werk 09 · applied AI in production

AI.

Applied intelligence in production — generative and predictive models put behind a product decision, with the data lineage, the evaluation and the responsibility that a regulated estate demands.

FamilyScale with AICarried byAdvisoryCarried byInterimProven lines4 / 9Detailed mandates4SectorsFinance & Insurance, Tech & Telecom

Werk 09 — who carries it

arms and mandates

An AI feature is a product decision wearing a model. The engineering that decides whether it survives contact with production is upstream and downstream of the model: the data it is fed, the evaluation that gates it, and the trace it leaves.

The offer

what is bought, and in what shape

Models in production behind a real decision, with the lineage, evaluation and governance a regulated estate has to show — not a pilot that never left the lab.

AI portfolio and business case

Use cases scored on value, feasibility and risk, with the two or three that justify a platform investment identified and sequenced.

Production AI build

Generative or predictive models integrated into the product path, with retrieval, prompts or features engineered against the real estate.

MLOps and model lifecycle

Training, versioning, deployment, monitoring and rollback, so a model in production is an operated asset rather than an artefact.

Evaluation and responsible AI

Golden sets, offline and online evaluation, explainability, bias review and the governance record an auditor can read.

AI diagnostic3 to 6 weeks

Use-case portfolio, data readiness verdict, platform decision.

AdvisoryFractional AI authority

Architecture, vendor arbitration, governance and risk posture.

Interim seat6 to 18 months, Chief Data & AI Officer

The AI portfolio delivered and operated in production.

DataikuSnowflakeNeo4jSodaMLflow-style lifecycle toolingModel evaluation harnesses

State of the proof

counted from the ledger below
9sub-capabilities
4proven · a published case carries a sourced figure
3held · carried by a named operator, no case published
2declared · in scope, no published proof today

Experiences that prove it

4 detailed mandates

Sub-capabilities and evidence

9 lines, each with its state
Generative AI in productprovenPeetchr
Applied machine learning and predictive modelsheldPhoeniks venture: an APSIM-DSSAT-ORYZA model ensemble for agricultural yield prediction.
Big data platforms for AI workloadsprovenAXABNP Paribas — FLOA
MLOps and model lifecycleprovenBNP Paribas — FLOA
RAG and knowledge retrievalheldRetrieval over a knowledge graph built from a 15,500-file engineering corpus.
Copilots and workflow automationprovenOrange Business
Fine-tuning and model evaluationdeclared
Responsible AI and model governanceheldOn-Kare: model versioning, training-data lineage, incident cases and explainability specified as platform-level aggregates.
AI due diligence and asset valuationdeclared

Publishable figures

named, sourced, attributable
2.5Months to market · blockchain parametric insuranceErwan Deschamps · Group CTO Digital & Big Data, AXA (2017-18)
-40%Delivery lead time · AXA digital factoryErwan Deschamps · Group CTO Digital & Big Data, AXA (2017-18)
1French DeepTech label · PeetchrErwan Deschamps · founding CPTO and AI founder, Peetchr (2024)

Questions

answered, in the open
What has the collective actually shipped?
A blockchain parametric flight-delay insurance product built at AXA and taken to market in two and a half months, on a group digital and big-data estate that was industrialised into a software factory in the same mandate. A conversational AI platform at Orange Business carrying 149 user paths. An AI conversational recruitment SaaS built from scratch at Peetchr, carrying a French DeepTech label.
What does the AI work depend on?
The data werk, always. A model on an ungoverned estate produces confident output from unverified input, which is worse than no model at all — it launders a data problem into a decision. AI mandates therefore start on the data platform and the lineage, not on the prompt.
What is deliberately not claimed here?
MLOps tooling and fine-tuning are declared, not proven: they are in scope for a mandate and no published case carries them yet. Model research is out of scope entirely. The collective applies and governs models; it does not build them.

Related werks

same family, shared proof
Discuss a ai mandateHow Advisory runs itAll eleven werks