Trench Group case study

4 min read

How Trench Group deployed 3 Claude-powered applications in production

Claude, Claude Code and Cowork at the Trench Group AI Hackathon, Berlin, 22 to 24 September 2026

Trench Group and AI Partners brought 33 managers and engineers together for three days to apply AI to real Trench processes, on real Trench data. Twelve teams built twelve applications, and three now run in production: a quality and EHS cockpit, automated reporting commentary and an SAP and LucaNet reconciliation engine.

12

applications built in three days

3

applications in production

8h → 1h

for the monthly reporting commentary

200+ h

of recurring work saved a year

Key facts

Industry: Industrial manufacturing.

The challenge: Claude was already deployed at Trench, with an internal App Hub to run applications. The ambition was to move headquarters from experimenting with AI to using it concretely and responsibly in daily work, with clear use cases and measurable relief for the teams under the heaviest load.

The solution: A three-day build hackathon designed and run by AI Partners: use cases scoped before Berlin, one day to learn, one day to build, one day to ship and pitch, with AI Partners coaches embedded in the teams.

Key results: Three Claude-powered applications deployed in production and integrated into Trench Group's operational workflows: automated monthly reporting commentary (8 hours reduced to 1 hour), SAP and LucaNet reconciliation (10 hours saved per month), and a quality and EHS cockpit replacing manual data consolidation.

The challenge

From experimenting with AI to using it every day

Trench Group had already done the groundwork. Claude was deployed, an internal App Hub was in place to host applications, and opportunities had been identified across the business.

The next step was harder: turning that potential into applications people use in their daily work. The goal set for the hackathon was explicit: apply agentic AI to one real Trench workload per team, hands-on, with the teams' own data, and leave with something that runs.

The approach

Learn, build, ship

AI Partners designed the hackathon around one rule: no ideation day. The use cases were scoped with Trench before Berlin, so that every team could start building from the first hour.

Day 1

learn

keynotes from Trench Group IT and AI Partners, then hands-on Claude sessions, with a business room and a technical room running in parallel.

Day 2

build

four sprints on real data, accessed through the App Hub services and MCP connectors.

Day 3

ship

final fixes, rehearsals and twelve six-minute pitches to the jury.

A team of five from AI Partners took part, four of them embedded as coaches rotating across the teams, with Trench Group IT acting as a floating resource. Participants worked with Claude, Claude Projects and Skills, Cowork, Claude Code and MCP connectors, each team on the tools that matched its level.

Three applications in production

Following the hackathon, three of the twelve applications were deployed into Trench Group's production environment, moving beyond experimentation to support real business processes.

AC Commentary

the monthly reporting commentary, automated

The application assembles the monthly reporting commentary automatically from the reporting data. The work that took 8 hours each month now takes 1 hour, around 84 hours a year.

ReconIQ

SAP and LucaNet reconciliation

ReconIQ finds the gaps between SAP and LucaNet and emails the factory concerned directly. It saves 10 hours each month, around 120 hours a year, on a task that used to be done by hand.

Operations Cockpit

quality and EHS in one view

A quality and EHS dashboard built over Databricks and SharePoint, which replaces manual consolidation and saves one person-week of work.

Together, the three applications take more than 200 hours of recurring work a year off the finance and operations teams.

Nine more applications in the pipeline

The other teams built applications that are now being taken towards production, including a finance chatbot on the budget templates, a PLM milestone approval portal, a single source of truth on sales ownership, a headcount approval gateway and a Teams briefing agent for marketing requests.

What the participants said

Satisfaction: overall satisfaction 8.6 out of 10, an NPS of +56, and no score below 7 (nine responses out of 33).

Motivation: an average of 8.9 out of 10 on feeling inspired to use AI more in daily work, with seven of nine respondents at 9 or 10.

Autonomy: 89% of respondents say they can now write effective prompts and use Claude Projects on their own, 78% use Skills and Cowork, 67% Claude Code.

Coaching: coaching next to each team on its own use case scored 4.2 out of 5, against 3.6 for the training sessions.

What comes next

With three Claude-powered applications already in production and nine additional use cases progressing toward deployment, Trench Group is now focused on scaling AI adoption across its operations.

Connecting Claude to company data and designing agent workflows are the two skills the teams now want to deepen, so that Trench can build the next applications on its own.

What they say

"Excellent team and support!"

Participant, Trench Group AI Hackathon

"Quick 45-60 min workshops in between are a great idea."