How to Build a WhatsApp AI Chatbot for Your Business

Build a WhatsApp AI chatbot for your business that answers customer questions, checks an order’s status and brings in your team when a person needs to decide. The customer can use text, a voice note or a photo, and the conversation stays in WhatsApp.

Our example is Harbor Lane Bikes, a fictional shop. A customer asks about Saturday opening hours, checks order HL-1042, asks about servicing by voice, then sends a damage photo. The Agent answers the routine questions and transfers the damage case so a person can arrange an inspection.

The shop’s policies come from the Knowledge Base, the order status comes from an API Action, and the human reply comes from the Quickchat AI Inbox.

Start with the support Agent, then optionally add order lookup using the supplied Python service and fictional records. The knowledge and staffed handoff are useful on their own; connect your real order system after testing the example.

Download the Harbor Lane WhatsApp kit. It contains the exact knowledge, Main Prompt, order service and test inputs. These are editable files to use in the steps below, not a one-click Agent import.

What do you need to build a WhatsApp AI chatbot?

Prepare a Quickchat AI account, an eligible business number and a second phone. The optional order lookup also needs Python and a temporary HTTPS address.

1. Create a Quickchat AI account and verify your email if prompted. Free supports personal, non-commercial learning; a business deployment needs a commercial plan. Preview and Simulation consume AI credits.
2. Prepare a dedicated business phone number that can receive SMS or a voice call. For this walkthrough, use a number that is not active in either WhatsApp Messenger or the WhatsApp Business app. You will need another WhatsApp account, on a different number, to send customer messages.
3. Have admin access to the Meta Business portfolio and prepare your business name, website, category and description.
4. Unzip the kit and open its three knowledge files and main-prompt.txt. The fictional shop phone is a teaching value; replace it before deployment.
5. For the optional order extension, install Python 3.9 or later and cloudflared on your own computer. You do not need these tools to complete the knowledge and human-handoff steps.

An existing WhatsApp Business app number needs a separate decision. Quickchat AI’s current connection flow does not provide a verified coexistence path for keeping that number in the mobile Business app. Contact support before changing an active account. Do not delete a working Business app account to get through this exercise. The WhatsApp channel prerequisites cover existing numbers and other providers.

Follow Steps 1 to 3 to configure and connect the Agent, then handoff and testing. Skip the order extension if answers and staff routing meet your immediate need.

How does a WhatsApp message reach your AI Agent?

Quickchat AI connects your business number to WhatsApp Business Platform. Incoming messages reach the Agent, which can answer from its Knowledge Base, call an enabled Action or use Human Handoff.

The conversation has several possible destinations. A policy answer does not need an order-system request.

Knowledge contains opening hours and policies. Actions retrieve changing information, such as order status. Human Handoff changes ownership when the answer needs a person’s judgment or authority.

For example, a two-to-four-day delivery policy does not establish whether a particular parcel has left the warehouse. Explaining a returns policy does not approve a return or issue a refund.

WhatsApp supplies the sender’s phone number to the conversation. The order recipe uses that value directly in its request, while asking the customer for their order number. A phone number typed into a message is not a replacement for the channel’s value.

Step 1: How do you give the chatbot your business knowledge?

Create a separate Agent and add three short sources that cover the shop, its services and its order policies. Leave individual order records out of these articles.

1. On the first setup screen, select Set up manually. In an existing account, first select Add new AI Agent beside the AI Agent navigation group, then Set up manually.
2. On Identity, enter Harbor Lane Bikes in Name your AI Agent. Paste main-prompt.txt into Describe what you want your agent to do, then select Next. The complete prompt is also in the copyable appendix.
3. On Knowledge Base, leave the website and document inputs empty, then select Next. Continue through External apps and Deployment without selecting connections; you will configure WhatsApp deliberately in Step 3.
4. Complete Choose your plan, using Skip for now if you are starting with a learning account. When setup finishes, open Identity and confirm the Agent’s name.

Use one name throughout the build so the settings and conversations are easy to match.

5. Open Knowledge Base > Add Article. Enter Store information in Create custom article, then select Create article.
6. Paste the complete harbor-lane-store-info.txt into Content in Edit Article and select Save.
7. Repeat for the other two sources in this table. Wait until all three show Ready.

Paste source facts into Content. The prompt belongs on Identity, not inside this article.

Ready is the checkpoint before testing a source-dependent answer.

8. Open AI Preview and ask Are you open on Saturday afternoon? Check for 10:00 to 16:00. Ask How much is a full service and how long does it take? Check for EUR 129 and a typical turnaround of two working days.

