AI operations · Software · EU · UAE · LatAm
Manual operations are the most expensive line you have.
We turn them into agent-run processes and publish the numbers. People stay where the law or the judgement requires them. One of our cases is a regulated payments platform inside our own group; the figures are below.
One thing, done end to end.
We take a process that costs you people every month — support, document review, compliance triage, reconciliation, reporting — and rebuild it so agents execute it inside your systems, a person signs where the decision has legal or financial effect, and every step is logged. You get the before and after in numbers. If a process should not be automated, the diagnostic says so and that is all you pay for.
Results, measured
What changed in one of our cases.
One regulated payments platform. May 2026 against September 2026. Internal management figures; the full table and what it leaves out are on the case page.
Cost is measured. Quality is being measured against an evaluation set per workflow: error rate, corrections, reopened work, time to resolution. We show the current figures under NDA and publish them at 90 days of run data. One case, four months: read it as evidence of method, not as a forecast for your operation.
Your software. Your operation.
Why rent another tool when you can build your own?
In one of our cases we rebuilt a financial operation across jurisdictions on software we own. We bring that to your build-or-buy decision: direct integrations, your own modules and agents around the work your business actually does.
Go closer to the source
Connect to your own databases, original providers and authorised APIs. Remove an intermediary when a direct connection delivers better value.
Build around your business
Your own reporting, purchasing, inventory, customer workflows or ERP modules. Open-source components where they fit; source access and ownership agreed in the scope.
Give the work to your agent
Ask for information, prepare an action, execute it within defined permissions. We build the system, show you how to use it and maintain it under an agreed service.
Fewer subscriptions where it makes sense. Clear costs for models, infrastructure, data access and maintenance.
01Four layers
Four layers. That is where the payroll is.
Customer operations
First-line support, onboarding follow-up, document chasing, status questions, refunds and disputes. Agents answer in the customer's language and escalate on rules you set.
Compliance operations
KYC and KYB file review, document validation, sanctions and PEP triage, transaction-monitoring alerts, regulator-ready case files. A person still signs; the person stops doing the reading.
Money movement
Cash-in and cash-out integrations, rate and spread logic, settlement reconciliation, exception handling, treasury movements. Errors here cost money, so this layer pays back fastest.
Engineering and DevOps
Database and balance migrations, automated pipelines, monitoring, code and review, internal copilots over your own data, reporting finance actually trusts.
02One of our cases, measured
From USD 96,636 a month to USD 34,589, on a live regulated platform.
May 2026: USD 96,636 a month across the comparable cost categories. September 2026: USD 34,589. Same platform, same banking rails, same regulatory obligations. The operating team went from 21 to 5. One line went up on purpose: the licensed roles the law assigns to a named person.
Be the second case.
We take a limited number of external diagnostics on one condition: the anonymised before and after can be published here, with your approval of every figure. In return the diagnostic is priced as a reference engagement. Say so in the first message.
03How we work
Three steps. You can stop after any of them.
Diagnostic
Fixed fee, two weeks. Process inventory, cost baseline, what to automate first and what to leave alone. You own the document.
Migrate one process
Fixed scope, fixed price. One workflow into production with evaluation set, escalation rules, rollback and a measured before and after.
Run
Monthly, optional. Monitoring, quality and cost per task, model changes, and training for the people who stay. Or we hand over and leave.
04
Regulated, high-volume, or both.
Payment and fintech operators, banks' operations teams, insurers, marketplaces, and public bodies with intake and case backlogs. Public-sector work is delivered as fixed-scope projects with procurement documentation and a record that survives an audit.
05Regulatory context
Built for the rules your operation already lives under.
A regulated operator does not need a vendor who discovers the regulation at delivery. This is how we position every engagement; the controls actually implemented are listed project by project.
EU AI Act
Each workflow is classified by use case before it is built. Where a decision falls in the Act's high-risk categories, a named person decides and the agent prepares. Logs, evaluation sets and model versions form the technical documentation.
DORA
For financial entities we deliver as an ICT third-party provider: data for the register of information, contractual clauses on access, audit and exit, incident-reporting support and testing evidence.
GDPR and data residency
Processor terms, data minimisation, records of processing, no training on client data. EU regions for model providers and infrastructure where the client requires it; own infrastructure or on-premise where that option does not exist.
Certifications
G66 does not hold ISO 27001 or ENS certification today and does not claim it. Controls are aligned to ISO 27001 Annex A and documented per engagement; the certification roadmap is available on request.
Compliance with these frameworks remains the obligation of the regulated entity. We build so that its evidence exists.
Tell us the process and what it costs you.
The diagnostic tells you, in two weeks and with your own numbers, whether it pays back in months or in years. If it does not, we say so.