Andrey Sabynin

Founder of FlyOptima, an AI-native developer of solutions for banks and fintech.

I build fintech and lead the people who make it:

people since 2007 and AI agents since 2024.

Former management board member of a universal bank · Bank CIO · Founder of a banking-software company

  1. IT costs more every year. Is it getting any more efficient?

    I took a third off a universal bank’s IT costs in a year - without stopping products or losing development speed; development efficiency rose 10% on an independent metric. I measure IT efficiency with metrics - delivery speed, cost of change, quality - and answer for them to the board, shareholders and the CEO.

    BCG, 2025: cost-to-income above 60% at traditional banks, about 35% at digital ones

  2. Everyone is rolling out AI. Is there a measurable effect?

    I build AI development that pays back: agents do the work of a team, checkpoints hold the quality, the board sees the result in numbers.

    PwC, 2026: 56% of 4,454 CEOs see no financial benefit from AI so far

  3. Or simply: need an outside expert view of your IT?

    How efficient and modern it is - let’s discuss your business’s tasks and the scope of my responsibility. 30 minutes on costs, delivery speed, quality and the use of AI.

    Pick a time →

Who I am

I have worked in banking technology since 2007. I served on a bank’s management board with responsibility for the operations and technology divisions and moved IT delivery to a product operating model. I secured shareholder and board approval for the IT strategy - and delivered it. I have led technology divisions of up to 1,250 people and IT budgets of up to $90 million a year.

In 2022 I founded FlyOptima. The company develops and implements solutions for banks and fintech, and helps fintech start-ups build modern IT processes with a focus on AI SDLC.

I know the technology hands-on, not from reports: FlyOptima’s products were built with my direct participation. Since 2024 I have run development where AI agents work as a team - under a protocol, with verification and measurement. This site, its design and its original typefaces were made the same way - by me together with the agents.

I worked as CIO of a neobank on a C-level-as-a-Service contract, scaled banking platforms and optimised costs. I measure the result by delivery speed, cost of change and quality.

Who I work with

Banks, fintechs and corporations that have a clear development strategy and measurable goals.

Those who are ready to grow, try, act and measure - and want to be at the forefront of transformation: AI, product and technology.

I give them industry solutions, experience of leading large teams and modern IT that reaches the goal within a limited budget.

What I offer

to a partner

To whoever is launching or modernising a fintech

  1. A ready, modern banking core

    The business starts on a ready platform: cards, lending, instalments, acquiring. Regulatory requirements are covered for your jurisdiction.

  2. A co-founder and partner

    I take the IT domain under my responsibility: a cheaper stack at the start, a team and processes in place from day one.

  3. Launch and growth experience

    Grew a bank’s IT division from 40 to 650 people. Over the same four years the bank’s profit grew almost fourfold. FlyOptima’s clients are in EMEA, LATAM and Asia.

  4. M&A expertise on the IT side

    Due diligence of platforms and teams, platform and IP acquisition, post-deal IT integration.

to a corporation or bank

To whoever runs a bank or fintech

  1. CIO / CTO as a Service

    A technology-leadership contract with measurable goals and pay tied to results: two to three days a week or full time.

  2. Audit and review of current IT

    Teams, infrastructure, technology, costs - where money and speed are lost. A third off IT costs in a year without losing speed - a proven case.

  3. A digital twin project

    We digitise and link the layers of the bank - from strategy and finance to systems, processes and metrics. Leadership gets what-if analysis and an AI adviser with the full picture of the company. My vision ↓

  4. AI transformation under board control

    My own processes: development, service and support, architecture analysis, due diligence, a single corporate archive. Checkpoints, evidence, data inside the perimeter. Processes ↓

to a board, CEO or shareholder

I discuss a management board/C-level role - for projects that need a mandate for transformation, not a consultant. I am interested in growing businesses with long-term team incentives.

Shall we discuss your task? Pick a time →

Career

  1. 2025 - 06.2026

    Novacard, MexicoCIO of a neobank, CIO-as-a-Service model

    Scaled the neobank’s platform, moved releases to a monthly cycle, launched an IT quality programme. In June 2026 I handed the team over to a permanent head.

