From Business Goal to AI-Driven Project: The Method for Building Digital Strategies with Real KPIs

I first read it in the McKinsey Global Survey of November 2025, 1,993 respondents across 105 countries, and I thought: this is the number that makes all the others pointless. Not because it’s surprising. Because it explains why almost nobody can answer the simplest question in the world: “and you — how much are you actually making from AI?”

While this is happening, search is reorganising itself around a new architecture. ChatGPT has 900 million weekly active users. Gemini has surpassed 750 million. 70.6% of the traffic coming from these generative engines is invisible to Google Analytics, classified as “Direct” and therefore ignored in most strategic decisions (Loamly, “State of AI Traffic 2026”, 446,405 visits analysed). And yet that traffic converts four times better than traditional organic traffic.

What follows is the method I use with Fortop clients to flip this order: first the objective, then the KPIs adequate to the AI era, then — and only then — the technology. It isn’t the only possible approach. It’s the one that produces measurable results.

The Paradox: Everyone Is Investing, Almost Nobody Is Measuring

The McKinsey Global Survey of November 2025, 1,993 respondents across 105 countries, captures an adoption that on paper looks mature. 88% of organisations regularly use AI in at least one business function. 72% use GenAI — more than double the 33% recorded in 2024. Three-quarters of the C-suite rank it among the top three strategic priorities.

BCG measured it across 1,250 companies: 60% extract no material value at all from AI. 35% scale, but with returns that struggle to justify the investment. The 5% that really works? They spend five times the average, have triple the engagement from leadership, and aren’t optimising processes — they’re rewriting the operating model.

The fastest way to understand why nearly everything fails is one figure, again from BCG: 60% of companies start implementing AI without having defined how they will measure the economic return. They pick the tool, kick off the project, allocate budget and people to it, and only afterwards — if ever — do they ask what it’s producing. This inversion explains the rest: abandonment rates, the zero impact on P&L, the boards that approve the next budget without knowing what the previous one delivered. It isn’t a technology crisis. It’s a methodology crisis.

Sources: McKinsey Global Survey Nov. 2025 (n=1,993); BCG AI Radar 2025 (n=1,250+); Gartner CMO Spend Survey 2025

Projects with pre-approved metrics have a 54% success rate compared to 12% for those without (BCG/Pertama Partners, 2025-2026). A 4.5x difference that no tool, no vendor and no additional budget can bridge after the fact.

The winner profile is precise and replicable. McKinsey’s “high performer” companies spend over 20% of their digital budgets on AI — five times the average — have triple the engagement from senior leadership, and pursue the transformation of the operating model, not just efficiency. BCG quantifies it: “future-built” companies achieve 1.7x the revenue growth, 1.6x the EBIT margins, and 3.6x the shareholder return compared to laggards.

The Real Problem Isn’t the New Ecosystem. It’s That Your KPIs Don’t See It.

If you want to understand how ChatGPT, Gemini and Perplexity are rewriting the rules of visibility, we have published a dedicated analysis. Here we start from the next step: in that already-changed scenario, almost no company is measuring the right things.

The concrete problem is this. 70.6% of the traffic coming from generative engines enters Google Analytics 4 reports classified as “Direct” — no referrer, no campaign, no channel. It’s technically invisible. And yet that traffic converts at a 10.21% rate, against 2.46% for non-AI traffic. The highest-performing channel in your digital mix is the one you know nothing about.

This isn’t a technical problem you can solve with a UTM tag. It’s a structural problem: the metrics you use to justify your marketing budget were designed for an ecosystem that no longer exists. And until you change the metrics, you can’t change the decisions.

GEO Isn’t a Technical Discipline: It’s a Business Choice

There’s a wrong way to frame GEO, and I see it often in client presentations: treating it as the “new SEO”, a technical update to be delegated to the content team. It isn’t. GEO — Generative Engine Optimization — is the strategic response to a shift in market structure: purchase decision-makers ask language models before they ask Google, and what the model answers depends on what the brand has built over time, not on what it optimised yesterday.

The point that most shifts business priorities is this: 86% of AI citations come from sources brands already control (Yext, 6.8 million citations analysed, July-August 2025). Websites (44%), listings (42%), reviews and social (8%). It isn’t a problem of discovering new channels. It’s a problem of optimising what already exists so an LLM interprets it correctly when someone asks “who’s the best supplier of X?”

The numbers on the time window are what convince boards more than the visibility numbers. The GEO market was valued at $886 million in 2024 and is forecast to reach $7.3 billion by 2031, a CAGR of 34%.

It’s worth pausing on a data point that changes operational priorities. The Ahrefs study of December 2025 on 75,000 brands documents that web brand mentions correlate with AI visibility at 0.664, while traditional backlinks stop at 0.218. Three times the correlation.

The Attribution Black Hole: 70.6% of AI Traffic Is Invisible

This is the problem nobody wants to confront head-on, because it touches the foundations of how budget decisions are made. The Loamly “State of AI Traffic 2026” report, 446,405 visits analysed, precisely quantifies the scale of the problem.

The paradox is this: the most invisible channel is also the highest-performing one. Microsoft Clarity documents that AI-referred visitors subscribe 2.4 times and register 10 times more than visitors from traditional search.