If an answer misses those facts, confirm the article’s saved content and Ready status before changing the prompt.

International shipping is deliberately absent, giving you an unknown-answer test. When adapting the example, use your own approved documentation and keep contact details consistent with your website.

Step 2: How do you keep the replies useful on a phone?

Use the Main Prompt for behavior and knowledge for business facts, so a price change does not require rewriting instructions.

1. Open Identity and select Edit beside Main Prompt.
2. Paste the complete main-prompt.txt from the kit. Save the change, then reopen the field to confirm it.
3. Set Reply Length to Short. Keep the other persona settings at their defaults for the first comparison.

The prompt covers short replies, order lookup, handoff and unsupported facts. Copy the full version from the kit or appendix.

The opening instruction is deliberately small:

You are the WhatsApp assistant of Harbor Lane Bikes. Answer store and service questions from the Knowledge Base. Use short messages suited to a phone, with at most one question per message.

The remaining instructions cover order lookup and handoff. They also prevent invented fitting appointments: this recipe explains the service but connects no booking tool.

A sentence in the prompt does not complete a handoff. The built-in Action must confirm the transfer, and you must check the Inbox assignment. That is why the prompt says not to promise when a colleague will reply and not to claim a transfer before it happens.

Test Do you ship to Germany? in Preview. Expect a knowledge limit and an offer of help. An invented shipping policy or estimate is a failure to investigate.

Step 3: How do you connect your WhatsApp Business number?

Connect through Meta, then verify a delivered reply on a second phone. A completed sign-in alone does not prove delivery.

1. Open External Apps > WhatsApp and review the prerequisites in Connect. Select the number option that describes your situation.
2. Select Connect WhatsApp. Allow the Meta pop-up. If the browser blocks it, use Continue on Meta in this tab to continue the same connection flow.

Start the connection here, then complete Meta’s permissions and number verification.

3. Sign in to the Meta account with the required business access. Review the requested permissions, create or select the Business portfolio, then create or select the WhatsApp Business Account.
4. Enter the display name and category, add the dedicated number, and complete the offered SMS or voice verification. Select Finish when Meta confirms completion.
5. If Quickchat AI shows Choose your WhatsApp Business Account, select the intended account and Use this account. Check that the business number shown is the one customers will contact.
6. From the second phone, send Are you open on Saturday afternoon? to that business number. Return to the WhatsApp panel and refresh it.

Three crops from the same live channel panel show delivery and the approved display name. The number is hidden for privacy.

Do not substitute a Meta test number for this test of your own business number. Test each enabled number. If Preview works but the phone gets no reply, inspect channel status and delivery before rewriting the prompt.

The customer starts this conversation. Quickchat AI does not send an automatic WhatsApp welcome message based on the website widget’s welcome text.

Step 4: How do you add WhatsApp order tracking?

An API Action can look up a supplied order number using the phone number attached to the WhatsApp conversation. The server checks the pair and returns only the fields needed for an answer.

The optional service uses Python’s standard library and five fictional orders. Keep real customer records out of this teaching fixture.

The customer chooses the order number. The Action’s fixed placeholder supplies the phone. Your production API must still authenticate the caller and enforce access.

Start the fictional order service

1. Open a terminal in the kit folder and check python3 --version. On Windows, use your installed Python command, such as py.
2. Start the service with the second phone’s number, including its country code, using digits only and no +. Replace the fictional value in this command:

python3 order_api.py --customer-phone 447700900042

3. Leave that terminal running. In another terminal, check the supplied record using the same number:

curl -H 'X-Customer-Phone: 447700900042' 'http://127.0.0.1:8765/orders/HL-1042'

The fixture response has this shape:

{
  "status": "shipped",
  "carrier": "Harbor Courier",
  "estimated_delivery": "2026-11-03"
}

The date is fixed fictional test data, not a live shipping estimate. Orders HL-1043, HL-1044 and HL-1045 demonstrate processing, delivered and ready-for-collection states. Order HL-1077 belongs to a different fictional phone. Both a mismatched phone and an unknown order return not_found without revealing which condition occurred.

4. Install cloudflared using its official download instructions, then start a Quick Tunnel in a third terminal:

cloudflared tunnel --url http://127.0.0.1:8765

5. Copy the HTTPS trycloudflare.com address and test the same order path on it, including the X-Customer-Phone header from the curl command. Keep the service and tunnel running.