  2. 09.2023 - 10.2024

    Renaissance Bank, MoscowSenior Vice President, Head of Technology, Management Board member

    Cut a third of the bank’s IT costs in a year without stopping products or losing development speed. Took over a technology division of 1,250 people - staff and contractors whom I managed and budgeted for; contractor spend more than halved, 45 contractor roles converted into permanent positions, development efficiency up 10% by an independent metric. Chaired the bank’s Technology Committee, moved key systems to open source, replaced departments with a flat structure. Completed the year-long programme in October 2024.

  3. 2022 - present

    FlyOptima, AlmatyFounder

    See the FlyOptima section.

  4. 10.2017 - 12.2021

    Alfa-Bank Kazakhstan, AlmatyCIO, reporting to the CEO

    Built the bank’s technology division from 40 to 650 people and an Open API ecosystem. Over the same four years the bank grew one and a half times by capital and almost fourfold by profit. 30+ projects: CRM, anti-fraud, AML, loan origination, digital channels; an in-house R&D lab. Secured board approval for the IT strategy in 2018. Completed the contract in 2021.

  5. 2012 - 2017

    Otkritie Bank, MoscowDirector, IT Development Department

    Team of 200+. Agile transformation, integration of acquired banks, Innovation Fintech Lab.

  6. 2010 - 2012

    Citibank, MoscowSenior IT Project Manager

    Data warehouse and CRM. Team of 20, offshore in Singapore and India.

  7. 2007 - 2010

    MBRD (now MTS Bank)from specialist to IT manager

Education: MESI, 2008, honours. PMP, 2013. SKOLKOVO Startup Academy, 2015. Languages: Russian, fluent English.

FlyOptima

Operating since December 2021 from Almaty, registered in Astana in May 2022, resident of AIFC and Astana Hub. Clients and projects across EMEA, LATAM and Asia. Software licences, development and testing, platform implementation, architecture audit, CIO as a Service.

The main product is FlyCoreBanking, an accounting and product core for cards and lending: products, tariffs and rules are data, not code. Implemented by FlyOptima and partner integrators. flyoptima.com

How I lead

I find competent people and build a system of management and control. Not a “strong team” in the abstract, but people for specific domains and clear rules that show who is accountable for what.

I give authority and autonomy within a domain - for the team’s goal, not for autonomy’s sake. I agree the result and the authority to achieve it - team, budget and access. I answer for the result within those limits.

The board sees status, risks and numbers. The team gets authority and answers for the result. One does not work without the other.

Shared understanding in the team is what secures the right result. “Have I understood you correctly?” is the phrase I use most in meetings: I clarify until the picture is shared. A result is something you can show, not something you can retell.

Publications and talks

Principles for leading people and AI agents

  1. Business first. The business goal and strategy run as a common thread through every process - in IT and in every other function.
  2. Verify rather than trust. I define the criteria and metrics of the result before the work starts, and measure afterwards.
  3. Transparency is the basis of trust. Status, risks and numbers are visible to those who make decisions.
  4. Fix the cause, not the symptom. I go down to the root of the problem, acknowledge it and fix it.

Contact

Book a meeting

30 minutes, slots in Moscow and Almaty time.

ceo@flyoptima.com · linkedin.com/in/andrewsab · flyoptima.com

Moscow · Almaty

Full profile, project list and references - on request.

My own AI transformation processesfive processes

AI transformation is not one tool but a set of processes in which agents work under a protocol and human control. I built and tested each process on my own products.

  1. Development

    From requirements to release: agent roles, checkpoints, a human decides at four points. 10× faster than classic development, measured. Process diagram ↓

  2. Service and support

    A bank’s support service, external and internal: customer and staff requests are handled by AI agents under a regulation, hard cases go to a human, every failure is traced to its cause.

  3. Architecture analysis

    C4 diagrams, module contracts and an architecture decision log “before → after” - the architecture is visible and verifiable.

  4. Technology due diligence

    Assessment of the platform, code and team before a deal or a core replacement: agents read the code and documents, the conclusion and accountability stay with the expert.

  5. A single corporate archive

    The company’s mail, documents and code in one index with search. AI agents work with the whole picture of the company, not with fragments of context.

