A private sportsbook analytics workbench I built to turn player activity, pricing, exposure and profitability signals into clearer commercial decisions.
Public-safe by design
Every entity, threshold, financial value and outcome shown on this page is synthetic. The live system and its underlying data remain private.
Context
Sportsbook decisions rarely depend on one clean metric. Player behaviour, profitability, market context, offer mechanics, exposure and uncertainty have to be considered together. I needed a more structured and repeatable way to move from raw evidence to a commercially useful recommendation.
The problem
A promotion can look attractive on turnover while weakening unit economics. A player can look profitable or unprofitable because of noise. A rule can appear effective in a simple before-and-after comparison while regression to the mean explains the change.
My role
I designed and iterated Molly primarily for my own analysis within an independent sportsbook advisory engagement. I own the problem framing, analytical logic, validation and AI-assisted build workflow, then use the output to prepare clearer recommendations and measurement plans.
Scope boundary: Molly supports my analysis. It is not presented here as a client-commissioned product or a client-wide deployment, and I do not claim ownership of campaign execution or the client's campaign tools.
Decision workflow
The workflow keeps the analytical steps visible so a recommendation can be challenged, refined and evaluated after the decision.
Start with the commercial question: an incentive, a pricing test, a player-value decision or a post-launch review.
Bring player activity, profitability, pricing context, open exposure and offer mechanics into the same view.
Use calibration, confidence intervals and shrunken estimates so noisy evidence is not mistaken for certainty.
Prioritise the cases worth reviewing, make trade-offs explicit and keep thresholds subject to expert judgement.
Decide what will be measured after launch or intervention before the result is known.
Representative case
One advisory contribution illustrates the kind of commercial reasoning Molly is designed to support. No real campaign figures or results are disclosed.
Would a planned volume-based rewards campaign create incremental value under the proposed mechanics?
I reviewed historical activity and profitability data alongside the proposed pro-rata and tiered reward structure.
The proposed mechanics had negative expected value for the operator, so I recommended changes to the structure and wagering requirements before launch.
Define participation, turnover, unit margin, net contribution and customer-value checks before evaluating the campaign.
Planning and analytical contribution only · no claim of end-to-end campaign execution
Advanced analytical views
These public-safe visuals recreate representative analytical workflows with synthetic data. They demonstrate the reasoning and interface patterns without exposing customers, commercial thresholds or live performance.
01 · Decision prioritisation
A representative prioritisation layer combining shrunken closing-line value, exposure, posterior skill probability, expected cost and uncertainty. The purpose is not to automate a restriction: it is to surface the cases that merit expert review first.
02 · Outcome review
A synthetic post-decision workflow using a frozen pre-decision baseline, matched controls, difference-in-differences and rule-level confidence intervals. This makes the evaluation logic visible instead of relying on a simple before-and-after comparison.
03 · Model trust
A reliability view for comparing predicted probabilities with observed outcomes. Calibration error, Brier score, drift and scope-level gates help separate a useful decision signal from a model that needs more evidence.
04 · Integrity analysis
A synthetic network view combining repeated co-activity, rare-selection lift, timing gaps and shared-funding evidence. Group-level recommendations remain traceable to the underlying signals and subject to human review.
What the work demonstrates
The value is in connecting quantitative evidence to a real decision, not in producing another collection of charts.
Implementation
I use AI tools to accelerate implementation, while retaining ownership of the commercial question, analytical logic, validation and the final output.
The production system stays behind authentication. This public page contains no source data, client identity, account information, credentials, live thresholds or proprietary implementation detail. The visuals preserve the analytical structure while replacing every entity and value with synthetic demonstration data.
Outcome and learning
Molly gave me a more consistent way to structure analysis, compare hypotheses and prepare evidence-based recommendations. Its most important design principle is simple: every view should connect to what could change, why it could change and how the result will be reviewed.
Commercial outcomes remain confidential and are not claimed on this page. A guided walkthrough using synthetic data is available for interviews and relevant professional discussions.