Insights
Iffiltra l-artikli
Regulation 18 ta’ Set, 2026
On 2 August 2026 the AI Act became almost fully applicable. But the Omnibus that entered into force in July moved the high-risk obligations to December 2027. Here is what applies today and what doesn't.
AI Strategy 11 ta’ Set, 2026
88% of organisations have adopted AI. 80% see individual productivity gains. Only 37% detect EBIT impact, and barely 6% are genuine high performers. The gap has an explanation.
AI Strategy 1 ta’ Apr, 2026
The gap between AI ambition and AI results isn't a technology problem. It's a strategy problem. After dozens of engagements, here's what we've learned from the projects that actually delivered — and what sank the rest.
Technology 15 ta’ Set, 2026
Capability has climbed steeply in twelve months, but unevenly. Stanford's AI Index carries an uncomfortable finding: the same model that solves competition mathematics fails to read an analogue clock.
Regulation 12 ta’ Set, 2026
If you have a customer-facing chatbot or publish generated content, Article 50 of the AI Act already applies to you. And if the system was deployed before August 2026, your deadline is 2 December.
Technology 8 ta’ Set, 2026
What cost around $30 per million tokens in 2023 costs under a dollar today. If you shelved a use case on unit economics, the numbers you used no longer exist.
Culture & Change 5 ta’ Set, 2026
There is no mass employment displacement. What there is, and it is measured, is an effect concentrated on 22- to 25-year-olds: 19% below where their roles would be without AI in the picture.
Technology 2 ta’ Set, 2026
62% of organisations are already experimenting with agents, but only 23% have scaled them in any function. The distance between those two numbers is the whole article.
Technology 28 ta’ Aww, 2026
In March 2026 the best open model landed 49 Elo points behind the best closed one. A 3.4% gap. And yet open-source share in enterprise deployments fell from 19% to 11%. That paradox has an explanation.
AI Strategy 21 ta’ Aww, 2026
20% of European companies with 10 or more employees now use AI. In Spain, large firms sit at 49.2% and SMEs at 8.7%. Almost a sixfold difference inside the same country.
Regulation 14 ta’ Aww, 2026
Fines of up to €15 million or 3% of global turnover frighten a lot of people the rule doesn't reach. Telling provider from deployer apart saves you an expensive audit.
Technology 7 ta’ Aww, 2026
OWASP keeps it at number one in its 2026 Top 10, and the reason is uncomfortable: it is not a bug, it is a property of the architecture. Instructions and data share one channel.
Culture & Change 1 ta’ Aww, 2026
Four in five university students use generative AI. Those people are already in your company. Banning it doesn't remove it, it only removes your visibility of what is happening.
Data & Analytics 31 ta’ Lul, 2026
Public benchmarks tell you whether a model is good in general. They tell you nothing about whether it works for your case. The distance between those two sentences costs most projects.
Data & Analytics 24 ta’ Lul, 2026
With enormous context windows, plenty of people declared retrieval dead. The evidence says otherwise: there is no silver bullet, and the choice turns on four measurable factors.
Regulation 17 ta’ Lul, 2026
Neither is mandatory anywhere. And yet both are showing up in vendor due-diligence questionnaires, which is a far more effective way of becoming mandatory.
AI Strategy 10 ta’ Lul, 2026
The MIT study that found 95% of pilots move nothing on the P&L points at something specific: the failure is almost never the model. It is the stretch between the demo and the process.
Data & Analytics 3 ta’ Lul, 2026
MIT identified data readiness as one of the three reasons 95% of pilots deliver nothing. It is not a problem a better model fixes, because it is not a model problem.
Technology 26 ta’ Ġun, 2026
On-device inference costs exactly zero per query after the model downloads. For 80% of a typical company's traffic, the quality difference no longer justifies paying per call.
AI Strategy 19 ta’ Ġun, 2026
With six frontier models within 25 Elo points of each other, 'build' no longer means training anything. The real decision is which layer gives you an edge and which is plumbing.
Culture & Change 12 ta’ Ġun, 2026
Enforceable since February 2025, applicable to practically any company using AI, and in the reviews we run almost none can evidence it. It costs little and cuts real risk.
Operations 5 ta’ Ġun, 2026
Most companies pick the most visible process. It is almost always the worst candidate. This is the filter we use to order the list before writing a line of code.
Operations 29 ta’ Mej, 2026
Optimise for percentage of tickets resolved without a human and you will build a system people hate. The right indicator is a different one, and it changes the whole design.
Sales & Revenue 22 ta’ Mej, 2026
80% of people report productivity gains from AI, but only 37% of companies see EBIT impact. In sales, much of that gap comes from automating exactly the wrong things.
Operations 10 ta’ Mar, 2026
Most companies underestimate how much their manual workflows cost them — not just in time, but in errors, delays and missed opportunities that compound every quarter.
Sales & Revenue 18 ta’ Fra, 2026
Your CRM holds data. AI finds meaning. Predictive lead scoring and conversational AI are changing the pipeline for B2B teams with lean budgets.
Data & Analytics 22 ta’ Jan, 2026
You don't need a data warehouse and a team of engineers to get real-time business intelligence. Here's the modern stack we recommend for growing companies.
Culture & Change 8 ta’ Jan, 2026
Technology is rarely the bottleneck. Resistance, misaligned incentives and poor communication are. A guide to managing the human side of transformation.
AI Strategy 12 ta’ Diċ, 2025
Finance leaders are skeptical for good reason — AI ROI is hard to quantify. We share the framework we use to build business cases that get budget signed off.