Want to know more? Pick a time →

My own AI development processprocess diagram

Development is run by an AI team with separated roles - analyst, architect, developer, tester, DevOps and domain experts - plus independent reviewers. A human decides at four points: goal, scope, final control, release. Every stage is closed by a checkpoint: beyond it only with evidence.

1Intake
whoOwner
whatSets the goal and priority of the task
artefactRequest
checkpointGoal and priority set
human decisiongoal
2Requirements
whoAI analyst
whatInterviews the owner, builds the catalogue of requirements and tests
artefactTracker ticket - opened by an agent
checkpointNo test or ticket - no work
3Architecture
whoAI architect
whatBuilds C4 diagrams, approves module contracts
artefactC4: context, containers, components
checkpointScope and architecture approved
human decisionscope
4Development
whoAI developer, DevOps, experts
whatWrite code and tests to the approved plan and standard
artefactCode and tests in a branch
checkpointReview and automated checks stop the build
5Quality
whoAI tester and AI reviewers
whatThe tester runs the tests, models of another class match the result to the requirements
artefactTest run and report with evidence
checkpointQuality Gate: 5 mandatory conditions
6Control and release
whoRelease agent → owner
whatThe agent builds the release, the owner checks and publishes
artefactRelease and changelog
checkpointRelease only with the owner’s “yes”
human decisioncontrol, release
7Feedback
whoOwner and users
whatVerify feedback before it becomes a requirement
artefactNew requirements → stage 2
checkpointA bug is fixed only via a requirement
Agent cycle · stages 4 ↔ 5Executor ↔ reviewer. Models of different classes and different vendors check each other until fact matches specification.
← Learning loop · 7 → 2Error → rework → lesson → rule in the agents’ instructions → changelog entry. Feedback returns to stage 2 only after verification.

AI approach

  1. Roles separated

    Whoever writes does not verify. Independent agents match fact to specification.

  2. Cross-checking

    Models of different classes and different vendors check each other’s work.

  3. Evidence

    A task closes only with the code version and the check output.

  4. The process learns

    Error → rework → rule. Every rule goes into the changelog.

  5. Own skills and MCP

    My own set of skills, agent roles and MCP integrations with the tracker, the repository and the checks. The process is built as a product and transfers to the client’s team.

  6. Human at 4 points

    Goal, scope, final control, release. Everything else is done by agents under a protocol.

10×
faster than classic development: 36 person-weeks instead of 358 or more by productivity norms
24
active weeks: two people and a team of AI agents
148k
lines of code and tests, 114 merged changes
22
product releases

Measured on the change history of the Voice Scribe product, 2026. The process is live: this is how FlyOptima’s products and this site are made. Model and tool names - at the meeting.

Want to walk through it on your case? Pick a time →

My vision: a digital twin of the corporationdiagram

AI models make it possible to turn a corporation into a digital twin: describe all of its layers - from strategy to sensors in the branches - and link them together. Once assembled and digitised, such a model gives AI a single picture of the company, and AI becomes an advisory body that helps run it.

  1. 1Business strategygoals, markets, priorities
  2. 2Business driverswhat moves revenue, costs and risks
  3. 3Finance and costwhat each layer costs: the link to spending
  4. 4Productswhat we sell and to whom
  5. 5Systemswhat the products run on
  6. 6Infrastructurewhat the systems run on
  7. 7Organisationwho is accountable for what
  8. 8Processeshow we work on the systems, in which roles of the organisation, under which rules and regulations
  9. 9Sensors and metricsbranches, sites, apps

AI as an advisory body

AI on the board of directors: it sees every layer and the links between them. What-if analysis and transparency for leadership:

  • What happens if the business strategy changes?
  • What happens if we close branches or launch a new product?
  • What will a decision cost and where will it show up in spending?

Klaus Schwab and the World Economic Forum, “Deep Shift”, 2015: the first AI on a corporate board of directors was expected by 2026

I arrived at this idea long ago; the models of recent years have made it feasible. The concept runs into the limited context window of today’s LLMs, but a multi-agent system covers a huge body of data and saves tokens by splitting the work between models of different tiers. Decisions stay with people - AI gives them the full picture.

Interested in discussing modern IT management practices? Pick a time →