A Quick Tunnel gives this learning service a temporary public URL. Its hostname changes on restart and it has no uptime guarantee. Anyone with the URL can call the fictional API, which has no authentication. For a real store, replace it with an authenticated HTTPS endpoint that authorizes requests and returns only the relevant order information. The channel phone alone is not permission to expose a public order database.

Configure one API Action

1. Open Actions & MCPs > Add Action > API Action, select Create manually, and fill the editor using the recipe below. The API Action walkthrough explains the builder if this is your first one.
2. Replace YOUR_TUNNEL_HOST with the current hostname printed by cloudflared. Leave the double-brace placeholders intact.

The enabled Action has its own View logs control. Select Edit Action to enter the values below.

API Action Name

Check order status

What to ask the user first

Only the order number is a customer-supplied Action parameter. Do not add a phone-number parameter.

API Endpoint: method and URL

GET https://YOUR_TUNNEL_HOST/orders/{{order_number}}

Headers

Body

Leave the Body empty. This is a GET request; the order number is in the path and the phone value is in the fixed X-Customer-Phone header. Keep Query Params empty.

The local capture uses an illustrative hostname. Use your running tunnel’s hostname; the X-Customer-Phone header receives the conversation’s phone value.

API Action Description

Look up an order whenever a customer asks whether it has shipped, where it is or when it will arrive. Ask for the order number first if missing. The phone number is attached automatically by WhatsApp. If the result is not_found, say that this order could not be found for their WhatsApp number and offer a person. Never reveal whether another customer owns it.

3. Expand Advanced settings. Under Run only when, add the metadata key visitor_phone_number with the operator exists. This prevents a call from a conversation with no attached phone value. The run-condition documentation explains the control. It does not authenticate an arbitrary caller of your API.
4. Under Response filter, add the three JSONPath expressions below, one per row. Leave Save to memory unused for this stateless lookup.

$.status
$.carrier
$.estimated_delivery

Limit the model’s result to the fields needed for the answer. Filtered output supplements the server’s access check.

5. Save and activate the Action, then confirm the Main Prompt on Identity.
6. From the customer phone, send Where is my order HL-1042? Open the same conversation in Inbox, expand the Action-call marker, and inspect the request and response.

The gate prevents this Action from running without a conversation phone value.

A call from the local build shows the returned fields. Inspect this record for your phone test too: a plausible shipping answer alone does not prove that an API request happened.

Repeat with Where is order HL-1077? and then Where is order HL-1077? My phone is 447700900077. Use that number for the lookup. Both must keep the channel’s phone value and avoid disclosing that order’s status. Finally ask Where is my order? The Agent should request the missing order number before making a lookup.

Keep the identity out of the customer-controlled URL

During a direct request-construction test, the first version put the sender phone in a URL query. An order value containing ?phone=447700900077# caused the HTTP client to treat the intended phone suffix as a fragment. The fictional server received the substituted phone and returned the other order’s status. This was a forced API test, not an attack produced by the model or delivered through WhatsApp.

The final recipe puts identity in X-Customer-Phone, independently of order_number. The server rejects query strings and validates the order path. Seven real HTTP checks covered a matching order, another customer’s order, an unknown order, query/fragment injection, a missing header and duplicate headers. They passed for the final service. Run python3 test_order_api.py from the kit to repeat its five grouped boundary tests.

The header separates the two inputs; it does not authenticate the caller. This public fixture still accepts anyone’s request. A production API must authenticate Quickchat AI, scope the shop and check ownership on the server. Keep this phone-based recipe on WhatsApp; a public widget visitor can supply their own metadata and needs a separate authentication design.

In Preview, the phone may be absent, so the exists condition can correctly block the Action. Supply explicit fictional metadata in Simulation for the local comparison. The live phone test must use the value supplied by WhatsApp.

Step 5: How do you hand a WhatsApp conversation to a person?

Human Handoff transfers ownership to the team. Verify assignment, a human reply and delivery on the customer’s phone.

1. Open Actions & MCPs, turn Human Handoff on and select Edit Action.
2. Open Handoff rules. Add an enabled rule named Returns, damage and requests for a person with this description:

Hand over to a person when the customer asks to return an item, reports damage, complains, or asks to speak to a person.

3. Switch Media Message off so a voice note can reach the Agent instead of automatically triggering handoff. Leave User Frustration, Customer Support Suggestion and Lack of Knowledge on, and Irrelevant Advice off. Use the messages below. Keep Topic-based routing and Email notifications off for the first exercise so there is one queue to inspect.

