HomeServicesLegacy Modernisation
Legacy → AI-Ready

Your most valuable data is locked inside systems no one knows how to evolve anymore.

We unlock it and make it AI-exploitable. No big bang. No production downtime. We release the value of your core legacy instead of rewriting everything.

4 wks
to map your legacy and identify your priority AI use cases
The reality

Why your AI projects stall

Critical systems 20 to 40 years old

COBOL, AS/400, IBM Z mainframe, DB2, VSAM, CICS, JCL… Your core applications have been running for decades. Stable but frozen, and the teams who knew how to evolve them are retiring.

Data trapped in legacy

Your most valuable data is locked in proprietary formats, obsolete interfaces, and nightly batch jobs. It exists but stays out of reach for modern AI models in real time.

AI projects failing for lack of data

Your AI POCs shine in demos and collapse in production because the data doesn't arrive, arrives too late, or arrives malformed. The bottleneck isn't AI — it's data access.

Methodology

Our approach in 3 phases

A progressive, costed trajectory from diagnosis to AI deployment, without disrupting your operations.

01
4 weeks

AI-Readiness Audit

Full mapping of your legacy, identification of AI-exploitable data and technical blockers, 18-month costed roadmap and priority use cases.

  • Application & data mapping
  • Technical blockers identified
  • 18-month costed roadmap
  • Priority AI use cases
  • C-level executive readout
PrixOn request
02
2 to 6 months

API encapsulation & extraction

We don't touch the core. We make it accessible: API gateway in front of the legacy, data extraction to a modern layer usable by your AI models and next-gen applications.

  • Secure API gateway
  • Real-time pipelines (Kafka, CDC)
  • Modern data lake / warehouse
  • API & data contract docs
  • Monitoring & SLA
PrixOn request
03
6 to 24 months

Progressive modernisation + AI

Progressive rewrite of components per business domain (strangler pattern), microservices replacing legacy modules one by one, generative AI layer and autonomous agents plugged into reliable data.

  • Microservices per domain
  • Strangler pattern, no big bang
  • AI overlay (RAG, agents)
  • Optional sovereign cloud
  • Continuous run & support
PrixOn request
Stack

Technologies we master

Legacy

COBOLAS/400IBM Z MainframeJCLCICSDB2VSAMPL/1

Modernisation

API GatewayJava / SpringMicroservicesKafkaKubernetesgRPCDebezium

Cloud

AWSAzureGCPOVH (sovereign)OutscaleScaleway

AI

OpenAIAnthropicMistralRAGAutonomous agentsLangChainVector DB
Legacy heritage

Decoding your technology heritage

The legacy technologies our teams master, their history, and the concrete trajectory to make them exploitable by artificial intelligence.

Modernizing a legacy information system is not just about replacing it. Most IT decision-makers we work with discover that their core systems rely on technologies first released in the 60s or 70s — COBOL, JCL, CICS, IMS, VSAM, AS/400, PL/1, NATURAL — yet still process billions of euros of transactions every single day. These systems are reliable, fast, and critical. They are also opaque to modern machine learning, real-time analytics, and generative AI tools.

Our job is precisely to unlock that trapped value. Not to rewrite everything, but to make your data and transactions accessible to LLMs (GPT-4, Claude, Mistral, LLaMA), AI agents, scoring models, digital twins, and next-generation data platforms. Below are the key technologies our teams master, the history of each, and the AI-readiness trajectory that applies.

COBOL

1959 · Banking, Insurance, Public sector

Common Business Oriented Language. Still in production at 70% of the world's banks. Versions ANS 68, ANS 85, COBOL 2002 and COBOL 2014. Over 200 billion lines in operation. Particularly common in end-of-day batches, credit scoring, insurance policy management, and actuarial calculations.

AI-ready trajectory

REST / gRPC encapsulation around COBOL programs, controlled exposure via API Gateway, real-time extraction via Change Data Capture to feed your AI models without touching the transactional core.

AS/400 · IBM i

1988 · Mid-market industry, Distribution

AS/400, then iSeries, then IBM i. Integrated IBM system with native DB2/400 database, RPG programs (II, III, IV, IV ILE), and CL control language. Widespread among wholesalers, manufacturers, and regional distributors. Stable for up to 20 years without downtime.

AI-ready trajectory

Open Access for RPG, modernization of 5250 screens to REST / Web, exposure of DB2/400 to a cloud data warehouse, AI agents plugged into the system via API.

Mainframe IBM Z · z/OS

1964 · Banking, Public, Telecoms

IBM Z platform (z14, z15, z16) running z/OS, formerly MVS and OS/390. Transactional core of major global banks, public agencies, and telecom operators. Runs JCL, JES2, RACF for security; handles tens of thousands of CICS transactions per second with 99.999% availability.

AI-ready trajectory

API encapsulation via z/OS Connect or IBM API Connect, data offloading to a cloud data lake (AWS, Azure, GCP, OVH sovereign), RAG layer for LLMs.

