June 2026 Archive

Most companies say they review strategy quarterly. In practice, the quarterly review is a status report dressed up as strategy.

Real strategy review asks: are the assumptions we made three months ago still true? If they are, double down. If they are not, change the strategy now, not next year.

The cadence depends on the speed of the market you operate in. In stable industries, biannual is enough. In AI, energy, or anything export-driven in 2026, monthly is closer to right. Some early-stage segments need a fortnightly check.

What kills strategy is not absence of review. It is review without honesty about what changed.

A useful litmus test: in the last three months, has any leadership decision overturned a previous one based on new information? If not, you are not running a strategy review. You are running a confirmation ritual.

In Norway, a meeting from 09:00 to 10:00 ends at 10:00. Anything else feels disrespectful.

In Brazil, the same meeting ends "when the topic is exhausted". Pushing for a hard stop reads as cold or transactional.

Both are correct, in their own context. Anthropologists call this the difference between monochronic and polychronic time. One culture treats time as a line you follow strictly. Another treats it as a pool you swim in.

If you sell, lead, or partner internationally, you are negotiating two things in every meeting: the substance, and the time-rules of the room.

A few practical signs you are in a polychronic context: meetings drift past their end time without anyone flinching; agendas are treated as suggestions; multiple conversations run in parallel.

The mistake I see most often is Norwegian or German leaders reading polychronic flexibility as disorganisation, and Brazilian or Mediterranean leaders reading monochronic rigidity as disrespect.

Neither is true. Both sides are operating by the time-rules they were taught.

Calibrate before you sell.

Founders and executives face this question constantly: is this not working because the strategy is wrong, or because we haven't given it enough time?

A useful heuristic: if your assumptions about the market were wrong, pivot. If your execution of a correct strategy is slow, persist.

The trap is pivoting away from a good strategy because execution is hard, and calling it a "strategic decision."

Before you pivot, audit: have you actually tested the strategy, or have you tested a weak version of it?

The hardest sales discipline is not closing. It is qualifying out.

In complex B2B deals, every account that does not close still costs you something. Pre-sales engineering hours. Senior leadership attention. Pricing concessions you offered to keep it warm. Time your team spent that they could have spent on a winnable account.

A deal you should walk away from will usually show three signals: the budget cycle does not align with your sales cycle; your champion has limited internal capital; and the buying group keeps adding stakeholders without removing any.

Stay too long and you teach your sales team that effort is the metric. Walk away too early and you teach them to chase the easy ones. Both are wrong. The discipline is to walk away on data, not on mood.

A practical question I ask: if this deal closes, does it become a reference customer we are proud of? If the answer is "no, but the revenue helps", the deal is probably costing you more than the contract is worth.

Build the discipline to qualify out. The pipeline you keep will be stronger.

Most SMEs still run a single annual planning cycle. Twelve months of budget. Twelve months of strategy. Reviewed on the calendar, not on the data.

In 2026, this is no longer good practice. Markets, costs, regulation, and supply chains move on cycles much shorter than twelve months. AI capability shifts every quarter. Energy prices shift every month. EU regulation lands every six weeks.

The replacement is not "no plan". It is a layered plan. A long-horizon strategy that changes rarely, two to three years out, with explicit assumptions written down. A medium-horizon plan that updates quarterly, six to nine months out, with measurable milestones. A short-horizon execution plan that updates monthly, six to eight weeks out, owned at team level.

The leaders I see thriving in 2026 are not the ones with the most ambitious annual plan. They are the ones who can change a milestone within four weeks of new information arriving.

Plans are still essential. Annual cycles, on their own, are not.

What a €2-7K budget gets you for local AI in 2026. Which chips actually matter for inference. The real difference between running models locally vs. using an API. Why some tasks belong on-device and others don't.

The prosumer AI hardware stack is moving fast, and in 2026 it is moving on one thing: memory. A global DRAM shortage has repriced the whole shelf, so here is what I'd buy today, and why.

At this budget the choice is raw speed on a discrete GPU, or memory capacity on a unified-memory box. The shortage has pushed those two further apart.

