Delivery Model Teardown · two weeks · fixed fee
Find what's slowing your AI-assisted delivery, in two weeks.
A fixed-fee review of your last ten initiatives. You get the funding rules, approval gates and role boundaries that hold delivery back, and who owns changing each one. For fintechs, banks and banking-IT providers.
Book a 30-minute call30 minutes. No pitch. You'll know whether a Teardown fits.
Led by Vitalijus Lykovas: 16 years in Nordic and Baltic banking at Swedbank, Luminor and Bankdata, including two project-to-product transitions run from the inside.
What the Teardown includes
How it works. You share what you already have on ten recent initiatives. We trace each one through your actual decision path, from idea to release, and mark where it waited and why.
What you get. The specific funding lines, role boundaries and approval gates that would block AI-assisted delivery, and who owns changing each one.
Price. Fixed, agreed before work starts.
For fintechs
Your teams use AI coding tools, but release lead times haven't moved, and an auditor may ask how AI-written changes are approved. We trace where funding, approval and ownership slow the work. If DORA is the trigger, the three-week Control Map maps your AI delivery against it.
Book a 30-minute callFor banks and banking-IT providers
Pilots work in a sandbox and stall on shared core systems, often during a merger, a core migration or a change of owner. We name which team owns each decision and where approvals wait. Our founder led this change from the inside, at Luminor and at Bankdata.
Book a 30-minute callAfter the Teardown
If you want help fixing what we find, an optional eight-week installation puts the changes in place. You decide after the Teardown.
How the installation works
The Installation — eight weeks. The evidence gate: no build without a measurable, evidence-backed problem. Work classification: run, grow, transform, debt — typed before anything enters a queue. The product–technology leadership contract that removes the translation layer between "what" and "how". A decision cadence of fixed sixty-minute sessions in place of steering committees.
Control Map
Three weeks, fixed fee, for regulated companies. Your AI delivery mapped against DORA obligations and EU AI Act classification — ending with the missing controls, the evidence each one requires, and the named role who signs it.
Tekhouse also builds what the Teardown finds is missing: see Agentic Delivery, and Software for Robot Cells.
Who leads it
Vitalijus Lykovas, founder. Sixteen years inside Nordic and Baltic banking at Swedbank, Luminor and Bankdata. Two project-to-product transitions run from the inside. Direct work with regulators and central banks in four countries on PSD2, AML and card-scheme compliance.
On the record.
Bankdata's CEO on hiring the company's first English-speaking director, responsible for Card & Payments.
ITWatch (2023)"Bankdata henter ny direktør til betalingsområdet."
Århus Stiftstidende (2023)"Ny direktør i virksomhed med milliardomsætning er hentet i Vilnius."
20+ recommendations on LinkedIn →
Our own agentic delivery system runs in production. We'll show you a real cost-ledger extract before you commit.
What is a product operating model?
The system that decides how work is chosen, funded, staffed and shipped — as opposed to what tools teams use. A product operating model funds outcomes and teams; a project model funds temporary initiatives. The distinction determines whether AI capacity compounds or evaporates.
How is an AI-native operating model different from a normal product operating model?
A standard model assumes people do the building. An AI-native one adds three things: an evidence gate strong enough to be worth automating behind, delivery capacity measured in cost-per-outcome rather than headcount, and governance that can account for work no human wrote.
How long does a transition from project to product take?
The structural change — roles, gates, funding — lands in a quarter. Behaviour follows over 12–18 months, and only when the old model is actually retired rather than run in parallel.
Do we need to reorganise before adopting AI delivery?
No — and doing it in the wrong order is the most common failure we see. Install the evidence gate and work classification first; they work in any structure. Reorganise once you can see where the work actually flows.
Is the Teardown a lead-in to a big project?
No. It stands on its own, and the installation is a separate decision you make afterwards.
Do we need to be using AI already?
It helps; that's where the stalls show.
Who is it not for?
Teams looking for a tool recommendation or a chatbot build.
A note on lineage: the vocabulary here was popularised by Marty Cagan and SVPG. The model is our own, built in regulated institutions where "just empower the teams" meets risk committees.
Thirty minutes. We come prepared.
Tell us what's stalled. No pitch, no obligation.