CICS

1968 · Online Transaction Processing

Customer Information Control System. IBM's transactional monitor, omnipresent on the mainframe. Handles online transactions, BMS (Basic Mapping Support) screens, and orchestration between COBOL, DB2, and VSAM. Still at the heart of real-time banking operations.

AI-ready trajectory

Exposure of CICS transactions as REST services via z/OS Connect, integration with Kafka for events, AI agent layer for automated customer service.

IMS · DB2 · VSAM

1966–1983 · Legacy databases

IMS (Information Management System), DB2 on z/OS, and VSAM (Virtual Storage Access Method) make up the bulk of legacy data in production. IMS hierarchical for historical banking, DB2 relational for more recent systems, VSAM for fast indexed files (KSDS, ESDS, RRDS).

AI-ready trajectory

Change Data Capture with IBM InfoSphere or Debezium to export changes to a data lake, vector indexing for LLMs, hybrid SQL + RAG queries.

PL/1 · NATURAL · Adabas

1964 · Banking, Insurance

PL/1 (Programming Language One) for scientific and financial calculations. NATURAL and Adabas (Software AG) for large-account systems in the 80s and 90s, particularly in insurance and industry. Pacbase for structured software engineering. Voluminous code with little documentation.

AI-ready trajectory

PL/1 → Java transpilers assisted by LLM, Adabas exposure via Event Replicator, gradual migration to PostgreSQL or cloud database, AI agents for automated documentation.

JCL · JES · REXX · ISPF

Mainframe operations & batch

Job Control Language to drive batches, JES2 / JES3 for the execution queue, REXX for automation, and ISPF for development. The operational backbone of mainframes. Often thousands of JCL jobs run every night for closings, reporting, and synchronizations.

AI-ready trajectory

Gradual orchestration to Airflow, Argo Workflows, or Dagster, batch replatforming to Spark / Databricks, unified monitoring with Grafana and AI observability.

Tandem · Stratus · Unisys

High-availability systems

Tandem NonStop (HP), Stratus VOS, and Unisys ClearPath. Fault-tolerant platforms used in payments (credit cards, interbank switching), stock exchanges, and telecom operators. Specific architectures with proprietary languages (TAL, COBOL85, PL/I-NonStop).

AI-ready trajectory

Encapsulation via MQ / Kafka middleware, gradual modernization to Java / Spring on Linux, preservation of 99.9999% availability constraints.

Target architecture: your heritage, plugged into AI

Beyond the technologies themselves, the challenge is to build an abstraction layer that makes your heritage exploitable by the modern AI ecosystem. This involves three pillars: (1) a secure API Gateway that isolates the legacy from modern consumers; (2) a Change Data Capture pipeline (Debezium, IBM InfoSphere CDC, Striim, Qlik Replicate) that exports changes from DB2, VSAM, IMS, or Adabas to a data lake usable by your models; (3) a RAG (Retrieval-Augmented Generation) layer or vector store (Pinecone, Weaviate, pgvector, Qdrant) that allows LLMs to query your business documentation, operational procedures, and transactional archives.

This progressive architecture, popularized by Martin Fowler's strangler pattern, allows you to replace one legacy module at a time, starting with the peripherals (front-office, reporting, partner integrations) before tackling the transactional core. At each step, you capture measurable business value: AI conversational agents plugged into real-time data, dynamic credit scoring, fraud detection via anomaly detection, contextualized recommendations, industrial digital twins, and predictive maintenance.

The sectors where this work unlocks the most value are historically banking (core banking, scoring, KYC/AML), insurance (underwriting, claims, actuarial), the public sector (tax, social, registries, citizen portals), telecoms (BSS/OSS, churn, customer service agents), and industry/energy (SCADA, predictive maintenance, energy optimization). In the French-speaking world, our teams operate in France, the Maghreb (Morocco, Algeria, Tunisia), and West Africa (Senegal, Ivory Coast, BCEAO and Bank Al-Maghrib compliance).

Related concepts & technologies

Modernization

Mainframe modernizationCOBOL to APICOBOL to JavaAS/400 modernizationLegacy refactoringStrangler patternAPI encapsulationCode translation LLM

Data & AI

Change Data CaptureDebeziumData lake mainframeRAG over legacyVector storeEnterprise LLMBusiness embeddingsKnowledge graph

Cloud & sovereignty

Mainframe offloadingz/OS to cloudIBM i to cloudSovereign cloudSecNumCloudGAIA-XOVHcloudOutscaleGDPR by design
Why UNIPOLE

What makes us different

Senior PMs + tech

Dual expertise: enterprise PM (banking, insurance, telco) and AI engineering. We speak to COMEX and we code.

Progressive approach

No big bang. Strangler pattern, incremental delivery, business value at every step, measurable ROI.

Trilingual FR · EN · AR

Mastery of three languages, cultural sensitivity for Maghreb and French-speaking Africa, ability to run transcontinental projects.

French anchoring

French company, data hosted in Europe (sovereign option), GDPR confidentiality and compliance by design.

FAQ

Frequently asked questions

Next step

Unlock the value of your legacy

Let's spend 30 minutes discussing your legacy, your AI use cases and the first costed phase we can commit to.