The path to effective AI: the method behind the Rautaki consulting programme

Effective AI adoption is not a question of the right tool, but of a structured path from an honest baseline assessment right through to secure production. The path to effective AI («Der Weg zu wirksamer KI») is the method Rautaki developed for exactly that: three phases, nine steps and two go/no-go gates, at which an organisation decides — with full cost control — whether to proceed. It deliberately leads from the question "Where do we stand?" to the question "How does AI become a repeatable capability?" — and not from one tool to the next disappointment.

The key points at a glance

• A path, not a tool. The method breaks AI adoption into three phases — baseline assessment, validation, scaling — with nine concrete steps.
• Two gates with cost control. At two decision points, you deliberately proceed or stop — with no long commitment upfront.
• Five design principles carry it. Evidence before intuition, focus over breadth, gates over a blank cheque, enablement over dependency, compliance from the outset.
• For organisations with responsibility. Developed for nonprofits, the social and public sectors and SMEs — wherever donor and grant money demands cost discipline.

The problem the method solves

The finding is sobering: according to the MIT GenAI Enterprise Report 2025, around 95% of AI pilot projects fail before they ever reach production. Not because of the technology — that mostly works — but because of the way organisations approach AI.

Three failure patterns recur. The first is frantic activity: a dozen parallel pilot projects with no focus, driven by enthusiasm rather than prioritisation. The second is tool-first thinking: a tool is bought and set down alongside the work without rethinking the workflow — the tool changes nothing because no one changes how they work. The third is a lack of anchoring: a pilot succeeds, but no one is accountable, the knowledge stays with external consultants, and when they leave the capability disappears with them. The method is built as an answer to exactly these three patterns.

The five design principles

Evidence before intuition. Before anything is changed, we capture the current state of the core processes — time, cost, quality — and set a benchmark. In the end, the benefit of a pilot is measured against precisely this baseline. That produces proof of impact in black and white, rather than a gut feeling that "it does some good". Anyone who does not measure where they start can never demonstrate where they have got to.

Focus over breadth. The biggest mistake is to begin everywhere at once. The method forces reduction: the AI vision is sharpened down to three or four priorities, and from many possible use cases the few are chosen by value and feasibility. A handful of the right initiatives with real impact potential beat a dozen gimmicks — in practice that means deliberately saying no.

Gates over a blank cheque. The path has two built-in decision points. At the first gate — after the baseline assessment — a joint decision is made before any investment in implementation. At the second gate, scaling happens only if the pilot has genuinely proven its impact. In between, the organisation retains full cost control: it starts small with the fixed-price baseline assessment and decides afresh at each gate, rather than committing for the long term upfront.

Enablement over dependency. The goal is not a mandate that renews itself, but an organisation that uses AI independently. The method trains the team, develops internal champions and defines an operating model with clear responsibilities — up to an accredited certificate on request. The knowledge stays in-house, because good consulting makes itself redundant.

Compliance from the outset. Governance is not an afterthought bolted on at the end, but built into every phase — from the first guardrail for data protection and responsibilities through to the operating model. The EU AI Act and the revised Swiss Data Protection Act (revDSG) are considered as a use case emerges, not once it is already in production. Compliance added after the fact is expensive and rarely watertight.

The structure in brief

The three phases build on one another:

• Phase A · Baseline assessment & foundation — maturity level, baseline and a board-ready strategy with secure guardrails.
• Phase B · Focus & validation — prioritised use cases, the right implementation route and a pilot with proven impact in real day-to-day work.
• Phase C · Embedding & scaling — enablement, a stable operating model and the roll-out to further teams and processes.

Between them lie the two gates: Gate 1 before implementation, Gate 2 before scaling. The full nine steps, with their guiding questions and outcomes, are set out in the detailed approach and in the booklet PDF (PDF in German).

Who the method is for

The method is deliberately developed for organisations with a particular responsibility: nonprofits, the social and public sectors and SMEs. That is no coincidence. Here, work is often funded by donor, grant or public money, frequently within volunteer-run (militia) structures with voluntary supervisory bodies and limited time for technology. Uncontrolled experiments in this environment are not merely inefficient — they are a risk to trust.

This is exactly what the cost-control gates answer: instead of one large budget decision into the unknown, there is a small, fixed-price entry point followed by deliberate yes/no decisions at defined points. A foundation board or an executive management can thereby account for what is spent — and demonstrate at any time what it was spent on.

The next step

The path begins with an honest baseline assessment — and that begins with a conversation. Arrange a free initial consultation: in 45 minutes we clarify your priorities and the most sensible next step, without obligation.

The method supplies the answers — the supervisory body asks the right questions. As the governance counterpart to it, our article on the AI questions for the board of directors shows how a body recognises a robust AI strategy.

Harry Witzthum

Founder of Rautaki · Doctor of Philosophy · NPO manager VMI