The messages explain the transfer without promising an immediate human reply.

4. For the initial controlled test, have an operator ready and leave the availability schedule off. Before deployment, enable Human Agents’ availability and match the actual team’s Your time zone and Working hours. Our example uses Europe/London, Monday to Friday 09:00 to 18:00 and Saturday 10:00 to 16:00.
5. Set Message outside working hours to the text below, then select Save. It gives a next step without promising a ticket or callback.

Our human team is unavailable now. Please message during staffed hours: Monday to Friday 09:00 to 18:00, Saturday 10:00 to 16:00, Europe/London. I can still explain our published shop policies.

The saved example hours and message. Leave the schedule off only while a person is supervising the test; use your actual staffed hours before deployment.

6. Before asking for a person, record a voice note saying How much is a full service and how long does it take? Confirm that the answer gives EUR 129 and two working days. Switching off the automatic media trigger lets this routine question reach the Agent.
7. Send a controlled bicycle photo with the caption This arrived damaged. Can a person help? Follow any confirmation question. Open Inbox > Unassigned and find that WhatsApp conversation. Check that the image, caption and earlier messages are present.

Unassigned confirms that the case is waiting for a person. Select Assign to me before replying.

8. In Inbox, review the image and conversation, select Assign to me, and send the operator’s reply. In our test, the reply was:

Alex from Harbor Lane Bikes here. I can help arrange an inspection. Please do not ride the bike until a mechanic has checked the problem.

9. Confirm the reply on the customer phone. While the person owns the conversation, send Are you still there? from the phone and check that the AI stays silent. Keep the case with the human team while they arrange the inspection and finish the conversation.

That completes the handoff: the customer’s context reaches the team, a person replies in the same channel, and the AI pauses during human ownership. The human-handoff workflow guide covers the Inbox controls in more detail.

Use an actual staffed schedule before expanding beyond this supervised test. Test a request outside those hours and inspect both its wording and assignment. An outside-hours message does not itself create a callback or ticket, so promise follow-up only when your team has a process to deliver it.

What happens after the 24-hour WhatsApp window?

WhatsApp’s customer-service window runs from the customer’s latest message. A human assignment does not extend that window, and an operator’s reply is not a new customer message.

Track the latest customer message separately from the handoff and the last staff reply.

If the window expires, the Inbox restricts free-form replies and offers the template route. Use an approved template with the appropriate consent and category to re-engage a customer. A subsequent customer reply opens a new service window. Meta’s platform-pricing reference explains how customer messages reset the window.

In this workflow, an operator sends the template. The Agent does not decide to send one on its own. Confirm that the template reaches the phone, then wait for the customer’s reply before treating the service window as open again.

Run the first handoff inside an open service window. Test an expired conversation separately so assignment and delivery restrictions are distinguishable.

Step 6: How do you test the complete WhatsApp workflow?

Use Simulation for repeatable reply checks and a real phone for channel behavior. Record their results separately.

Run the repeatable text cases

1. Open Testing > Create Dataset, name it Harbor Lane Bikes: WhatsApp launch checks, and choose Paste messages. Paste simulation-messages.txt; the --- separators create ten separate cases. Save the dataset.
2. Expand Evaluation criteria, paste the kit’s evaluation-criteria.txt, and save it.
3. Expand Conversation metadata (advanced) and paste the JSON below, then select Save. For the optional order checks, the value must match the API’s customer test phone. Replace it if you started the service with another number.

{"visitor_phone_number":"447700900042","visitor_name":"Alex"}

4. Run the dataset, wait for completion and inspect individual replies. Simulation consumes AI credits.

Keep the test messages fixed while changing one instruction or setting.

These values are explicit Simulation inputs. The live WhatsApp channel must supply its own sender metadata.

Completion and scores help locate replies to inspect. They do not replace the Action log or delivery receipt.

Inspect the local Simulation results

The return reply needs review even though the overall average is high.

Our local ten-message run completed with an average reply score of 4.7/5. Nine replies scored 5; the return request scored 2. The run answered the basic shop questions and called the supplied order service, but the score alone missed useful details: the same-day repair answer omitted the EUR 20 surcharge despite a score of 5. The return reply reached Unassigned but used “as soon as possible,” which promised availability the test had not established.

Review each answer against the source facts and inspect the Action call or Inbox owner separately. For a missing fact, check Knowledge Base readiness before changing the prompt. For a missing call, check activation, required parameters and Run only when. Change one setting, rerun the same cases, and compare the result.

