AI Product Lab · CPO · CTO · Product · Engineering

Decide what to build. Then build it differently.

In two days from AI use case to a production-ready decision – on the real backlog, the real architecture and, where cleared, the real code.

Triage a use case now

Thirty minutes with an operator. Then you get scope and a fixed price.

For
  • CPO and CTO
  • Head of Product, Head of Engineering
  • Product managers and product owners
  • Engineering managers and tech leads
  • Cross-functional product teams

The starting point

More generated code is not a product outcome. Without solid problem evidence, AI-ready data, quality gates and an operating plan, AI may only be accelerating the wrong part of the system.

Evidenced by

In the AI Monitor, usage sits at 66 out of 100. The organisational groundwork beneath it sits at 50. People moved faster than data, integration and governance.

Dream outcome

The team knows which AI use cases get built now, tested first, deferred or deliberately stopped – and one of them is specified production-ready.

Your first result

55 working days after a complete intake

Product and tech red-flag map

Within five working days of a complete intake, the team receives a red-flag map with the three biggest uncertainties for the decision sprint.

Decision snapshot

Triage a use case in two minutes.

Pick an archetype and rate four dimensions. The verdict appears immediately – with the criteria that produced it.

AI Use-Case Decision Snapshot

Your rating
Impact

How much does this move a result you already measure?

Feasibility

How confident are you that you can build and run this on today's stack?

Evidence

How well evidenced is it that users actually have this problem?

Risk

How well do you control the consequences of a mistake? High means: well controlled.

Basis: 162 valid responses, 106 of them with demographic detail. AI Monitor 2026, published by Teklens with ETH Zurich and the University of St. Gallen. Self-assessments, normalised to 0–100. Pre-release.

What the Decision Lab includes

Every component names the obstacle it removes, the artefact you receive and when it is delivered.

Lead time: about 10 working daysLab: 2 intensive daysParticipants: ideally 5–10Roles: product, engineering, data and security

We work on the real backlog and the relevant technical material.

  1. Obstacle

    Signals, decisions and technical reality sit apart

    Artefact

    A shared baseline from backlog, architecture, data and code context

    Delivered

    before the lab

Included accelerators

  • AI evaluation starter
  • Data readiness checklist
  • Kill criteria template
  • Human-in-the-loop decision guide

Templates created during the lab that your team keeps using afterwards. We don't attach invented individual values to them.

When the evidence holds: Execution Track

6-Week AI Product Shipping Sprint

The Decision Lab plus six weeks of execution on the prioritised work – in small batches, with review and evaluation.

  • Execution of 1–2 prioritised pieces of work, where scope and readiness are confirmed
  • Small batches with review, tests and evaluation
  • Monitoring plus cost and failure handling
  • Updated Production Passport
  • Handover of the routines to the product team
  • Updated Context Workspace

We don't promise «two shipped features» as a blanket commitment. What ships in six weeks depends on scope and technical dependencies – we settle both in Decode before committing to anything.

Proof

Problem proof

  • 66 vs 50
    AI usage runs ahead of the groundwork beneath it: data, integration, governance and the ability to adapt.
    Basis: 162 valid responses, 106 of them with demographic detail
    Limitation: Self-assessments, normalised to 0–100. Pre-release.
  • 60%Projection
    of AI projects will be abandoned through 2026 for lack of AI-ready data.
    Source: Gartner (2025)
    Basis: Analyst projection through 2026
    Limitation: A projection, not a measured figure.
  • −19%
    That is how much slower experienced open-source developers were with AI tools – while feeling faster.
    Source: METR (2025)
    Basis: Randomised controlled trial with experienced open-source developers on their own repositories
    Limitation: Small sample, very specific context. Not transferable to every kind of task – the finding about self-perception is the most robust part.

ETH Zurich and the University of St. Gallen are research partners on the AI Monitor. They are not Teklens reference customers.

Mechanism proof

Anonymised exampleAI Production Passport

Anonymised example · B2B SaaS, around 40 engineers

GateCriterionStatusOwner
ValueMetric defined and measured todayMetPM
DataCoverage, freshness and provenance settledOpenData eng
ReliabilityEvaluation set and error budget in placeMetTech lead
SecurityData residency, roles and logging settledMetCISO
OperationsMonitoring, cost and rollback definedOpenPlatform

Delivery proof

What we have actually delivered: cases, customer quotes or an anonymised before-and-after. Where nothing is cleared yet we show a clearly labelled anonymised example – not an empty placeholder.

We do · You provide

We do

  • Product and tech synthesis
  • Preparing the use-case triage
  • Facilitation and decision artefacts
  • Production Passport and roadmap
  • Workspace structure

You provide

  • The real backlog
  • Architecture and data material
  • Optionally cleared code access
  • Two lab days with the core team
  • Named product and engineering owners

Investment

Two options, one scope. The Execution Track sits first because it shows the full extent – the recommended entry is the Decision Lab.

