Winnfield Chat: AI-first customer support
Winnfield Chat is an optional add-on, included in some plans, which aren’t published yet. It’s a support chat you add to your app or website, and it handles the conversation with your signed-in customers. Its AI answers them, and when it needs a person, the conversation becomes a support ticket for your team in Winnfield Chat’s inbox.
Like other AI support tools, its AI answers from your documentation. It also goes two steps further. When the docs don’t cover a question, Winnfield itself checks that customer’s settings and account, and then your code, read-only. Winnfield Chat’s AI never sees your code or data.
You don’t have to use it. You can keep your current help desk and have your team ask Winnfield in Slack.
Example with a fictional company
Three tiers, then a person
Winnfield Chat goes only as deep as a question needs. If one tier can’t answer, the next one tries. If none can, a person on your team takes over.
Each tier below shows a case from Larkspur, a fictional company whose AI booking assistant answers its guests.
- Tier 1 · Your documentationWinnfield Chat’s AI answers from your docs
- Like other AI support tools, it answers from your approved help articles and cites them.
- It has no tools and no path to your code or data.
- At first it only drafts, and your team sends or edits each draft. It answers customers directly only on the topics you turn on. Topics are the subjects your staff tag in your help articles, such as checkout times.
- When it’s unsure, it says it’s checking and tells your team.
Example A guest asks what time checkout is. It answers “11 AM, unless you’ve booked late checkout” and cites Larkspur’s Checkout times article.
- Tier 2 · The customer’s own settingsWinnfield checks the customer’s own settings and account
- When the docs don’t cover it, Winnfield itself, not Winnfield Chat’s AI, looks at that customer’s records and settings, read-only.
- The reply to that customer uses only that customer’s records, the ones you’ve allowed, and never your logs.
- Winnfield’s investigation for your own team, such as when your staff ask in Slack or in an internal note, can look across other conversations, records and logs, read-only. What it finds stays internal.
Example Jordan asks why he was charged $45. Winnfield reads his booking, LS-48213, and the fee setting, and finds that the $45 fee has applied since October 14.
- Tier 3 · Your codeWinnfield reads your code to find the cause
- When the cause is still unclear, Winnfield reads your code, read-only.
- If it’s a bug or a missing feature, it files it in Jira, once your admin has turned on the File stage.
Example Larkspur’s booking assistant had also told Jordan late checkout was free. Winnfield finds that its instructions still quote the old policy, and files bug LSP-1294 for Larkspur’s engineers.
- Hand-off · A person on your teamThe conversation becomes a support ticket for your team
- Winnfield writes what it found in an internal note, which only your team sees.
- Your team writes the reply itself, or approves a draft made from the part of the note marked safe to share, and Winnfield Chat sends it. On topics you’ve turned on, Winnfield Chat sends that draft without approval.
- Your team sees every conversation and can take over any support ticket.
Example Jordan’s conversation becomes support ticket 1043 in Winnfield Chat’s inbox. Maya reads Winnfield’s note and approves the reply.
When the docs aren’t enough
When Winnfield Chat’s AI can answer from your docs, on a topic you’ve turned on, the answer goes straight to the customer. When it can’t, the question follows the path below.
Winnfield Chat’s AI has no access to your code, database or logs, so Winnfield investigates and writes an internal note for your support team. Winnfield Chat then sends the customer the part of the note marked safe to share, after your team approves it or, on topics you’ve turned on, automatically.
Example Jordan’s question reaches Winnfield, which writes its answer for Larkspur’s team in an internal note. Maya approves the reply, and Jordan gets the answer in Winnfield Chat.
- Asks in Winnfield Chat
- Can’t answer from your docs, so asks Winnfield
- Writes a note only your team sees
- Writes or approves the reply
- Gets the answer
Sees every conversation and can step in at any point
Your team writes or approves each reply that comes from Winnfield’s notes, until you let Winnfield Chat send those replies on its own for topics where its drafts have proven reliable.
A bug or a feature request? Once your admin turns on the File stage, Winnfield files it in Jira. A bug is filed with the evidence, and a feature request becomes a story, a Jira ticket with acceptance criteria, for your product team. Someone on your team can also ask Winnfield a question from the same conversation, in an internal note.
