Subject: My promo code won't apply at checkout
Hi — I'm trying to use SPRING20 on my order but checkout keeps saying the code isn't valid. It's not expired, I got the email yesterday. What am I doing wrong?
— Maya
Now recruiting design partners · Launching this fall
AI-native customer support that enhances your team, doesn't replace it.
Starling triages every ticket, finds the right answer in your knowledge base, and drafts — or sends — replies, all under your control.
See it work
Watch Starling handle a ticket.
Product demo showing three tickets in the Resolve agent workspace.
- Drafts for approval: a customer's promo code won't apply at checkout. Starling reads the ticket, finds the matching knowledge base article, and drafts a reply grounded in it. The suggested reply waits for an agent to apply and send it.
- Replies on its own: a customer asks about weekday and weekend support hours. Starling finds the answer in the knowledge base, its confidence (0.86) clears the configured threshold (0.75), and it replies automatically. The ticket moves to Awaiting customer.
- Holds for a human: an upset customer reports a duplicate charge and threatens a bank dispute. Confidence (0.66) is below the threshold and the ticket is a high-risk billing dispute, so Starling doesn't send anything. Its draft waits in the approval queue, and the ticket shows "Starling has a draft pending" until a person reviews it.
Ready for your approval — nothing goes out until someone on your team sends it.
What you just watched.
Three things Starling does on every ticket — and why your team stays in charge of all of them.
Grounded, not made up.
Starling answers from your knowledge base — the articles your team already trusts. The related articles sit right beside every suggestion, and Starling won't send anything on its own that it can't ground in one.
- Related articles shown with every suggested reply
- No knowledge-base match, no automatic send
You decide how much it does.
Pick a mode for your team and change it whenever you like.
- Off
- No AI at all.
- Run assist
- Analyzes every ticket and drafts replies for your agents. Never sends.
- Auto-reply
- Sends on its own when a draft clears every safeguard. Human review is on by default, so drafts wait for approval until you switch it off.
One setting for your whole team, with a confidence threshold you control.
It knows when to step back.
When confidence is below your threshold, or a ticket touches a billing dispute, a legal question or account security, Starling doesn't send. The draft waits for a person, and the ticket shows it's pending so nobody doubles up.
- High-risk topics are always held for a person
- Once someone on your team replies or takes over, Starling stops
In Command
Every held reply lands in one queue.
With human review on, every draft Starling holds back lands in Command's approval queue, with the reason it was held. Approve it, reject it, or take the ticket over in Resolve.
Screenshot-style mock of Starling Command's AI Drafts approval queue showing two drafts awaiting review, each with the ticket subject, the reason it needs review, a preview of the drafted reply, and Approve and Reject buttons.
Everything else you'd expect — already built.
Starling is a complete support platform, not an AI layer bolted onto someone else's.
-
Email ticketing
Inbound email becomes a ticket, with threading, attachments and send-status tracking.
-
Knowledge base
Write articles or import a CSV. Keyword and semantic search find the right one.
-
KB upkeep suggestions
Starling groups resolved tickets and proposes new or updated articles. Nothing goes live until you approve it.
-
AI triage
Summary, sentiment and intent on every ticket — when it arrives, or when it's opened.
-
Suggested replies
A ready draft in the agent panel, related articles beside it. One click to apply.
-
Auto-reply
Sends when a grounded draft clears your threshold — or holds it for approval. Your call.
-
SLAs
First response, next response and resolution targets, with breach alerts.
-
Reporting
Volume, response times, CSAT and per-agent stats, a backlog heat map, CSV export.
-
CSAT surveys
Sent automatically after a ticket closes. One click to rate — no customer login.
-
Automation rules
Route and prioritize new email by subject, body or sender — or send an instant reply.
-
Macros
Team-wide and personal templates that can also set status, priority or assignment.
-
Spam filtering
Allow and block lists, automated-sender and mail-loop detection, flood quarantine.
-
Web forms
Hosted or embedded on your site, with topics that route to the right department.
-
Customer portal
Customers sign in with an email link to see and reply to their tickets.
-
Slack notifications
New tickets, assignments, SLA breaches and CSAT responses, posted to a channel.
-
API & webhooks
Create tickets from other systems, and get signed webhooks when tickets or CSAT change.
Built right Strict tenant isolation at the data layer · Verified inbound email · Real-time updates, no refresh · Duplicate-safe sending
Why I'm building Starling.
I spent years on support teams, first as an agent and then as a manager. I watched our knowledge base go stale because nobody had time to update it. I watched agents copy-paste the same answer fifty times a week. I watched leadership get excited about an AI tool, turn it on, and turn it off two weeks later because it was sending replies nobody had approved.
The problem with most AI support tools is that they're built by people who've never taken a ticket. They look impressive in a demo. They fall apart on a Monday at 4pm when the queue is on fire and the AI just sent something embarrassing to your biggest customer.
Starling is what I wish I'd had. An AI that actually knows our knowledge base. Confidence scores I can tune. Modes I can dial up or down as I learn to trust it. A platform that treats my agents as the experts and gives them better tools — instead of trying to replace them.
If you've run a support team, I'd love to talk. Apply to be a design partner, ask for a call, and tell me what's broken about your current setup. I read every reply.
Questions you probably have.
- We're launching this fall. Right now we're working with a small group of design partners, who get in first.
- Only if you turn that on. New workspaces start with AI off. Run assist drafts replies for your agents and never sends. Auto-reply can send on its own, but human review is on by default, so every draft waits in an approval queue until you decide otherwise.
- Your data is never used to train AI models. Conversations are processed by Anthropic (Claude), retained up to 30 days for safety, then deleted. To power search, your knowledge base and ticket text are sent to OpenAI's embedding API, which returns mathematical vectors that our search index is built from. OpenAI does not train on API data and keeps API logs up to 30 days for abuse monitoring.
- Yes. Write articles in Starling or import them from a CSV. You can also send resolved conversations from another system through our ingestion API, and Starling will propose articles from them for your review.
- We're still finalizing pricing. Design partners get early pricing locked in for their first year.
- Today: email, embeddable web forms, and a ticket API — all landing in the same queue. Slack gets notified about new tickets, assignments, SLA breaches and CSAT responses. A customer-facing chat widget, two-way Slack, SMS and WhatsApp aren't built yet; design partners help decide which comes first.
- Starling uses PostgreSQL with strict per-organization isolation. Attachments are stored on AWS S3 or Cloudflare R2. We'll publish a full security and data-handling page before launch.
- Starling is built by a former support manager who got tired of bad support software. More on that above.
We're taking on a small group of design partners.
We're keeping the group small so every partner gets real attention before Starling launches this fall. Tell us about your team — if it's a fit, we'll reach out personally.
- A direct line to the founder Talk to the person building it, not a sales team.
- A real say in the roadmap What your team needs moves up the list.
- Early pricing, locked in Your first year at design-partner pricing.
Thanks — we've got it. We review every design-partner request and will reach out personally. If you asked for a call, expect to hear from the founder soon.