Recommended entry

2-Day AI Product Decision Lab

Diagnosis, a shared decision and a concrete execution plan. For teams that need clarity before they commit capital or capacity.

  • Role-specific diagnosis
  • First-value snapshot
  • One to two workshop days
  • Decision Pack
  • Context Workspace
  • 90- or 100-day plan
First result
5 working days after a complete intake
Full result
5 working days after day 2
Effort on your side
2 lab days plus preparing the material
Fixed price after a 30-minute scoping call

The scope is set; the price is fixed in the scoping call and confirmed in writing. No line items appearing later.

Decision Pack guarantee

When the agreed inputs are complete, we deliver the defined Decision Pack by the agreed date. If an agreed component is missing, we keep working at no extra fee until the pack is complete.

The team knows which AI use cases get built now, tested first, deferred or deliberately stopped – and one of them is specified production-ready.

Review a use case with an operator
Full system

6-Week AI Product Shipping Sprint

The Decision Lab plus real execution, reviews and embedding into how the team works.

  • Everything in the Decision Lab
  • Execution on the real portfolio, backlog or GTM system
  • Agreed working assets or agents
  • Quality and governance gates
  • Review cadence
  • Handover and ongoing development in the Context Workspace
First result
5 working days after a complete intake
Full result
5 working days after day 2
Effort on your side
2 lab days plus preparing the material
Fixed price after a 30-minute scoping call

The scope is set; the price is fixed in the scoping call and confirmed in writing. No line items appearing later.

Decision Pack guarantee

When the agreed inputs are complete, we deliver the defined Decision Pack by the agreed date. If an agreed component is missing, we keep working at no extra fee until the pack is complete.

The team knows which AI use cases get built now, tested first, deferred or deliberately stopped – and one of them is specified production-ready.

Review a use case with an operator
Decision Pack guarantee

A decision you can act on – or we keep working.

When the agreed inputs are complete, we deliver the defined Decision Pack by the agreed date. If an agreed component is missing, we keep working at no extra fee until the pack is complete.

We guarantee defined outputs and dates that we control. No revenue, transformation or autonomy guarantee – that depends on your execution, and we can't stand behind it.

A Decision Pack counts as complete with
  • a validated baseline
  • prioritised decisions
  • reasoned wait and stop decisions
  • named owners
  • success criteria or metrics
  • a roadmap and the next review point
  • an updated Context Workspace
Prerequisites
  • Complete inputs by the agreed date
  • The agreed stakeholders take part
  • Decisions stay within the agreed scope
  • Access and permissions are in place
  • No external blocker prevents the work

Fit gate

After Decode and Shape, you decide – not us.

The Execution Track does not start automatically. A reasoned wait or stop is a valid result – and often the most valuable one.

  1. Amplify

    The evidence holds. We start the Execution Track.

  2. Experiment

    A bounded test resolves the open uncertainty before capacity is committed.

  3. Wait

    Fix the groundwork or the input first. With a criterion and a review date, not as a postponement.

  4. Stop

    No sensible business case. Recording that with reasons saves more than any pilot.

Shared operating model

Decode → Shape → Amplify → Evolve.

A loop, not a transformation with an end date.

  1. 01

    Decode

    · Discover

    Understand reality, context and the bottleneck

    Where do we actually stand?

  2. 02

    Shape

    · Define

    Set decisions, priorities and limits

    What do we change – and what don't we?

  3. 03

    Amplify

    · Build

    Turn decisions into real work

    How does measurable impact happen?

  4. 04

    Evolve

    · Operate

    Embed learning, governance and routine

    How does the system get better?

The Decision Lab covers Decode and Shape and hands over to Amplify and Evolve. The Execution Track runs the whole loop.

Your company leaves with a system, not a slide deck.

Every lab leaves behind a versioned Context Workspace: evidence, decisions, work, rules and what you learned, in one place. Human-readable, machine-readable for approved agents, exportable.

Minimum requirements
  • Versioned, with traceable sources and assumptions
  • An owner per artefact; decisions dated and reasoned
  • Human-readable and machine-readable for approved agents
  • Exportable – no artificial lock-in
  • Permissions and data residency settled, update rhythm defined in Evolve
Structure
/product-ai/
  00_customer_context/
  01_opportunities/
  02_use_cases/
  03_architecture_data/
  04_delivery/
  05_operate/
  06_roadmap/

Common questions

Not sure which Lab fits?

Thirty minutes with the founding team. You get a clear recommendation, the right scope – or a reasoned no.

All labs