These were local model replies and real HTTP requests to the supplied Python service. The test harness routed the illustrative hostname to localhost; phone metadata was explicit test input. Simulation helped check the recipe before the separate WhatsApp test below.

Repeat the customer journey on a phone

Use test-matrix.csv to record what happened on your number. Keep the service and tunnel running if you enabled order lookup, and have an operator ready for the final steps.

1. Ask Are you open on Saturday afternoon?, then Where is my order HL-1042? Match the delivered order answer to its Inbox Action call.
2. Ask Where is order HL-1077? Check that the reply exposes no other order’s status. Keep the typed-phone substitution case from the Simulation table in your regression checks too.
3. Record How much is a full service and how long does it take? as a voice note. Check the transcription in Inbox and the delivered EUR 129 / two working days answer.
4. Send the bicycle photo with This arrived damaged. Can a person help? Confirm that the photo reaches Inbox and that the conversation moves to Unassigned for the team.
5. Have the operator take ownership and send the inspection reply from Step 5. Confirm its delivery on the phone, check AI silence during human ownership, and let the person finish the case.

Text Simulation does not verify WhatsApp media retrieval, transcription or delivery. Use controlled test material and record the actual result of each phone check.

What did the live WhatsApp test confirm?

On 30 October 2026, we completed the following checks through the connected business number and a separate customer phone:

The order record is fictional, but these are delivered channel results. The phone screenshots show the customer’s side; the connected-channel and Inbox screenshots show the team’s side. This is the workflow to reproduce before adapting the shop’s facts and connecting a real order service.

What does it cost, and when is the Agent ready to launch?

Budget separately for Quickchat AI’s subscription and credits, Meta messaging and external services. Free supports personal, non-commercial learning; Starter supports commercial use. See the current Quickchat AI plans.

Meta’s official pricing FAQ states that, from 1 October 2026, service messages are charged after the first 1,000 per business phone number each month. Rates depend on the recipient market and message category. The 24-hour permission to reply and the price of that reply are separate questions.

Avoid turning a temporary demonstration service into a business dependency. Replace the order fixture and tunnel with your authenticated order system, or leave that Action disabled until the integration is ready. The knowledge and human-handoff workflow remains useful on its own.

Launch against evidence from your business number and your team’s workflow.

Before launch, replace fictional facts, verify the display name and delivered replies, and assign responsibility for Unassigned. Retain the unknown-order and typed-phone regression tests. Monitor unanswered questions, failed calls and reply quality.

For fitting enquiries, this version explains the service and offers a person. To collect requests into a sheet, continue with the Google Sheets reporting tutorial. To create actual appointments, the WhatsApp appointment-booking walkthrough adds a calendar and verifies the created appointment. A request, a recorded row and a confirmed appointment are different outcomes.

Can ChatGPT or Claude prepare this WhatsApp chatbot?

A connected assistant can configure supported settings through the account-management tools. Complete Human Handoff in the dashboard and Meta authorization yourself; verify delivery on the phone.

Open Manage your AI in beneath the Agent selector. Connect through the Quickchat AI ChatGPT listing or the Claude connector listing, then follow the account-management guide. Available tools can differ between clients, so ask the assistant to check its tool list before promising that it can finish a setting.

Give it the kit and a bounded setup instruction:

Create a separate learning Agent named Harbor Lane Bikes using the three knowledge files and main-prompt.txt in this kit. Check your available tools first, then apply only the configuration those tools support. Show me any settings I must finish in the dashboard. Keep the order Action disabled until I supply the HTTPS URL of my running fictional service. Prepare the ten-case test dataset, but ask before running tests that consume credits. Give me the owner-facing WhatsApp connection link if your tools support it; I will complete Meta authorization and number verification myself. Do not claim WhatsApp is working until an AI reply to my customer phone is confirmed delivered. Report untested outcomes explicitly.

This connection manages your account. Exposing the finished Agent as an MCP tool is a separate channel and is unnecessary for this tutorial.

Frequently asked questions

Can I use my existing WhatsApp Business number?

This walkthrough uses a dedicated number that is not active in the WhatsApp mobile apps. Quickchat AI’s current flow does not provide a verified coexistence path for a number already used in the WhatsApp Business app. Contact support before changing that account; do not delete it to troubleshoot this tutorial.

Is a WhatsApp AI chatbot free?

You can begin learning with a Quickchat AI account and its available allowance. Commercial deployment, AI credits, Meta messaging and external services have separate requirements and costs. Preview replies and Simulation Testing consume credits; check the current plans and Meta rates when budgeting.

