In-house training vs. open AI courses: what works for the Mittelstand

In-house training or open AI course is rarely a didactic question, it is a procurement decision. Purchasing and L&D want to know: what does it cost per head, is it eligible for funding, does it fit our framework agreement, and do we get the proof of competence. This post answers the question from exactly that perspective, not a trainer's.

I (Sebastian) regularly see in our engagements that the format question gets decided too late and on the wrong criteria. Whoever looks only at the day rate misses the two most expensive line items: the staff travel time and the transfer loss when what was learned does not fit your own codebase or tool stack.

The 5 criteria that really decide

It is not the list price that decides, it is these five axes. On each axis the answer tips one way or the other.

1. Number of participants. From roughly five to six people from the same team, in-house undercuts the per-head cost of an open course, because the day rate spreads across more heads. Below that, the open course is cheaper.

2. Context binding. The more the training depends on your real codebase, your data policy and your tool stack, the more clearly in-house wins. A generic open course cannot deliver that because the context is missing.

3. Scheduling and availability. Open courses have fixed dates, good for individuals and bad when a whole team has to be available at once. In-house sets the date around your calendar but needs lead time.

4. Proof and compliance. If the training has to deliver the AI Act proof of competence (Art. 4), per-employee documentation is decisive. Both formats can do it, but in-house bundles it more cleanly for the whole workforce.

5. Funding eligibility. QCG funding (see below) does not depend on the format but on the measure duration and content. What matters is that the format fits the funding logic, more on that below.

When in-house training wins

In-house pays off when a whole team with a concrete productivity goal is to be trained, on its own codebase and with its own data policy. The output is not just knowledge, it is a shared way of working, a pattern or rules repository, and a proof for the whole group. How we set up in-house projects is under how we work.

Typical in-house cases: a development team that wants to make Copilot, Cursor or Claude Code productive, or a whole department that needs the AI Act proof of competence.

When the open course wins

The open course is the right choice for individuals, for a first look, for staff spread across several locations, and for the baseline literacy tier that is the same for everyone. The content is the same as in-house, the context is more general. You can find our open formats under our courses.

Rule of thumb: individuals and first tests go into the open course, a whole team with context binding and a productivity goal gets in-house.

The procurement vocabulary that belongs in the offer

So that purchasing and L&D can compare cleanly, an offer should state these points: per-head cost and participant-days, an optional framework agreement for recurring waves, cancellation terms, the funding logic (QCG eligibility), the proof artefacts (participant certificate, workforce overview) and a clear success measure instead of mere attendance. With these points in the offer, you can honestly compare in-house and open against each other instead of just comparing day rates.

Funding through the Qualifizierungschancengesetz

Both formats can be eligible through the German Qualifizierungschancengesetz (Section 82 SGB III) when the measure teaches skills beyond short-term, purely job-specific adjustment. The funding rate for course costs is staggered by company size, with thresholds at 50 and 500 employees: under 50 employees up to 100 percent (the employer cost contribution can be waived), 50 to 499 up to 50 percent, 500 and more up to 25 percent. The wage subsidy is staggered at 75, 50 and 25 percent.

The decisive point for the format choice: the funded measure must exceed 120 hours. A single open course day or a short in-house session does not reach that. It becomes eligible when the training is embedded in a larger qualification programme. The circulating 10/250/2,500 tables are wrong, the numbers are in Section 82 SGB III. The full overview is in AI funding for the Mittelstand 2026. (As of May 2026)

Decision matrix

Situation	Recommendation
Single person, first test	Open course
Distributed team, same baseline literacy	Open course
Team of 5-6+, productivity goal	In-house
Binding to own codebase / data policy	In-house
AI Act proof for whole workforce	In-house, baseline tier possibly open
Over 120 hours, QCG funding planned	Programme (formats combinable)

FAQ

Is in-house always more expensive than an open course? Per head, no. From roughly five to six participants, in-house undercuts the per-head cost, plus travel time falls away. The comparison must include travel time and transfer loss, not just the day rate.

Can we mix formats? Yes, and often that is the best solution: baseline literacy as an open course for everyone, role and deep literacy as in-house for the teams that should become productive.

Do we get the proof of competence with both? Yes, provided the offer states the proof artefacts. In-house bundles the workforce overview more cleanly.

Is a single course day eligible for funding? Through the QCG no, the 120-hour threshold applies there. It becomes eligible as part of a larger programme. (As of May 2026)

Read more

• AI Act training: proving AI literacy under Art. 4
• GitHub Copilot training for engineering teams
• Cursor training: from tool access to a productive agentic workflow
• Rolling out Claude Code in the team
• AI funding for the Mittelstand 2026
• Workforce AI training pyramid (Bitkom)
• AI skills in the team: the 2026 role shift

Sources

• Section 82 SGB III, https://www.gesetze-im-internet.de/sgb_3/__82.html; Federal Employment Agency, administrative directive on Section 82 SGB III, effective 01.01.2026
• EU AI Act, Regulation (EU) 2024/1689, Art. 4 (AI literacy)
• Sentient Dynamics workshop aggregate (DACH Mittelstand clients, 2025-2026)

How Sentient Dynamics can help

We help purchasing and L&D do the format decision properly: an offer with per-head cost, participant-days, funding logic and proof artefacts, plus a recommendation per use case instead of a blanket answer.

Book a demo