Your warehouse already has the answers.
Now it can speak.
I wire AI into your data — with your house rules encoded, your caveats attached, and nothing invented. Questions answered in Slack in minutes. Decks that build themselves. For casino and sportsbook operators.
how did FTDs do last week?
FTDs, Wk 29: 214 — down 9% WoW (236 in Wk 28), flat on the 4-week average. The dip sits in one affiliate that paused; organic and PPC held. Watch it one more week before touching anything.
ftd_date · test accounts excludedwhich affiliate IDs caused the drop?
One ID — 64 paused mid-week. The rest held or edged up. Same week, same caveats as above.
| Affiliate ID | Wk 28 | Wk 29 | Δ |
|---|---|---|---|
| 64 | 31 | 4 | −27 |
| 2 | 26 | 28 | +2 |
| 13 | 18 | 21 | +3 |
| all others | 161 | 161 | 0 |
Every operator has the same bottleneck. The data exists. The analysts exist. But between a question and its answer sits a queue — days of it. So decisions get made on gut, and the warehouse sits there, knowing.
AI closed that gap for me. Not AI that sounds right — AI that is right, because it follows the same rules a good analyst would: which players to exclude, how a VIP is defined, which columns lie. I build that setup inside your operation.
Asteri — Greek for "star". Also what happens when a theoretical astrophysicist spends thirteen years in iGaming and then teaches a machine to do the repetitive half of the job.
Three things. Set up to run
on their own.
Ask your data in Slack
A channel where anyone — CEO, CRM lead, product, sportsbook, whoever needs the number — asks a question in plain English. The AI writes the SQL, runs it against your warehouse, and posts the answer in minutes.
Follow-ups work like they do with a person: reply in the thread, it carries the context. An ambiguous question gets asked back — which metric, which window — not picked on your behalf. An empty result gets flagged as a probable wrong filter, never reported as a finding.
Every answer carries its caveats: which table, which window, what was excluded. A number without its caveats isn't an answer — it's a future argument.
The AI analytics stack
Claude, connected to your warehouse over MCP — BigQuery, Postgres, Redshift, whatever you run — with your business encoded as rules it must follow. Your metric definitions, your exclusions, your known-bad columns.
That encoding is the whole product. It carries your definitions the way a senior analyst carries them — how a VIP is scored, which players sit outside financial metrics, which columns look right and lie. Skip it and you get answers that look right and aren't.
It plugs into the warehouse you already have. No migration, no new BI tool, no six-month project. And if the data underneath is a mess, that's the first fix — I've done it on top of a thousand raw tables.
Reporting that builds itself
Weekly trading decks — week-on-week, month-to-date, year-on-year, VIP versus core. Retention dashboards. Monthly scorecards. Bonus-cost breakdowns. Whatever your Monday meeting runs on.
Generated on schedule, validated against totals you already trust before anyone opens them, saved to Drive — and the link posted in Slack before the meeting starts.
Your analysts stop assembling and start analysing.
Business question first.
Then the machinery.
read-only
caveats on every answer
One brain in the middle. Everything it says traces back to your data, under your rules.
Map
What decisions is the business trying to make? What questions keep hitting the analyst queue? We start there — never with the tooling.
Encode
Your house rules become instructions the AI follows on every query. How you define a VIP. Which players get excluded from financials. Which tables to trust and which to never touch.
Wire
Warehouse connected read-only — the AI can query, never write. Slack connected, schedules set.
Guard
Caveats travel with every answer. Empty results get flagged, not reported as zero. Sensitive channels get a QA pass before anything posts. And the AI never fills a gap with an estimate — an error says it errored.
Run
Validated against numbers you already trust, then live on schedule. Done means automated and signed off — not "ready to review".
Not a concept. A working setup
I run every day.
A Slack channel that answers data questions on its own
At a multi-brand operator group, leadership asks questions in Slack — FTDs, deposits, retention, whatever the week raises. The AI runs the SQL and answers in minutes, caveats included. Sensitive channels pass a QA gate before anything posts.
Weekly trading decks that build themselves
The Monday deck — week-on-week, month-to-date, year-on-year, VIP versus core — assembled on schedule from the warehouse and validated against known totals before anyone opens it.
A clean layer over a thousand messy tables
One client sat on 1,000+ raw tables nobody could query with confidence. Consolidated into a single analytics layer the AI — and every analyst — now works from. The value wasn't the cleanup; it was everything that became cheap afterwards.
Angeliki Kiakotou, PhD
Theoretical astrophysicist turned iGaming analytics leader. Twelve-plus years inside operators — Rank, then Entain across BWIN, Ladbrokes and Gala, including a stretch managing global bonus spend across the group. Founder of The Bonus Doctor, an iGaming analytics consultancy, since 2024.
Asteri is the AI half of that work, made its own thing. The systems I built because my own weeks demanded it — now set up inside your operation, with your rules, for your team.
One rule carries over from everything else I do: the analysis is sacred. No invented numbers, no spin, no claim the data doesn't support. The AI I set up holds the same line.
Want your warehouse to start talking?
Tell me what your team keeps waiting on. I'll tell you what a first setup would look like.
DM me on LinkedIn