Set Up a Technical Support Bot: Start to Finish

This walkthrough builds one bot that answers general website questions, looks up orders, walks customers through fixes from your manuals, and hands off to your team. It uses the Troubleshooting Knowledge add-on for the manuals part. Moshky can walk you through any step: ask her "help me set up a support bot from my manuals."

The plan in one picture

You build two AI steps in one workflow. "Product & Orders" answers product, price, order, and general website questions. "Troubleshooting" answers how-to and fix-it questions from your manuals. Each step has a jump to the other, so the customer can move back and forth in one chat. Either step can hand the customer to a person.

Keep them as two steps. One large step with every collection mixed together pulls procedures into price answers and the reverse.

Step 1: Create your collections

Open AI, then Knowledge Setup, and answer the questions. The wizard creates three collections. Each one holds a different kind of content:

• General: your website pages, help articles, and product pages. Used for product, price, and order questions.
• Troubleshooting: the manuals themselves. The full text of each manual, kept in large pieces so a procedure stays together. Used for how-to and fix-it questions.
• Manual catalog: not the manuals. A short list with one line per manual: model, version, and public link. It is small on purpose. Both bot steps read it, so either one can answer "do you have a manual for this model?" and share the link without pulling in pages of procedure text.

Why keep them apart: mixing manuals into your website content makes a price question pull back repair steps, and the reverse. Separate collections keep each answer to the right source.

Your plan sets how many collections you can have. The wizard shows your limit before it creates anything.

Step 2: Scan and load your content

Website: start a scan of your site and publish it to the General collection. Leave Smart Scan on so a re-scan only re-reads pages that changed.

Manuals on your website: start a second scan. Set the start URL and path to where the manuals live, turn on "Follow linked PDFs," and publish it to the Troubleshooting collection. Pages that go to Troubleshooting stay out of General.

Manuals you have as files: upload them with their model and version. If they land in General, use the banner on the training page to move them to Troubleshooting without rescanning.

Manual catalog: add one short entry per manual with its model, version, and link. This is a list of your manuals, not a copy of them.

Image-only scans of paper manuals need text recognition first. Otherwise they index as blank.

Step 3: Build "Product & Orders"

Add an AI step in your workflow. In its Knowledge Sources, tick General and Manual catalog. Leave nothing else ticked. If nothing is ticked, the step falls back to the bot's default collection, which is not what you want here.

Add the skills this step needs: order lookups, customer lookups, and your store or CRM connection. Start with read-only skills and add ones that change data only when you need them.

In the instructions, say what the step covers (products, prices, orders, hours) and when to jump to Troubleshooting (any how-to or fix-it question). Add the jump to Troubleshooting.

Step 4: Build "Troubleshooting"

Add a second AI step. Tick Troubleshooting and Manual catalog. Give it no skills that change data.

Tell it to ask which model the customer has, answer only from the manual text it finds, give one procedure at a time, name the manual and share its link, and say so plainly when the manuals do not cover the question. Tell it to offer a person when the customer is stuck. Add the jump back to Product & Orders for price, order, and product questions.

Step 5: Choose where each visitor starts

Use the Conversation Opened trigger. Add a Condition on the page URL. A visitor on your support or manuals pages starts in Troubleshooting. Everyone else starts in Product & Orders. The bot can still jump either way once the chat is going.

To route by group, add a Condition on a contact attribute or tag, for example existing customer or plan level, and send that group to a specific team, persona, or step.

Note the difference between the two triggers. Widget loads on page only controls whether the chat button shows and which chat window design appears. It runs before the visitor starts chatting, so it cannot start the bot or pick a team.

Step 6: Hand off to your team

From either AI step, add a transfer to a team. Use Route by Intent to choose the team, such as Sales, Support, or Billing, after the bot finishes. Add branches for business hours and availability so an after-hours request leaves a message or opens a ticket instead of waiting in an empty queue.

Step 7: Test before you go live

Run these in a test chat, in order:

1. Ask a price question. The answer should come from your website, with no manual text.

2. Ask how to clean or fix a specific model. The answer should be step by step and name the manual with a link.

3. Ask "do you have a manual for this model?" from both steps.

4. Ask a fix question, then a price question, then a fix question. Confirm each answer uses the right source.

5. Open a support page and a product page, start a chat on each, and confirm where the bot starts.

6. Ask it to change a therapy or device setting. It should give the fixed referral and no advice.

7. Ask for a person during and after business hours.

8. Ask something your manuals do not cover. It should say so and offer a person.

After a jump, check that the bot answers the question that caused the jump instead of asking it again.

What the bot will not do

It will not change or recommend changes to therapy or device settings. It sends those requests to your customer's doctor or equipment provider, and it tells anyone describing an emergency to call 911 first. See the Troubleshooting Knowledge guide.