Data science & analytics
Decision engines, not experiments — statistics before hype, measured against the baseline they have to beat.
The challenge
Decisions still run on gut feel
Credit, pricing, stock, retention — the calls that move your P&L are made daily, and mostly from experience and habit. The data is there; it just doesn't reach the moment of decision. And where models do exist, too many stay in notebooks — impressive in the demo, absent from the workflow, indefensible in front of an auditor.
Our approach
Decision engines, not experiments
We build models that live inside the decision: scored in seconds, explainable to the applicant and the regulator alike, measured against the money they move. Each engine starts from a pattern we've already run in production — credit, fraud, churn, pricing, stock — trained on your data, using the simplest method that wins.
Results
Models you can defend on paper
Lower risk losses, more accurate forecasts, and spend targeted at the customers, SKUs and channels that actually return.
What we do
Approve, decline, limit and terms — decided in seconds, not committees. Scorecards, rules and affordability checks in one engine, explainable to the applicant and the regulator alike.
- Scorecards, rules and affordability checks combined in one decision flow
- Decisions in seconds, explainable to the applicant and the regulator alike
- Champion / challenger built in, so the engine keeps improving in production
One price for everyone means your best customers subsidize your worst. Price each loan by its actual risk — margin protected on the risky end, growth unlocked on the safe one.
- Each loan priced from its actual risk, not one blended rate
- Margin protected on the risky end, growth unlocked on the safe one
- Elasticity tested, so pricing moves are evidence, not guesses
Non-performing loans are a portfolio, not a pile. We score recovery likelihood per exposure, match each segment to the right strategy — restructure, collect, sell — and price the portfolio for either outcome.
- Recovery likelihood scored per exposure, refreshed as behaviour changes
- Segments matched to strategy — restructure, collect, sell
- The portfolio priced for either outcome: hold or sale
Scoring every transaction in the moment it happens — catching the fraud without strangling the checkout. Tuned to your loss data, monitored for drift from day one.
- Every transaction scored in the moment, tuned to your loss data
- Thresholds balanced against checkout friction — fraud caught without strangling conversion
- Drift monitored from day one: fraud adapts, so does the model
Who's about to leave, why, and what offer changes their mind — lifecycle-aware models feeding next-best-action, not a slide deck. Retention spend goes where it actually retains.
- Lifecycle-aware models: who is leaving, why, and what changes their mind
- Next-best-action feeds the channels you already run — not a slide deck
- Retention spend measured against retained revenue
The right upsell or cross-sell at the moment the customer is listening — in the app, at the till, in the call. Propensity models served in real time, measured against revenue, not clicks.
- Propensity models served in the moment — in the app, at the till, in the call
- Offers ranked by expected revenue, not click likelihood
- Every recommendation measured against uplift
Some customers take ten others with them when they go. Network analysis over calls, contracts and households finds who actually sways whom — so retention protects the customers who matter beyond their own bill.
- Network analysis over calls, contracts and households — who actually sways whom
- Influence factored into retention priority and offer size
- Built for telco-scale data volumes
Ad inventory priced by demand, audience and moment instead of a static rate card — yield managed continuously across channels, with the floor prices your sales team can defend.
- Inventory priced by demand, audience and moment — not a static rate card
- Yield managed continuously across channels
- Floor prices your sales team can defend
Forecast-driven stock levels and allocation per SKU and location — less capital sitting on shelves, fewer empty ones. Reorder points that move with the season, not with last year's spreadsheet.
- Forecast-driven stock levels per SKU and location
- Reorder points that move with season and demand, not last year's spreadsheet
- Allocation balanced between shelves, warehouses and cash
Proven in production
AI & Data
Kvika — Five generations of a credit decision engine
Netgíró's credit decisions moved from a bureau score and human judgement to models that learn — halving risk losses, then improving through five model generations without a single rewrite.
Data
Arnarlax — Iceland's largest farmer and producer of Atlantic salmon
Iceland's largest Atlantic salmon producer replaced its source systems and its data platform at once — migrated to Microsoft Fabric without the business losing a reporting week.
Web & Mobile
Netgíró — Iceland's national payment method
Iceland's leading buy now, pay later platform — built and scaled over more than a decade of partnership.