AI has been on every board agenda for a few years now. That's long enough to stop guessing and look at what's working and what isn't.
A new discipline is taking shape in front of us, so it's normal that nobody has settled yet on the best ways of working, the frameworks or the technology. Every few weeks a new narrative takes over: RAG, knowledge graphs, semantic layers, agent loops, MCP. That churn is what a disruptive technology looks like in its early years. The hard part is separating the signal from the noise, and that's what this newsletter is for.
Where the name comes from
In September 2025, which in AI news already feels prehistoric, BCG published a study of more than 1,250 companies. Only 5% were getting value from AI at scale. BCG called them "future-built". At the other end, 60% reported hardly any material value, despite substantial investment.
That 5% is where "The 5%", the name of this newsletter, comes from. The exact number will keep moving. The name stands for the question behind it: what do the few who get it right do differently?
A year later
I went through the latest reports to see whether that number has moved. It has, but not that much:
- BCG, September 2026 (1,330 executives): future-built companies went from 5% to 7.5%. Over the same period, AI spending more than tripled, from about 1% of revenue to 3.3%.
- McKinsey, August 2026 (1,719 respondents): AI "high performers", companies attributing at least 5% of EBIT to AI, are still about 6%, unchanged from 2025. Only 37% can attribute any EBIT impact to AI at all, also flat.
- Teradata and Wakefield Research, June 2026 (1,000 technology and data leaders, VP and above): only 7% have reached the most mature stage of agentic AI. 63% have yet to see meaningful returns.
Three studies, three different definitions, and they land in the same place: somewhere between 6% and 7.5%. We've been told AI is going to break the economy at lightspeed, but it seems there's some friction on the way.
To be fair, the middle is moving. According to BCG, almost half of companies now generate measurable value from AI, and the other share stuck with nothing shrank from 14% to 4.5% in a year. More companies are getting something. Very few are getting a lot.
Some sectors are further ahead than others. In technology, 70% of companies already get measurable value from AI. In chemicals, 38%. But BCG found at least one future-built company in every one of the 20+ sectors it studied, so your sector doesn't decide which group you end up in. In BCG's words, leadership choices do.
How many companies are getting AI to work
Share of companies already getting measurable value from AI.
- 1R&D
- 2Device and software development
- 3Customer service
Show the data table
| Sector | Getting measurable value | Agentic AI in production | AI spending | Revenue increase from AI | Cost reduction from AI |
|---|---|---|---|---|---|
| Technology | 70% | 76% | 4.1% | 2.0% | 2.1% |
| Banking | 60% | 61% | 5.7% | 1.2% | 1.6% |
| Telecom | 56% | 62% | 3.7% | 0.6% | 1.1% |
| Biopharma | 49% | 50% | 3.3% | 0.8% | 1.0% |
| Aerospace | 52% | 52% | 3.6% | 1.2% | 1.4% |
| Insurance | 53% | 48% | 3.3% | 0.6% | 1.2% |
| Retail | 48% | 49% | 2.7% | 1.2% | 1.5% |
| Power & utilities | 46% | 36% | 3.1% | 1.0% | 1.0% |
| Machinery | 46% | 55% | 3.2% | 0.9% | 0.9% |
| Chemicals | 38% | 36% | 2.6% | 0.7% | 0.9% |
So what are they doing right?
Reading the different reports side by side, five patterns show up in more than one of them.
1. They pick a few big bets and own them from the top. Future-built companies run AI as one multiyear program instead of a collection of initiatives: 61% commit to a multiyear effort, against 17% of laggards, and all of them have clear ownership of the AI agenda. They also finish what they start. On average they have 4 AI workflows stuck in planning and 20 running at scale. Laggards have 10 stuck and 4 at scale.
Example At a European bank, every business unit was running its own AI projects, and each team was rebuilding the same pieces. It cut back to about ten key initiatives, built on one shared platform with common standards and guardrails. Execution got faster, and the bank projects productivity gains of up to 10% of its cost base within three years.
