ai customer support agent

live on the client sitehubspot ai agent · chat widget · hubspot crm

problem
Every customer issue went straight to the support team, even the ones a customer could fix in a few minutes with the right steps.
what i built
An AI agent in the chat on the client-facing website. It identifies the customer from their email, reads their company's service history in HubSpot, and troubleshoots with them. When it can't fix the problem, it opens a HubSpot ticket with everything that was tried.
result
Live on the client-facing website, where customers chat with it directly.

demo

support chat · AI agent

the screen in our lobby has been black since this morning

Sorry about that. What's the email address on your account?

ops@harbourfitness.example

Thanks. A screen at your site went black once before, after a firmware update. Is the small light on the media player blinking amber, or solid green?

solid green

Please unplug the media player for 10 seconds, then plug it back in. Does the screen come back within 3 minutes?

no, still black

I've opened ticket #4931 with everything we tried, so you won't need to repeat it. Our support team will follow up by email.

hubspot · what the agent sees

contact
found by email
company
Harbour Fitness
past tickets
2 · closest: black screen after a firmware update, solved by re-pushing the playlist
next step
guided flow: black screen

ticket #4931 · new · support team

issue
Lobby screen black since this morning
tried
Light is solid green. Power-cycled the player; still black.
history
Same symptom before; fixed by re-pushing the playlist.
handed over
The steps didn't fix it.
Illustration of the chat widget on the client-facing website, beside the HubSpot record the agent works from. The customer, email and tickets are made up.

how it works

  1. chat widget. A customer starts a chat on the client-facing website.
  2. identify. The agent asks for their email and finds the contact and company in HubSpot.
  3. if there's no match. It opens a ticket for the support team instead of guessing who it's talking to.
  4. service history. It reads the company's past tickets, so it knows what has gone wrong before and what fixed it.
  5. troubleshoot. It follows a guided flow for common issues, and searches the support knowledge base for everything else.
  6. if it can't fix it. It opens a HubSpot ticket with a summary of the chat and what was tried.
  7. resolved. The customer is back up and running, and no ticket is needed.

technical breakdown

built inside hubspot

The agent is HubSpot's own AI agent, running in the chat widget on the client-facing website. Building it inside HubSpot means it uses the same contacts, companies and tickets the support team already works from, with nothing to keep in sync.

identifying the customer

Before troubleshooting, the agent asks for the customer's email and matches it to a contact and company in HubSpot. It then reads that company's service history, so a problem that has happened before starts from what fixed it last time.

two sources of answers

when it hands over

The agent opens a ticket and stops when:

what the ticket contains

The ticket carries a summary of the chat and everything that was tried, so the support team picks up where the agent stopped and the customer doesn't have to explain the problem again.

stack

status

Live on the client-facing website of a digital-signage company, where customers chat with it directly. I don't quote usage or resolution figures here.

more projects