// industry · iGaming

Hold the players
your retention team
cannot see going.

We are an AI / ML engineering studio for iGaming. We build per-player retention scoring, media-buying agents, generative content pipelines, and the custom engineering operators and platforms need on top.

Book a scoping callSee Retivo→
ML, not LLM, on money decisionsDecision per player, not per cohortiGaming domain, not generic outsourceLean team, real unit economics
Per player
Scoring decisions, not cohort blasts
~2 weeks
Scoring core, Claude-assisted dev
0 LLMs
Anywhere money is on the line
Self-written
Platforms supported via adapter
// what we do

What you actually get from us.

Six work directions across retention, agents, content, data, custom development and team augmentation. Each card is the outcome, not the label.

Retention

Hold VIPs before they go silent

Per-player churn and reactivation scoring on raw transactions. Gradient boosting, not LLM. Output is a prioritised worklist into your retention CRM.

Agents

Run media buying without scaling the team

Agents with bounded autonomy on top of your ad accounts. L1 recommendations, L2 confirmed, L3 autopilot inside a KPI envelope.

Content

Ship affiliate creative at volume

Generative video and creative pipelines for UBT and affiliate traffic. Script to scene to cut, briefed in plain text, delivered ready to publish.

Data

Give the retention team a real player 360

Player cards, behavioural segmentation, dashboards and BI on top of your platform schema. The base layer everything else is built on.

Custom build

Build what your platform does not have yet

Backend, API integrations into platforms and payment providers, frontend that serves the model. We ship the system around the model, not just the model.

Staff aug

Plug ML engineers into your roadmap

ML and backend engineers embedded in your team for the duration of a project. For when the bottleneck is hiring speed, not direction.

// products

We ship our own products, then build the engineering around them.

Retention CRM

Retivo

Per-player ML scoring of risk and reactivation in real time. One decision per player, returned with the features that drove it. Built for online casino operators running a self-written platform.

  • Gradient boosting on raw transactions
  • Score plus feature-level explanation per player
  • Adapter for self-written platforms
Content pipeline for affiliate and UBT volume

AI video studio

Script to storyboard to scene to cut, automated end to end. Used to land affiliate and UBT creative volume without scaling a production team.

  • Roughly $2 to $4 of generation cost per finished video
  • Around 5x faster than a manual edit cycle
  • Briefed in plain text, delivered ready to publish
Available as a service · landing in progress
// who we work with

Same team, different emphasis per segment.

Online casino

B2C operators

// pain

Manual retention is expensive, cohort blasts burn bonuses, VIPs are leaving before anyone notices.

// what we offer

Retivo as the retention layer, analytics on top of your platform schema, custom work for the gaps.

Whitelabel and turnkey vendors

Platforms and PAM

// pain

Your operators need AI and CRM. Building it in-house pulls focus from the platform itself.

// what we offer

Whitelabel ML retention module that plugs into your stack, plus customisation for your largest operators.

Performance and arbitrage teams

Affiliates and media buying

// pain

Manual campaign operations do not scale. Creative volume caps growth before budget does.

// what we offer

AI media-buying agent with bounded autonomy and a generative content pipeline that ships at the volume traffic teams actually need.

// recent work

Proof, not slide decks.

One case is public. Three more are under NDA - we walk them through live on the scoping call. We do not fill the page with fabricated numbers.

Per-player retention

Online casino: per-player churn and reactivation scoring

Gradient boosting on raw transactions, scoring core built in around two weeks, pilot running on six-plus months of player history. The shipped instance of the Retivo approach.

Read case →
// under NDA
Operator analytics · Media-buying agent · Platform integrations

Reviewed live on the scoping call. We do not put fabricated metrics on a website to fill space.

// approach

ML on money decisions. LLMs only at the edges.

Player payback is five to seven months. A hallucinated risk score is real loss. Three rules we engineer to.

01

Formulas on money decisions

Retention, decisioning, risk scoring - anywhere a wrong answer is real money. Auditable gradient boosting and reinforcement learning, not generative scoring.

02

Bounded agent autonomy

When an agent has authority to act, another layer enforces the bound. KPI envelopes, sanity checks, and rollback paths are part of the build, not an afterthought.

03

Protect the payback window

Player payback in this industry is months. One bad offer or a hallucinated risk score is real loss. We engineer for the payback period, not for the demo.

// team

Small team, real roles.

We list roles, not headshots. Names and CVs available on request once a project is scoped.

ML and data engineering
Gradient boosting, sequence models, feature pipelines
Backend and integrations
FastAPI, Postgres, platform adapters
Product and iGaming ops
Retention, VIP, payback economics
Media buying and content
Affiliate, UBT, generative pipelines
// FAQ

Questions operators and platforms actually ask.

How is this different from a generic AI agency?
We only work inside iGaming, and we only build for decisions that touch unit economics - retention, VIP holds, media buying, content volume. A generic agency takes the brief. We challenge it against the payback window.
Do you use LLMs for scoring or decisioning?
No. Gradient boosting and other auditable ML for anything that drives money. LLMs only for copy drafting, analyst summaries, and explanations of an already-auditable score.
We are a platform, not an operator. Can you whitelabel?
Yes. An adapter decouples the retention layer from your schema. Your operators get the ML output through your stack with your branding. Per-operator customisation is part of the engagement.
What stack do you run on?
Python, XGBoost or LightGBM for tabular, PyTorch for custom networks, FastAPI for inference, Postgres for storage. We integrate via your API or direct database read. No migration required.
How is a pilot scoped?
Two phases. Phase one: data audit and adapter against your platform - we verify the data can support the model. Phase two: model and pilot against six-plus months of history. We quote after phase one, not before. If phase one shows the data is not ready, we stop and tell you.
How is responsible gaming handled?
Self-exclusion, deposit limit, time-played and other RG flags are read from your platform and checked before any retention trigger fires. Player data stays on infrastructure you control.
// contact

Bring the data and the target.

Send a short brief. Ivan reads everything and replies within one business day. On the first call we will tell you if a pilot makes sense - or where Retivo, custom development, or just a referral fits better.