseph are closing the shop or baking for the next day. The same questions repeat constantly: Is it eggless? How much for 1 kg? Can you deliver? How many days’ notice? By the time they reply, often the next morning, some customers have ordered elsewhere.

They decide to add a website chatbot. This page follows exactly what they do, step by step, as an example of a chatbot built properly by a tiny team with no technical background. The details are invented, but every step is one you can copy.

Piece 1 Β· The facts

One page of knowledge, written in an evening

Before touching any settings, Anita writes down everything customers ask about.

How she built the list

She scrolled back through two months of WhatsApp messages and wrote down every question, then grouped them. Twelve groups covered almost everything.

What went into the bot’s knowledge

– Cake menu: flavours, eggless options, sizes, prices – Notice needed: 2 days for custom cakes, same day for menu cakes before noon – Delivery: within 6 km for β‚Ή80, free above β‚Ή1,500; pickup any time in opening hours – Opening hours: 8 am to 9 pm, closed Mondays – Payment: UPI, card, cash on pickup; 50% advance for custom cakes – Custom designs: photo cakes, themes, message on cake, price varies – Allergens: nuts used in the kitchen; no guarantee for severe allergies – Cancellation: full refund of advance if cancelled 24 hours before

Why this matters most

Every good answer in the final bot came from this page. Every bad answer in week one came from something missing from it.

Writing tips Anita followed

– Numbers, not adjectives. “β‚Ή1,100 for 1 kg” rather than “reasonably priced”. – Customer words. She included “eggless”, “without egg” and “veg” because customers use all three. – Honest limits. The allergy line is deliberately cautious, so the bot never promises a nut-free cake.

She added this page to the AI Agent‘s knowledge and also saved the eight most common answers as saved replies, so the free keyword bot could answer even without AI replies.

Piece 2 Β· First impression

The widget, the name and the welcome line

🎨

The look

The bakery’s warm brown as the main colour, the logo in the header, and the name ‘Crumb & Co’ rather than ‘Support’. Took ten minutes using the customisation settings.

πŸ‘‹

The welcome line

First draft: ‘Hello! How can we help?’ Final: ‘Ordering a cake? Custom cakes need 2 days’ notice. Ask me anything.’ It answers the most common question before anyone asks.

🀝

The honesty line

The bot introduces itself as the bakery’s assistant, not as Anita. When Anita or Joseph join, a line says so. No pretending.

Piece 3 Β· Flows

Two proactive messages, and only two

Flows let the bot start a conversation when a visitor does something specific.

πŸŽ‚

Flow 1: Custom cakes page

Trigger: visitor spends 30 seconds on the custom cakes page.

Message: “Planning a celebration? Tell me the date and how many people, and I’ll suggest a size and price.”

Why: visitors on this page are close to ordering but unsure about size and price.

πŸ”

Flow 2: Returning visitors

Trigger: visitor returns to the website for the second time or more.

Message: “Welcome back! Ready to order, or still deciding on a flavour?”

Why: returning visitors are often comparing bakeries. A friendly nudge helps.

Both were built from templates on the chat Flows page in a few minutes. Anita resisted adding a third: one proactive message per visit is plenty.

Piece 4 Β· Handover rules

When the bot steps aside for Anita or Joseph

The partners agreed three rules for when the bot should stop and pass the conversation to a person. Writing them down took five minutes and prevented most potential disasters.

  1. Custom designs. Any request for a photo cake, theme cake or unusual design goes to a person, because price and feasibility depend on the details. The bot collects the date, size, design idea and phone number first.
  2. Allergies. Any mention of a severe allergy goes to a person. The bot says the kitchen uses nuts and that the team will call to discuss.
  3. Problems. Any complaint, late order or mistake goes straight to a person, with an apology and a promise of a call.

Outside opening hours, the bot collects a phone number and says the team will call after 8 am. Messages left overnight become tickets, so nothing is forgotten in the morning rush. The guide on opening hours shows how to set this up.

Piece 5 Β· Testing

An hour of trying to break it

Joseph tested from his phone, pretending to be different customers.

