Use Cases

What it actually looks like, six times.

Six projects delivered for manufacturers. All anonymized. No client logo, no quote identifiable: until we have written authorization, we publish nothing. Confidentiality is also a contractual commitment.

Case 01 3 weeks

Full Data and AI Diagnostic

European premium-equipment manufacturer, ~3.5M euros revenue, thirty employees.

Nine on-site interviews with management, sales, production and procurement. Mapping of the existing IT system. Identification of fourteen candidate use cases, prioritized on a value-feasibility matrix. 24-month roadmap in four waves. Quick wins deployed by month 3. Stack chosen: mixed cloud (managed LLM) for non-sensitive tasks plus local models for industrial data.

Results

  • 14 quantified, prioritized use cases
  • 24-month budgeted roadmap
  • Quick wins deployed by month 3
  • Measurable ROI by month 12
Case 02 6 weeks

Automatic dimension extraction from 3D CAD drawings

Plastics processor, technical-mold specialist.

A task that blocked two designers for three weeks on a 55-drawing batch. Vision model deployed locally to preserve trade secrets. Full pipeline handed over to the engineering office team with training, documentation and non-regression test set. The client can now process similar batches in less than a day, fully autonomously.

Results

  • From 3 weeks to under one day per batch
  • 100 percent local vision model, zero cloud data
  • Pipeline transferred and maintained in-house
  • Full cost amortized on the first batch processed
Case 03 10 weeks

Automation of first-article inspection reports

Multi-site metallurgy manufacturer.

Word report generated automatically from ERP bill of materials, CMM measurements and material certificates. Fully local LLM, hosted on client servers, zero data leaving the plant. Integration with existing quality workflow. Administrative delay on shipments, which regularly blocked critical deliveries, was reduced by more than 70 percent over the first two months.

Results

  • Minus 70 percent administrative shipment delay
  • 100 percent local LLM, trade-secret compliant
  • ERP + CMM + materials integration
  • Quality file generated in minutes
Case 04 8 weeks

Technical copilot for pre-sales support teams

Industrial security-opening manufacturer.

Unified search across 20 years of technical documentation: product sheets, standards, after-sales feedback, historical client files. On-premise vector database, role-based access control, full query audit. Average response time on complex technical requests divided by four. New hires reach autonomy in half the time.

Results

  • Response time divided by 4
  • 20 years of documentation queryable
  • On-premise vector base, full audit
  • New-hire time-to-autonomy minus 50 percent
Case 05 12 weeks

ML-based optimization of a multi-head filling machine

Regional food processor.

Model trained on existing production data to continuously optimize machine settings. Reduced volume variability, lower material overconsumption on the priority format. Six Sigma statistical training for production teams in parallel to lock in the gain and give them tools to extend it to other formats without us.

Results

  • Volume variability significantly reduced
  • Material overconsumption down on priority format
  • Teams trained in Six Sigma in parallel
  • Extension to other formats possible autonomously
Case 06 8 weeks

Sensor data processing pipeline — real-time aggregation

Multi-site mid-cap manufacturer, continuous and discrete process.

Fifteen sensors (temperature, vibration, pressure, part counting) collected via Modbus TCP and OPC-UA. Edge pipeline: EWMA aggregation (lambda = 0.2) for the continuous process, Xbar subgroups for the discrete one, triplet storage (min, mean, max) per minute in TimescaleDB. Real-time Grafana dashboards plus ML features auto-computed for the predictive maintenance model. Aggregation reference documented and transferred — the client data team manages new sensor onboarding autonomously.

Results

  • 15 sensors integrated via Modbus TCP + OPC-UA
  • Data volume reduced 600x with no loss of useful signal
  • ML features auto-computed for predictive maintenance
  • Aggregation reference transferred to data team

FAQ

Frequently asked questions

Why are the use cases anonymized?

Industrial discretion. Company names, logos and sensitive data are only published with explicit written client authorization. Five projects are documented with context, problem, solution, stack used and quantified results.

What size of company is typical?

French manufacturing SMEs and mid-cap companies. Sectors represented: textile, paper, plastics, metallurgy, food processing, composites. Typical headcount from a few dozen to several hundred employees.

How long does a typical project take?

Data & AI Diagnostic: a few weeks. Use case integration: from a few weeks to several months depending on complexity. Long transformation roadmap: spread over two years with quarterly milestones.

What does the client get at the end?

Full source code, trained models, technical documentation, team training, skills transfer. No dependency on BCUB3 to operate or evolve the delivered solution.

How are you paid without creating dependency?

Flat fee on the diagnostic, day rate on integration, day rate on training. Public pricing, written framing before each mission. No royalty or recurring license on the code delivered to the client.

Do you take projects beyond the five shown?

Yes. The five cases illustrate method and diversity, not scope. Any industrial or cross-functional use case documented on the Expertise page is eligible. If your need isn't there, a forty-five-minute framing call qualifies feasibility.

Your case looks like one of these?

Or you have something completely different? 45-minute call, we see if it applies, we honestly tell you whether it's for us.

Book a framing call