Raft unveils "AI KOMP-AS" framework for profitable, safe AI adoption

Raft has unveiled AI KOMP-AS, a framework that treats AI adoption not as a set of fashionable experiments but as a manageable business transformation with measurable value, clear constraints and a route from idea to scale planned in advance. The approach rests on the team's 14 years of practical experience developing and deploying AI/ML solutions for clients in Russia and abroad.

The main thesis is simple: access to models is no longer enough for a project to deliver results. For executives and business owners, AI literacy is becoming an applied skill on which a company's economics, speed of change and ability to avoid spending budget on initiatives with no payoff all depend. Adopted deliberately, the technology can drive growth. Otherwise, it quickly turns into an expensive, poorly controlled line item.

Raft illustrates this gap between expectations and actual returns with data from the MIT report The GenAI Divide: State of AI in Business 2025: over a one-year horizon, only one organization in twenty achieved a positive economic effect from AI adoption.

The causes are familiar to most companies: it is unclear where exactly AI will bring the most benefit, how to match the technology's capabilities to real processes, what risks lie hidden in data, infrastructure and the operating model, and at what point a pilot makes sense to turn into a product. Against this background, the framework looks less like an abstract methodology and more like a way to reduce the likelihood of costly mistakes at the start.

AI KOMP-AS breaks AI transformation into six sequential steps.

• The first, "K": where the company wants to get to, meaning which business goals, growth points and constraints must be taken into account.
• The second, "O": where the route starts, including the current state of processes, technology and pain points.
• The third, "M": whether the organization can make this journey at all, given the maturity of its processes, its culture, data quality, available infrastructure and regulatory risks.
• The fourth, "P": how to prioritize initiatives by value and economic viability, so as not to automate everything indiscriminately.
• The fifth, "A": architecture and product design, where hypotheses are tested, requirements are formulated, and risks, cost of ownership and a roadmap for taking the solution from prototype to a minimum necessary production are assessed.
• The sixth, "S": scaling, where the team prepares a strategy for moving from MVP to full AI automation of a function without losing manageability and quality.

An important part of the approach is rejecting the logic of "first the model, then we'll figure out why we need it." AI KOMP-AS reverses the order: business context and success criteria first, then an assessment of readiness and economics, and only then solution design and deployment. For companies, this means tougher filtering of ideas at the entry point, a clearer choice of use cases, and less chance of falling into the trap of demo pilots that look impressive but do not survive contact with real processes, security, integrations and maintenance costs.

Raft separately stresses that a roadmap must account not only for functionality but also for non-functional requirements, constraints and the scaling scenario. Otherwise a local success does not turn into a systemic result.

What this means in practice: the market is gradually moving from talk of "adopting AI at any cost" to talk of adoption discipline. AI KOMP-AS is interesting precisely as an attempt to package this shift into a clear management scheme: from goal and diagnostics to priorities, architecture and scale. For companies that are tired of chaotic experiments and are looking for a way to tie AI initiatives to ROI, TCO and real business constraints, such a framework can be a useful entry point. It is not a guarantee of success, but a way to noticeably improve the odds that AI investment delivers not just a demonstration of capabilities but a measurable economic effect.