AI support automation · for small SaaS teams

You stop opening the inbox.

Switching on AI support takes a day. The work starts after that: answers drift from the facts, the tone slips, some tickets fall through. I take that work on for the long run.

Get a free queue teardown See how it works ↓
30%

of tickets close without a person

1,000+

tickets a month in the system I run

10

years inside support and QA

VK (ex-Mail.ru Group) Ecommpay IT Artlogics / pooling.me
01

An audit of what already runs

I look at what your AI actually tells customers and which emails it never picks up.

02

The parts a vendor won’t build

I build what the box doesn’t include: a knowledge base that updates with every release, rules for when the AI should stay quiet, and a path for hard tickets to reach a person.

03

Then I run support myself

Every correction goes into the base, so the answers get more accurate over time. Your time stays on the product.

What happens to a hundred tickets

  • 30close themselves.
  • 70reach a person, already with context and a draft reply.

Sound familiar?

What the work looks like

The system runs every day on a live product.

The product is over ten years old and gets about a thousand tickets and five hundred reviews a month. Its name is under NDA, so below is the same mechanism on open data.

Ticketarrives in the queue
Contexthistory, receipt, logs and known bugs are pulled in before the reply
Draftonly facts from the base
Decisionboilerplate goes out on its own, a person sends the rest
Ruleevery correction goes back into the base

A correction at the last step goes into the base and changes the next answers.

Reply box: a draft in German built from the 'Delete a Joplin Cloud account' template, confidence 90% Why this reply: mood, confidence 90%, source: the template

Five kinds of email go out without a person. This is one of them: a template reply adapted to the customer. Everything else waits for review.

Demo on public Joplin data · real tickets are under NDA · not affiliated with Joplin

I built and run a test automation factory on the same principle. The repositories are under NDA, but the activity shows on GitHub.

Customers thank us for being human. A model writes the replies.

The support I received was quick, clear and sorted the issue straight away★★★★★ · Google Play · EN
Customer support walked me through it step by step. Excellent service★★★★★ · Google Play · IT
Suporte rápido e eficiente★★★★★ · Google Play · PT

These are public store reviews. Support affects the rating, and the rating affects installs.

Tested on your tickets first.

There is no slide deck. I run every change on your past tickets first and only then release it to customers.

Who I am

Natural born helper.

A reply takes seconds to generate now, but working out what actually hurts behind an email is still hard. I started on the phone line and built support twice as its head, so I see both the customer and the team. I work remotely in English, Russian and Portuguese.

Send me your last 100 tickets.

I’ll send back a breakdown: how many of them can close without a person, where your current bot stumbles and how many hours a week it takes. Free, no call.

Send the tickets → hi@iliasupport.com