Data integration
Customer, finance and operational data spread across systems that were never designed to agree.
A reconciled, documented data layer: one definition for each number, and checks that flag breaks instead of hiding them.
Practical AI and automation that gives your team its time back.
We help small and mid-sized businesses connect their data, redesign slow workflows and apply AI where it genuinely adds value, with people accountable where judgement matters and every step open to review. Led by Fabien Vernaz, engineer and former CTO.
Where we usually start
- Data integrationSales, finance and operations each hold a different version of the truth.
- Business intelligenceThe management pack takes days to build and nobody quite trusts it.
- Workflow redesignA process that grew by accretion, with hand-offs nobody can fully explain.
- Unstructured materialContracts, reports and correspondence hold answers no one can query.
- Human and AI boundariesUnclear which decisions a model may prepare and which must stay with a person.
If one of these sounds familiar, it’s worth a conversation.
01 · What we work on
The structural problems behind the busywork.
Tidying an inbox rarely changes a business. The bigger gains sit in data that doesn’t agree, processes nobody has redesigned in years, and decisions taken on slow or doubtful numbers. And wherever AI is involved, it has to be auditable, reviewable and monitored. These are the areas we work in.
Business intelligence
Management reporting assembled by hand, late, and open to dispute.
Reports that refresh from source and show their working, so the discussion moves from “is it right?” to “what do we do?”.
Competitive intelligence
Competitors, prices, tenders and regulation keep moving, and the watching falls to whoever has a spare hour. Called veille concurrentielle in French.
A standing watch on the sources that matter. AI sorts and summarises, every item links to its source, and a person reviews what reaches a decision.
Workflow re-engineering
Processes that grew by accretion, with steps whose purpose nobody remembers.
We map the process as it really runs, remove what doesn’t earn its place, and automate what remains.
Reading at scale
Contracts, filings, reports and correspondence that contain answers nobody can query.
Structured data and searchable knowledge extracted from documents, every answer linked back to its source.
AI-driven processes
AI is easy to demo and hard to run: it is wrong in confident ways, and nobody can say afterwards why it did what it did.
A process where every step is explicit: what the model reads, what it prepares, which checks run on its output, and where a person takes over.
Auditable decision frameworks
Uncertainty about what AI may decide, what it may only draft, and who answers for it.
Rules you can review: the model prepares, a person approves anything consequential, every action is logged and every output traces back to its inputs.
Monitoring and early warning
Problems discovered at month-end that the data could have shown weeks earlier.
Continuous checks across data feeds and processes, raising a flag with context rather than noise.
Monitoring the AI itself
A model that quietly degrades, or drifts from its instructions, until a customer notices.
Sampled output checks, comparison against independent references and alerts on drift, so a system catches the failure first.
And when AI is the wrong answer, we’ll say so. Sometimes a cleaner process or a better database saves more time than any model. Where we do use it, we assume it will fail somewhere, and we build the checks for that. See our failure file.
02 · Working together
Start small, prove it works, then decide.
Every engagement begins with one well-chosen problem. You see a working result before committing to anything larger.
-
Time review
A two-hour working session, on site around Geneva or by video call. We look at what can be done: how work and data really flow, where time and quality are lost. Then we spec the work and agree how success will be measured, in a short written plan with effort, risks and expected benefit. The plan is yours, whether or not we build it. Ask us about terms.
-
Build
We build the first piece: small, documented and tested. We use software you already have wherever possible, and custom code only where it pays off. You own the code and the accounts.
-
Handover and training
Your team learns to run it, check its output and change it. Explaining technology to non-specialists runs through Fabien’s whole career, from training users abroad to teaching mathematics.
Part-time technology lead
For businesses without a CTO: senior judgement on technology decisions, vendor and quote reviews, and oversight of projects, without a full-time hire.
03 · Principles
Trust is the whole job.
A slow tool is an annoyance. An agent nobody can hold to account, whose work nobody can check, can end a business. These rules apply to everything we deliver.
-
Start from the work, not the technology
Before proposing any tool, we want to understand the process, the people doing it, and what a mistake would cost.
