Tacit expertise capture
Seven senior staff leaving within 18 months. Procedures never written down, customer relationships held in one person's memory, tolerances judged by eye. Tacita captured it all before it was too late.
The context
Ferretti Meccanica has made precision components for the oil & gas sector from Brescia, Italy, since 1981. Ninety-two employees, revenue of 26 million euros, customers spread across Italy, Germany, Norway and Saudi Arabia.
Ferretti's strength had always been continuity. The same sales manager for the Nordic market for 19 years. The same workshop supervisor for 23. The same quality manager since the Ospitaletto plant opened in 1997.
In October 2024 the HR manager updated the retirement plan. Seven people would be leaving by March 2026. Three in engineering, two in sales, one in quality, one in materials logistics.
The problem was clear. Those seven people carried skills no internal manual covered. They knew things their colleagues had heard mentioned in the corridor, but that nobody had ever formalized.
"When Mauro retires, who knows that the customer in Stavanger wants the specifications in UNS format and the certificates translated into Norwegian? It's all in his head."
Elena Ferretti, Chief Executive Officer
The problem
01
Three senior operators applied tolerance margins different from the ones written in the technical specifications. They had done so for years, with better results. The reason lay in the specific characteristics of the materials supplied by their usual suppliers. Without documenting it, those margins would have reverted to standard, with a predictable rise in scrap.
02
Mauro Silvestri, head of the Nordic market, managed 14 customers across Norway and Sweden. He knew each buyer's format preferences, their approval cycles, the periods when it paid to ease off commercially. None of that information was in the CRM.
03
Carla Bruni, quality manager, had developed over the years a personal system of visual pre-inspection on incoming batches. That pre-inspection caught defects before they entered the production line. The certified ISO quality system made no provision for it.
04
Giorgio Valli, warehouse manager, knew which suppliers delivered late after August, which accepted orders below the minimum if you called them directly and which applied off-list discounts on specific volumes. The ERP system tracked none of it.
The process
Each phase involved conversational AI interviews, human validation and the production of working documents. Click on each phase for the details.
Week 1 — 3 days
Together with the HR manager and the area heads we identified the critical skills at risk of being lost. For each senior person leaving, we built a map of the knowledge to be explored.
Weeks 1–3 — 14 sessions
Each senior person took part in 2 interview sessions with the AI agent. Sessions lasted between 35 and 55 minutes. The agent followed an adaptive protocol, starting from the mapped areas and digging deeper based on the answers.
Week 3 — 5 days
The content gathered was organized into working documents, validated by the senior staff themselves and by the colleagues who would inherit those responsibilities.
Week 4 — 4 days
The final documents were integrated into the company's systems. Technical procedures into the production ERP, commercial information into the CRM, quality protocols into the ISO system.
The results
The metrics refer to the first quarter after the retirement of Silvestri (sales), Bruni (quality) and Moretti (workshop).
The comparison
Only Mauro knew each buyer's specific document format requirements. His replacement took 3 hours per quote, with an error rate of 22%.
The tolerances for special alloys sat in the heads of three operators. At shift changeover, the parameters reverted to the manual values, pushing scrap up by as much as 15%.
Carla's visual pre-inspection wasn't documented. The quality team worked only from the ISO checklists, catching defects later in the production chain.
Information about how suppliers actually behaved (seasonal delays, flexibility on minimums, off-list discounts) existed only in Giorgio's memory.
38 customer records in the CRM with format requirements, approval cycles and relationship notes. The new salesperson quotes in 45 minutes with an error rate below 5%.
47 operating procedures in the MES with parameters for each alloy and each supplier. Scrap on special alloys fell from 15% to 9% in the first month alone.
Carla's visual pre-inspection protocol was codified into 22 checklists integrated into the ISO system. Defects caught on arrival rose by 34%.
36 supplier records updated with real operational information. The warehouse cut supply delays by 28% in the first quarter.
The voices
Mauro ran the Nordic market on his own for almost twenty years. He knew the buyers by name, he knew when to call and when to wait, he remembered the implicit conditions of every contract.
Over the two interview sessions, the AI agent reconstructed with him how he managed each individual customer. Document preferences, technical formats required, shutdown periods, habits in approval cycles.
"I thought some things just couldn't be explained. But when someone asks you the right questions, you find out you can tell the whole story."
Over the years Carla had built a visual pre-inspection system that appeared in no manual. She looked at incoming batches and could recognize the signs of problems further downstream. Micro-scratches, color variations, packaging defects that foreshadowed defects in the part.
The AI agent worked with her to translate that visual know-how into criteria that could be described. Every signal became a checklist item with a threshold, an action and a reference photograph.
"I did this job for 27 years without ever having to explain it to anyone. Explaining it was the best way to understand how much what I know is worth."
Luca was the workshop supervisor. He set the machines by feel, adapting the parameters to the batch, the material supplier and the ambient temperature. His settings produced less scrap than the ones specified in the technical manuals.
The AI agent reconstructed with him the logic behind the settings for every material-supplier combination. The result was a set of 47 operating procedures with specific parameters, validated by the production manager.
"They asked me why I turn that dial half a turn further on Nimonic. I'd never thought about it. Now it's written down in black and white, with the numbers."
Giorgio managed the relationships with 28 suppliers of critical materials. He knew who delivered late after the summer break, who accepted orders below the minimum without putting it in writing, who granted verbal discounts on certain volumes.
The interview with the AI agent produced 36 supplier records with operational information the purchasing system had no field for. Each record captures actual behavior, not contractual terms.
"Those things, only I knew them. If I'd left without saying them, my replacement would have needed years to piece them back together."
"The project changed the way we think about succession. It used to be an org-chart problem. Now we know it's a knowledge problem."
Davide Reali, HR Manager, Ferretti Meccanica
The deliverables
Documents structured by functional area, ready to be integrated into company systems. Each procedure includes context, steps, known exceptions and references to specific materials or customers.
Operational profiles for the Nordic market with document requirements, communication preferences, decision cycles and implicit conditions. Ready to load straight into the CRM.
Real operational information on the behavior, flexibility and seasonal patterns of suppliers of critical materials. Complementary to the contractual terms held in the ERP.
Visual pre-inspection protocols codified with criteria, thresholds, corrective actions and reference photographs. Integrated into the ISO 9001 system.
Setting parameters for every material-supplier combination, with specific tolerance margins validated against the production results of the past 5 years.
An updated map of the critical knowledge areas still uncovered, with priorities indicated for any future work on other roles.
What we learned
01
People think their know-how can't be put into words. With the right questions, every operational skill becomes describable. What it takes is specific questions about real activities, about exception scenarios and about the reasons behind choices.
02
None of the seven pushed back. All of them welcomed the interview as recognition of their contribution. The process had a positive effect on the internal climate, visible even among colleagues who weren't involved.
03
Company systems track transactions. Operational information (how you work with a customer, how you set a machine, how you handle a supplier) lives outside those systems. Capturing it requires a dedicated process.
04
Ferretti started the project 14 months before the first departure. That made it possible to run validation sessions with the senior staff still on the payroll and to hand things over gradually. Waiting until the last month would have reduced the quality of what was captured.
Let's talk about how to capture operational knowledge before it walks out of the door.
Talk to the team