A knowledge base that Winnfield writes and you approve
Winnfield drafts help articles for your customers from your code and documents. Only the versions you approve reach Winnfield Chat’s AI.
Example Larkspur’s Checkout times article, which Winnfield Chat cites, would be drafted this way and approved by Larkspur’s team.
- SourcesYour repositories, Confluence and uploads
- DraftsArticles written for your customers
Internal paths, IDs and Jira ticket keys are blocked
- You approveReviewed as a diff
Nothing is published automatically
- Tier 1Winnfield Chat’s AI answers and cites them
When your code changes, Winnfield proposes updates
Your team stays in charge
Your people work in Winnfield Chat’s inbox, in Winnfield’s web app, where every conversation appears. Internal notes from Winnfield and replies to customers are kept apart, and you can pause Winnfield Chat’s AI. Replies drawn from Winnfield’s notes go through three reply modes.
- You choose which topics Winnfield Chat may answer without approval. A topic qualifies only after at least 100 drafts over two weeks, with 95% of them sent unedited and none marked wrong.
- Your staff review a sample. They check 5% of the replies sent without approval, and at least one a day, after the replies go out.
- Any problem puts approval back. A reply marked wrong, a report that a reply shared something it shouldn’t have, or a change to what customers may be told returns that topic to drafts at once.
- You can make these rules stricter, but never looser.
Example At first, Maya writes or approves every reply to Larkspur’s guests. Once a topic’s drafts have earned it, she can let replies on that topic go out without approval.
- Internal onlyYour team reads Winnfield’s note and writes the reply.
- DraftA reply is drafted from the note. Your team approves each one.
- Sent without approvalFor the topics you choose, once their drafts have earned it.
Jordan Ellis · support ticket 1043
Pause chat AIFile in JiraEach customer’s answers use only their own records
You approve a map, a list of the records and fields each signed-in customer’s answers may use. A reply to a customer uses only the rows and fields you’ve allowed for that customer, and it can also draw on your code, configuration and approved articles. Winnfield’s investigation for your own team can look wider, read-only, and what it finds stays in the internal note.
Example Jordan’s answers may use his own bookings, such as LS-48213, but not another guest’s booking or Larkspur’s payouts to hosts.
How Winnfield Chat and your data are kept apartWhen it’s a bug, a setting or a feature request
After reading the evidence, Winnfield works out whether the cause is in your agent, in the software around it, in a wrong setting, or a missing feature. A wrong setting goes back to your support team with the change to make, and they can fix it without an engineer.
Once your admin turns on the File stage, Winnfield files a bug in Jira with its evidence, and a request for something your product doesn’t do yet becomes a story for your product team. When someone assigns the Jira ticket to Winnfield, the fix or the feature follows, with its tests and a pull request. The Jira ticket’s key, such as LSP-1294, never reaches the customer.
Example Jordan’s question became bug LSP-1294, and Ana’s request to split a payment became story LSP-1301.
See one bug from start to finishBooking assistant quotes the old late-checkout policy
Impact: the booking assistant has quoted the old late-checkout policy in 37 conversations since Oct 14
Questions
All questions about Winnfield ChatDoes Winnfield Chat’s AI see our code or data?
No. Winnfield Chat’s AI sees approved articles and navigation links, messages that customers can see, and user details your product supplies. It never sees internal notes. Only Winnfield’s read-only investigation touches your code and data.
Can Winnfield write to our customers?
Only through Winnfield Chat. Winnfield writes internal notes for your team, which your customers never see. Winnfield Chat sends a customer the part of a note marked safe to share, after your team approves it or, on topics your staff have turned on, automatically, after the required checks. Topics are the subjects your staff tag in your help articles. Your staff can also reply directly.
Who can use Winnfield Chat?
Your signed-in users. Your backend signs a short-lived token for each one, and the chat can be embedded only on the sites you register.
Join early access as a design partner and bring us your hardest escalation.
A design partner is one of the first companies to use Winnfield, before it launches. An escalation is a customer problem your support team had to pass to engineers. Requesting early access is how you apply. The founder leads your onboarding, and requesting access commits you to nothing.