In AI, the problem is not technical — it is one of judgement
By 2026, almost any technical team knows how to use AI models. Documentation is public, APIs are accessible and the technical entry barrier has dropped radically. What has not dropped is the decision barrier: knowing where to apply AI, where not to, with which stack, against which metrics and with which regulatory defence. That barrier is not crossed with juniors, courses or frameworks. It is crossed with senior judgement accumulated over years of real projects.
In a German utility, this difference is especially visible. AI lands on critical systems (SAP IS-U, FI-CA, Powercloud, market communication) operating under BSI, GDPR, KRITIS and continuous scrutiny from the Bundesnetzagentur. A poorly calibrated decision does not generate a bug: it generates regulatory exposure, expensive rework or loss of internal trust in AI itself.
“In AI, the difference is not using the right model — it is deciding whether the model should be used at all.”
What is lost when an AI project runs with junior or rotating teams
When an AI project in a utility runs with junior profiles without real senior mentorship, or with teams that rotate every six months, three patterns keep coming back.
01 · Over-engineering
The biggest, newest or trendiest model gets applied even when a simpler solution would fit. The pilot becomes expensive to defend to finance and to compliance.
02 · Poor scope calibration
Without track record, the team says yes to everything. A senior recognises when discovery needs to be extended and when to say “this is not an AI problem”.
03 · Loss of continuity
Discarded decisions, validated hypotheses and identified data biases live outside the docs. When teams rotate, the next cycle pays twice for the same learning.
What a Senior Consultant brings that a model or a junior cannot
A Senior Consultant in AI brings three capabilities that neither a model nor a junior can replicate.
✓ Judgement
Knowing where AI belongs (classification, forecasting, anomaly detection, document summarisation) and where it only adds risk (binding decisions, deterministic processes, regulatory-exposed calculations). That reading saves months of wasted piloting.
✓ Sector context
How MaKo works, how BNetzA regulation evolves, how SAP IS-U integrates with FI-CA and the portal, which data can leave the environment under KRITIS or GDPR. Not taught in any LLM course.
✓ Saying “no”
A senior says no to a poorly framed use case before it turns into a lost project. In AI, saying “no” on time protects the client’s ROI far more than saying “yes” with enthusiasm.
Four critical moments where seniority makes the difference
Four moments in an AI project in a German utility determine the final outcome depending on the seniority of the team.
01 · Scope definition
Focusing the project on two or three use cases with demonstrable short-term value, instead of a catalogue of disconnected pilots.
02 · Technology selection
Applying real AI-agnostic criteria: open-source for sovereignty, hyperscaler for scale, small fine-tuned models for cost — chosen per use case.
03 · Regulatory validation
Anticipating BSI, internal compliance and auditor questions, and documenting decisions and architecture at the moment they are made.
04 · Industrialisation
Designing what an operations team actually needs to keep the solution in production for five years, not just what the sponsor needs for go-live.
Low attrition as an operational variable, not a badge
In AI, continuity matters even more than in other technologies, because part of the project’s knowledge lives outside the documents: in the discarded decisions, in the identified data biases, in the conversations with business and compliance. A partner with high attrition loses that knowledge every time a profile leaves. A partner with low attrition accumulates it.
For a utility planning a three- to five-year AI roadmap, this difference is not a supplier HR detail: it is an operational variable that directly affects total cost of ownership and programme risk.
Where Principal33 fits
At Principal33, AI projects in German utilities are led and delivered by Senior Consultants from start to finish. There is no handover from a commercial team to a junior team after signature. The same profile involved in discovery is present in the architecture, in the regulatory validation and in the industrialisation. This model is sustained by a very low attrition rate, supported by the internal Senior + Junior development policy and by the exclusive focus on the DACH market.
Teams are senior, German-speaking and nearshore DACH, which reduces the risk of regulatory misunderstandings and accelerates decision-making with IT, compliance and business inside the utility.
The shift in perspective
For a German utility evaluating partners for its AI strategy, the useful question is not “how many data scientists do you have?”. Almost everyone will claim many. The question that anticipates the success or failure of the programme is less comfortable: “who is going to decide where AI belongs and where it does not, and has been making that kind of decision in German utilities for years?”. The answer to that question is what separates a shiny pilot from an AI programme that reaches production and holds.