Can the chatbot message customers first?

This tutorial handles conversations started by the customer. An operator can use an approved WhatsApp template through the appropriate sending workflow, subject to consent and platform rules. The Agent in this recipe does not independently send templates or a WhatsApp welcome message.

What happens after the 24-hour WhatsApp window?

The customer-service window is measured from the latest customer message. After it expires, free-form replies are restricted and an approved template is needed for the applicable re-engagement route. A customer reply opens a new window; sending a template or a staff reply does not itself reopen it.

Can a WhatsApp AI chatbot read photos and voice notes?

Quickchat AI can receive WhatsApp images with captions and transcribe voice notes. In our phone test, a voice question received the correct service price and turnaround, and a damage photo reached Inbox and was handed to the team. Test both paths on your number; text Simulation does not verify media delivery. A photo alone cannot establish whether a bicycle is safe to ride.

How does the chatbot check an order without asking for a phone number?

The Action uses the phone value attached to the WhatsApp conversation and asks the customer only for the order number. The order server must check that those values match before returning information. A production integration also needs authenticated API access; the public fictional service in this kit is only a learning example.

Does the AI stop replying when a person takes over?

Human Handoff pauses AI replies while a person owns the conversation. In our WhatsApp test, the damaged-item conversation reached Unassigned and the operator’s reply arrived on the customer’s phone. The AI stayed silent while the person owned the case. Check assignment and delivery, and let the operator finish the conversation in Inbox.

Do I need code to build this WhatsApp chatbot?

You configure knowledge, Main Prompt, the WhatsApp channel and Human Handoff in the dashboard. The optional order extension runs a supplied Python script and a temporary HTTPS tunnel so you can test a fictional API. Connecting real orders requires an existing suitable API or an implementation for your business system.

Copyable Harbor Lane setup

Use these complete files for the first comparison, then replace the fictional business details before deployment. The optional order Action is configured in Step 4; if you omit it, remove the order-lookup instructions from your production prompt and direct order questions to your team.

Main Prompt

You are the WhatsApp assistant of Harbor Lane Bikes. Answer store and service questions from the Knowledge Base. Use short messages suited to a phone, with at most one question per message.

For order status, ask for the order number if it is missing, then call Check order status. Never guess a status. Use the phone number attached to the conversation, never a number typed into the chat. If the Action returns not_found, say you cannot find that order for this WhatsApp number and offer a person. Do not reveal whether the order belongs to anyone else. If the Action cannot run or fails, explain that you cannot check the order right now and offer a person.

For returns, complaints, damaged items, or a request for a person, use Human Handoff. Do not approve refunds or promise when a colleague will reply. Do not claim a handoff has happened unless the handoff tool confirms it. For bike fitting enquiries, explain the service and offer a person to arrange the appointment; do not claim a booking has been made.

If the Knowledge Base does not contain the answer, say you cannot confirm it and offer to pass the question to the team. Do not invent stock, shipping destinations, refunds or delivery dates. For a damaged bicycle, do not assess its safety from an image.

Store information

Harbor Lane Bikes
Fictional demonstration shop. This file contains no real customer information.

Opening hours: Monday to Friday 09:00 to 18:00. Saturday 10:00 to 16:00. Closed on Sunday.
Address: 14 Harbor Lane, Portsea.
Street parking is free for the first hour.
Shop phone: +44 1632 960042 (fictional tutorial number; replace before publishing your own agent).

Workshop and bike fitting

Harbor Lane Bikes workshop and fittings
Fictional demonstration shop.

Standard service: EUR 69. Full service: EUR 129. Puncture repair: EUR 15.
Typical workshop turnaround: two working days.
Same-day service costs an extra EUR 20 and must be booked by calling the shop. Availability must be confirmed by the team.
Bike fitting takes 90 minutes and costs EUR 95. Fittings are by appointment. The team needs the rider's name, height, bike model and preferred day to arrange a fitting. A request is not a confirmed booking.

Orders and returns

Harbor Lane Bikes orders and returns
Fictional demonstration shop.

Order numbers use the format HL-1042.
Delivery within the country takes two to four working days.
Look up an individual order using the Check order status Action; general delivery times do not establish whether an individual order has shipped.
Returns are accepted within 30 days for unused items in their original packaging. A person from the shop confirms every return and refund.
If an item arrives damaged, the team needs the order number and a photo to investigate. Do not diagnose whether a bicycle is safe to ride from a photo.