Test questionFirst resultFix
“eggless choc cake price”βœ… Correct price and sizeNone needed
“can u deliver to koramangala”❌ Said ‘delivery within 6 km’ without knowing distanceAdded a list of covered areas by name
“cake for 25 ppl”⚠️ Suggested 1 kg (too small)Added a size guide: 1 kg β‰ˆ 10, 2 kg β‰ˆ 20, 3 kg β‰ˆ 30 people
“is it nut free”βœ… Cautious answer and offer of a callNone needed
“I want a Spider-Man cake”βœ… Collected details, handed overNone needed
“my cake was late yesterday”βœ… Apologised, collected number, handed overNone needed
“do you have gluten free”❌ Said yes (wrong!)Added: no gluten-free cakes currently

Two wrong answers out of seven would have caused real problems: a delivery promise to an area they don’t cover, and a gluten-free claim they can’t make. Both were fixed by adding facts, not by changing any technology. This is the single most useful lesson from this example of a chatbot: test with real questions before launch.

Piece 6 Β· The first month

What happened after launch, week by week

The bot went live on a Thursday evening. Here’s how the first month went, as the partners reviewed conversations each Sunday.

Week 1

Evening questions now got instant answers. But reading the chats, Anita found visitors asking “what flavours do you have?” and getting a long list of every flavour and price in one message. She split it: the bot now lists flavours briefly and asks which one they’d like prices for.

Week 2

Several visitors asked about “photo cakes” using different words: “picture cake”, “cake with photo print”. She added those phrases to the knowledge. She also noticed the custom-cakes Flow was getting replies but many people left after the bot asked for their phone number too early. She changed the order: size and price first, phone number only when they were ready to book.

Week 3

Orders through chat picked up. Joseph started using the Android app to reply to handovers from the kitchen instead of waiting until closing time.

Week 4

The partners checked their one number: how many evening WhatsApp messages they still had to answer by hand. It had dropped noticeably, because most of those questions were now answered on the website before customers ever messaged. They decided to keep the setup as it was and review monthly from now on.

Lessons

What this example of a chatbot teaches any small business

πŸ“‹

Facts first

Three hours of building was mostly writing down facts. The technology part took less than an hour.

πŸ’‘

Answer before they ask

The welcome line answered the most common question (notice needed) before anyone typed.

✌️

Two Flows are enough

One on the key page, one for returning visitors. More would have felt pushy.

🎨

Hand over the creative part

Custom designs, allergies and complaints went to people. That’s where the partners add value.

πŸ§ͺ

Test with real questions

Two of seven test questions exposed answers that could have cost money or trust.

πŸ”

Review weekly at first

Each weekly review fixed one or two things. Small fixes added up fast.

Your turn

Building your own version

You can follow the same six pieces to build your own example of a chatbot for any small business: write down your facts, set up the widget and welcome line, add one or two Flows, agree handover rules, test for an hour, and review weekly for a month. A salon, a tuition centre, a repair shop or an online store would end up with different facts and Flows, but the same shape.

Everything Crumb & Co used runs on LiveChatBot’s free plan: the widget, the AI Agent with the keyword bot as fallback, Flows, handovers, tickets and the Android app, for up to two websites with 100 visitors per website each month. AI-written replies include a small free monthly allowance per website. See the pricing page for the details, browse more chatbot examples by industry, and start your own build on livechatbot.in.

Frequently asked questions

Is Crumb & Co a real bakery?

No. It’s a fictional business used to show every step of a realistic build. The steps, settings and lessons apply to real small businesses.

How long does a build like this take?

About three hours of setup, mostly writing down facts, plus around twenty minutes a week to review conversations in the first month.

Do I need technical skills?

No. Adding the widget is a copy-paste or a plugin install. Everything else is writing and choosing settings.

What if my business is very different from a bakery?

The six pieces stay the same. Only the facts, the welcome line and the Flows change. Our industry examples show different patterns.

What’s the most important piece?

The facts. Every good answer comes from them, and nearly every bad answer comes from something missing.

Keep reading

Related guides

πŸ“˜

Customer Care Chatbot for Complaints: A Calm, Step-by-Step Intake That Customers Don’t Hate

Read the guide β†’

🧭

Chatterbot Meaning: Where the Word Came From, How It Differs From ‘Chatbot’, and Which Term to Use

Read the guide β†’

πŸ› οΈ

Chatbots Meaning Through History: How the Word and the Idea Changed From 1966 to Today

Read the guide β†’

Build your own in an afternoon

Follow the six pieces above on livechatbot.in. Free to start, no card needed.