The speed route. An NVIDIA RTX 5090 with 32GB of very fast memory, now around €3,000 to €3,500 (the €2,000 launch price did not survive the shortage). It is the quickest tokens you can buy without an enterprise contract, and by some distance the best local image and video generator. You can even run it as an external GPU over Thunderbolt 5 or OCuLink, though at 575W it needs a serious enclosure, so in practice it stays a desktop-tower card. The catch is unchanged: 32GB is the ceiling on a single consumer card, so a 70B model already spills past it.

The capacity route. An AMD Ryzen AI Max+ 395 mini PC, the GMKtec EVO-X2, with 128GB of unified memory for around €1,900 direct. It holds 70B-class models a 32GB card cannot, sits silently on a desk, and sips power next to a 575W discrete card. The honest trade is speed: it is bandwidth-limited, so it flies on mixture-of-experts models and crawls on dense 70B, single digits to low teens of tokens per second. You run bigger models, just slower, on Linux rather than CUDA.

The Apple route. A Mac with 128GB of unified memory is the quiet, efficient option: near-silent, low power, comfortably over 20 tokens per second on a 70B model with Apple's MLX stack. The asterisk is price. The 128GB box here is a Mac Studio with the M3 Ultra chip, 32-core CPU and 80-core GPU, around €6,600, and the DRAM shortage is a big part of why that much memory costs what it does. You pay a real premium for macOS and silence.

None of this top end is needed for everyday work. Most business tasks, drafting, summarising, classifying, and answering questions over your own documents, sit happily on 14B to 32B models, which run on far cheaper hardware.

The honest part. Local wins on privacy, full control of your data, and zero cost per token once the box is paid for. The API wins on frontier reasoning and the occasional heavy lift you cannot match at home. For most teams the answer is hybrid: a small local model for the private, high-volume, repetitive work, and a frontier API for the hard problems.

So buy for the model you run every day, not the one you run once. And in 2026, buy the memory before the shortage decides for you.

The pattern repeats every fifteen years. New technology arrives. Vendors sell it. Companies buy it. Two years later the system sits half-used while the team works around it.

ERP in the 1990s. CRM in the 2000s. Cloud in the 2010s. AI now.

Every time, the failure mode is the same: the technology was treated as a procurement decision, not an operating-model decision. Tools were chosen before workflows. Pilots ran in isolation from the actual business. Rollouts stalled when the org chart did not change to match.

What works for AI in 2026, just as it worked for ERP in 1996: pick one workflow that matters, change the operating model around it first, and only then bring in the tooling. Measure adoption, not licence count.

Most of my conversations with SME leaders this year follow the same arc. They have bought tools. They want help making them load-bearing. The bottleneck is rarely the model. It is the workflow it sits inside.

In over a decade delivering training courses globally, I've noticed something consistent: participants from certain cultures give perfect scores almost automatically. In others, you'd be lucky to see an 8.

Neither group is lying. Both are being culturally honest.

In some cultures, giving low feedback, even anonymously, is a form of disrespect to the trainer, the host, or the institution. The 10/10 is a courtesy.

In others, 10/10 means "this changed my life." Anything less is just... normal.

If you're designing global training programmes, calibrate your feedback instruments culturally. Or you'll spend your time optimising for scores that don't mean what you think they mean.

When we set out, Stormberry had one goal: to be the operational AI partner that sits beside a business, not above it. The kind that turns AI from a buzzword into everyday work, across sales, strategy, culture and delivery.

A hundred of you now follow that idea with us: founders, operators, the curious and the sceptical, from Vestland to Brazil and beyond. That means a great deal.

It is a small number with a large meaning. Every follow says practical, hands-on AI is worth building well, in plain language, with real outcomes, owned by the people who actually run the business.

We are only getting started. The next hundred will be earned the same way as the first, one honest conversation and one shipped result at a time.

To everyone who has read, commented, shared, or simply followed along: takk, obrigado, thank you.

A bad sales hire costs more than their salary. It costs you: accounts mismanaged, relationships damaged, market feedback ignored, and 6-12 months of lost pipeline.

The root cause is almost always a misalignment between what the role actually requires and what the candidate was evaluated on in the interview.

Hiring for charisma when you need process discipline. Hiring for industry knowledge when you need learning agility.

A sales competency framework aligned to your specific sales motion is not HR bureaucracy. It is risk management.