Case studies

Systems in production. Real results.

These aren't proofs of concept. Each case is a live system in insurance, banking, or investments, with its pillar, industry, and numbers.

Some of the companies we work with
Inter Quálitas Blue Capital Dynamic Re Yellowstone AON MRS Alya Lifeflex Sekura Promass Lockton SumaPay Innovazione BTG Marsh Era GMX Seguros Atlas Sura XS Global SEOS Tribal Credit GNP Actinver CXC Cerca Seguros Collado Cuprum Brove WTW Principal Toku Unisap Afirme Seguros HDI Global Vitro Dock Meryto Insound GP Plus Zurich Unitam
How much of the underwriting process a person does measured
Insurance Reaseguro

AI reinsurance underwriting

The challenge

An MGA with a facultative underwriting desk receives risks ceded by brokers across Latin America, the Middle East, North Africa, and Europe, each with its own slip, loss history, SOV, and inspection report in scattered formats. A senior underwriter had to read all four by hand and cross-check them against appetite and available capacity before deciding whether to quote.

The solution

A ten-phase AI protocol takes each case's documents, extracts and structures them, synthesizes the risk profile, matches it against available capacity, and calculates technical premium with a deterministic actuarial engine, delivering a recommendation report for the underwriter to decide on.

suscriptor AI Deterministic actuarial engine
97%
Of the underwriting process automated
4:36
Average minutes per completed case
1,546
Hours saved vs. the manual process
Insurance Contract management

Contract Management System

The challenge

Vendor contract management depended on email, files duplicated for every request, and in-person wet-ink signatures: no stage traceability or document expiration control.

The solution

End-to-end platform, with a single vendor file by tax ID and online electronic and biometric signature, now running across the group's six companies.

Custom development E-signature Single case file
+30,000
Legal documents under management
2.6x
Demand absorbed in 3 months
189
Requests in the record month
See full case study
Insurance Claims verification

End to end auto claims verification

The challenge

Fraud-flagged claims verification lived in spreadsheets and scattered tools, with no central audit trail, and a flagged claim could take up to 30 days to be assigned to an analyst.

The solution

A platform that governs the full cycle: an engine of 12 weighted rules assigns every flagged claim to an analyst the same day, then manages the document file, typed rulings, and fee control with tiered approvals, synced back to the client's core.

Custom development Rules engine Single case file
Same day
Assignment to analyst · previously up to 30 days
7,194
Claims verified in 30 days
12,250
Claims handled since go-live
Banking · Corporate payments Wallet

End-to-end digital wallet

The challenge

A Mexican corporate payments operator was already moving the money: a company funds a pool and balance flows down to its employees' cards. What did not exist was control over who is allowed to do what: no role and permission layer, and no single point of administration over the main balance.

The solution

We built the governance layer on top of the platform already in operation: roles and permissions with a super administrator over the main balance, and the cardholder app rebuilt end to end: balance, activity, disputes, bill payments, withdrawals, and travel expenses. The figures show the scale of the platform this layer governs, measured in July 2026.

Custom development Roles and permissions 3-D Secure

Scope note: Suma Pay, the payments platform in this case, is developed by rocket code.

$49.4M MXN
Disbursed to cardholders · July 2026
11,903
Active cards · July 2026
99.8%
Approved operations · July 2026
The next case?

Your operation could be the next case study.

Tell us which process you want to solve. We'll show you how we'd do it and what result you could expect.

Run the discovery with rocky

By submitting you accept our privacy notice.

No spam. The engineering team replies, not a bot.