2. They redesign the work instead of adding AI on top of it. Nearly three quarters of McKinsey's high performers have fundamentally redesigned workflows because of AI. Among everyone else, one in four. BCG sees the same thing from another angle: future-built companies expect about 70% of their AI value to come from reshaping processes, and 62% of their initiatives reshape or invent a process rather than deploy a tool (44% for laggards).
Example At a US regional health insurer, customer service reps searched more than 20 documents per call, and calls averaged 16 minutes. The insurer rebuilt the service process around three agentic AI solutions on a new platform that links documents, structured data and live systems. Agents now handle about 60% of call types, handling time is down 35%, and the insurer projects over $60 million in call-center savings.
3. They measure the value in the P&L. 95% of future-built companies track AI with clear KPIs or directly in the P&L. Among laggards, 47% measure it directionally at best (BCG). McKinsey's high performers are twice as likely to have a defined process to quantify AI impact. If you can't tell which initiative is paying off, you can't stop the ones that aren't.
Example A global investment bank runs AI as a single funded program with fewer than ten high-value use cases. A program office reporting to senior leadership tracks each one straight to the P&L. Its targets: $500 million in financial upside by 2027 and $1 billion by 2030.
4. They fix what their data means. BCG reports that basic data structuring and access controls are now common at every maturity level, between 58% and 72% of companies. The sharper divide is whether you can rely on the semantics: data that is current, consistently defined, trustworthy and has a clear owner. McKinsey's high performers are more than twice as likely to have a semantic layer or knowledge graph over their data. Most companies know they're behind here. 77% of leaders say 20% or less of their enterprise data is described and contextualized well enough for agents to act on (Teradata). Among data practitioners, 84% run into conflicting versions of the same metric, and 80% put a semantic layer with standardized definitions in their top three AI enablers, ahead of every other option (Modern Data Report).
Example A global resources company received over 10,000 master-data change requests a year. The process was manual, the backlog reached six months, and people stopped reporting errors, so data quality kept degrading. An AI agent now turns a plain-language description into a checked, ready-to-submit request: 3 steps instead of 68. Employees fix data errors when they spot them, at the source.
5. They treat governance as an accelerator. BCG looked at six basic rules for running AI agents: what context and memory an agent can use, what it's allowed to do and when a human has to step in, how it's tested and rolled back, how it connects to tools, who owns it, and how its actions are secured and audited. Companies that apply all six across the whole company get three times as much value from agents as companies that apply only one. Just 5% have all six. The rule that weighs most is the first one, what data and context the agent works with.
Example A global outsourcing company decided, process by process, how much freedom its AI agents get. There are three levels: the AI only analyzes the work, the AI assists a person, or the agent runs the task on its own. Each process spells out where decisions are made and when a person has to step in. Rolled out with a central team of about 50 people and 100 local teams, the pilots improved conversion rates by up to 40%.
None of this is new technology. A program with an owner, workflows redesigned around AI, clear KPIs, data people can trust, guardrails for agents.
BCG has a rule of thumb for this kind of transformation, the 10-20-70 rule: put 10% of the effort into algorithms, 20% into technology and data, and the remaining 70% into people, organization and processes. Most of what the top performers do sits in that 70%. AI may be a disruptive technology, but in the end what sets them apart is how they organize around it.
My commitment with this newsletter: something worth reading slowly and information you can actually use. If that's what you're after you can subscribe at revale.tech/newsletter.
Hernan
Sources
- BCG, The Widening AI Value Gap (Build for the Future 2025), September 2025, n = 1,250. bcg.com
- BCG, The Formula for Agentic AI Value (Applied AI Index 2026), September 2026, n = 1,330. bcg.com
- McKinsey, The State of AI in 2026: On the Road to ROI, August 2026, n = 1,719. mckinsey.com
- Teradata and Wakefield Research, Arrested Automation: Why Agentic AI Stalls at the Enterprise Level, June 2026, n = 1,000. teradata.com
- The Modern Data Company, The Modern Data Report 2026: The Data Activation Gap, February 2026, n = 540+ data practitioners.