-
Your data stays under your control
Where confidentiality matters, tools can run on your own machines or on European hosting. Nothing goes to an external AI service without your agreement.
-
People stay where judgement matters
Automation handles the routine and a person reviews anything consequential. A good tool says when it is unsure.
-
No lock-in, plain documentation
Every tool comes with a short guide to what it does, how to run it and how to change it, written for the people who will use it.
-
Every action leaves a trail
What the system did, on what input and why is recorded in a form a person can read, so a result can be questioned and a mistake traced to its cause.
-
Trust is extended in steps
We start with a narrow scope and independent checks. A system gets more autonomy only after it has shown, on your own data, that it earns it.
04 · From the workbench
Small tools, each automating one workflow.
A few snippets from our own work. None is extraordinary, and that is the point: each takes a single workflow and automates it, which gives an idea of how we approach the problem. We’re happy to walk through any of them.
Debtalk
Hotkey dictation for the Linux desktop. Speech is transcribed on the computer itself with the faster-whisper engine and typed into whichever application is in use. No audio leaves the machine.
The pattern: dictated notes and meeting records where recordings can’t go to a cloud service.
Python · faster-whisper · Source on GitHub
vidchat
Turns a long video into a research session: a structured summary, follow-up questions answered from the source, and an explicit decision about what is kept or deleted afterwards.
The pattern: getting the substance of recorded calls, webinars and training material, with answers tied to the source.
Python · Gemini Notebook · Source on GitHub
Procedure-driven assistant
An AI assistant that works from written procedures: scheduled routines, voice-note transcription and report publishing. Risky actions need explicit approval and every change is logged.
The pattern: giving an AI agent real work safely, with documented procedures, human approval and an audit trail.
Python · scheduled jobs · LLM agents
Portfolio monitor
A P&L analysis tool for option and volatility trading: broker trade files in, positions and profit and loss out, with FIFO accounting and an automated test suite.
The pattern: replacing a fragile spreadsheet with a small, tested application built on data you already export.
Vue 3 · FastAPI · SQLite · pandas
05 · Background
Twenty years of technology where mistakes are expensive.
Solve for Time is led by Fabien Vernaz. He spent most of his career in financial technology, where systems have to run every day and errors cost real money. He started as a software consultant, deploying risk-management software for financial institutions and training their users across Europe, Asia and the Americas.
He then ran support and development for an investment fund in London, and went on to a decade as a chief technology officer: first at an investment firm in Geneva, responsible for its technology, budget and teams, then as co-founder of a start-up applying AI and natural-language processing to financial analysis. He also advised a US investment firm remotely on quantitative projects.
More recently he trained as a coach and taught mathematics in a secondary school. The thread through all of it: understand the problem properly, build something that works, and make sure people can use it.
Why “Solve for Time”? In algebra, to solve for t is to isolate the one unknown that matters. For most businesses, that unknown is time.
- 2024–2025Mathematics teacherSecondary school, France
- 2021–2024Certified coachIndependent practice
- 2019–2024Partner, quantitative advisoryUS investment firm, fully remote
- 2016–2019CTO and co-founderFinTech start-up, Geneva: AI and NLP for financial analysis
- 2008–2016Chief technology officerInvestment firm, Geneva: all technology, budget and teams
- 2006–2008VP support and developmentInvestment fund, London: team across London, New York and Bangalore
- 2000–2006Software consultant and business analystFinancial software vendor: client deployments and user training worldwide
- Education
- Engineering degree (MSc equivalent) in networks and information systems, CPE Lyon
- Also
- MITx cybersecurity course; certified life coach
- Languages
- French and English
- Based
- Haute-Savoie, minutes from Geneva. On site around Geneva, remote elsewhere.
06 · Contact
Tell us what’s slowing you down.
Send a few lines by email. You’ll get a personal reply from Fabien, and if it’s something we can help with, we’ll arrange a two-hour working session, on site or by video.
contact@solvefortime.com
or message Fabien on LinkedIn. We work in English and French.
Useful to include
- The process or problem, and who it affects
- How often it happens, and roughly what it costs today
- Which systems, tools or files are involved