Klarna was the first global brand to publish a retrospective on how its AI rollout went, and the most useful thing about the document is what it does not flatter. Ninety percent of staff are using AI daily, which is the number that gets quoted. The patterns underneath that headline are more interesting, and most of them are inside reach of a winery with thirty people.
The pieces that translate are the quiet ones. A quarterly fifteen-minute interview with each staffer. A shared prompt library with names on the entries. A spinup process for ad-hoc tools so that the moment someone has an idea, the path from idea to working artifact is not a six-week procurement cycle. These are the parts that look the most boring in the writeup and turn out to be load-bearing.
The pieces that do not translate are the ones a winery owner should skip on purpose. A dedicated AI strategy team is a scale-dependent move. A 30-person company that copies it ends up with one person who has been pulled away from real work to write an AI strategy document nobody reads.
What the retrospective says
The Klarna retrospective came out after a few years of public AI experimentation that had its own embarrassing moments — the company replaced customer service roles, then walked some of that back when service quality dropped. The retrospective does not bury that. It frames the rollout as a culture project that worked because of the human-side work, with the tooling sitting underneath as a secondary layer.
The numbers Klarna published are useful because they are unusual in their specificity. Ninety percent of staff are using AI daily. Internal expertise development on AI workflows runs 3.2 times faster on teams that share a prompt library compared to teams where prompts stay on individual machines. Both numbers are reported from inside, by a company with skin in the game. Neither is a vendor benchmark.
The patterns Klarna highlights as the operational backbone are where the lessons sit for a smaller team. Four are inside a winery's reach. One is not.
Pattern 1: The quarterly deep interview
The piece of the Klarna play that has the cleanest transfer to a thirty-person winery is the quarterly deep interview. Every employee gets a fifteen-minute one-on-one each quarter, and the agenda is the same every time. Half the conversation is an engagement check on how the staffer is using AI day-to-day, what is working, what is friction. The other half is roadmap input — what would she build if she had the time, what task does she wish the tool would help with that it currently does not.
Fifteen minutes per staffer, quarterly. For a thirty-person winery that is 7.5 hours a quarter, split across the GM and one or two function leads. It fits inside a normal management rhythm without needing a project plan.
What it produces is two assets that are hard to get any other way. The first is a current map of where adoption sits, told by the people doing the work, not the dashboard. The second is a roadmap of small tooling ideas with named champions attached. The DTC coordinator who has been thinking about how to handle wine club hold requests faster gets to say so to the GM, and the GM walks out of the meeting with one specific thing to build.
The version a winery would run does not need a survey tool, a structured form, or a manager-coaching framework. A notebook open on the table works. The reason it works is the cadence and the consistency, not the instrument.
The interview is also the structural answer to the involvement gap I wrote about in the four numbers every winery leader should know about AI. HBR's research has 78% of employees fearing job loss to AI while only 12% feel involved in deployment decisions. The Klarna interview pattern is a low-cost repeatable move that lifts the involvement number. A staffer who knows she is going to be asked every ninety days what she would build is a different staffer than one who is waiting to be told.
Pattern 2: The shared prompt library
The second piece is the shared prompt library. The 3.2x faster expertise development number is the headline, but the mechanism underneath is simple. When a staffer figures out a prompt that produces a good output for a recurring task, she puts it in a shared place. When the next staffer hits a similar task, she starts from that prompt instead of from a blank window.
The thirty-person winery version of this is a folder in whatever shared-doc tool the winery already uses — Google Drive, Notion, SharePoint, a shared Obsidian vault if the winery is doing that — with one document per prompt and a flat naming convention. No custom tool required. "Wholesaler late-payment nudge." "Cancellation reply, mid-tier club." "Pour-list draft for restaurant placement." Each document holds the prompt, two or three sample outputs, and a one-line note on when to use it.
The library gets a champion at the start. The Klarna number — 3.2x — sits inside an organization with a small group of people who maintain the library and review additions. A thirty-person winery does not need a librarian role. It needs the tasting room manager who is already the first AI champion to spend ten minutes a week keeping it tidy. The depth of the library is the depth of adoption, more reliably than any usage dashboard. If the library has thirty real prompts in it after a quarter, the rollout is working. If it has six and the rest are placeholders, the rollout has stalled.
Pattern 3: Names on the library entries
The third piece is the one most easily missed. Klarna's prompt library carries author attribution. The DTC coordinator who wrote the cancellation reply prompt gets her name on the document. The wholesaler ops person who wrote the late-payment nudge gets hers on that one. When a teammate uses the prompt, the author is visible.
The recognition mechanism is permanent and low-key. There is no Slack shoutout, no badge. The author's name sits at the top of a working document that other staffers open every week. The reward is being usefully visible to the team for something concrete.
This is the recognition pattern that fits a wine industry team better than most of the alternatives. A bottle from the cellar is a fine occasional reward. A handwritten note from the GM is a fine occasional move. Neither one compounds. The named entry in a library that gets reopened week after week compounds, and it works for staff with quiet temperaments who would actively dislike a public Slack shoutout. The Carnegie Mellon work on workplace gamification has a long thread on how individual leaderboards erode moral agency. The named-attribution-in-the-library pattern works as recognition without the competition layer a leaderboard adds.
A winery owner who wants to copy one pattern from Klarna this quarter should make this the one. It costs nothing, it transfers cleanly, and it solves the recognition problem most rollouts get wrong.
Pattern 4: The internal sandbox
The fourth piece is the spinup infrastructure. Klarna describes what they call an internal sandbox, where staff with ideas can stand up a working prototype quickly, without going through procurement or a formal IT project. The design goal is to shorten the path from "I wonder if I could automate that" to "I will try it on Friday afternoon" so the idea survives the trip.
The full Klarna version of this involves engineering support a winery does not have. The transferable piece is the principle: keep the path from idea to working artifact under a week, ideally under a day. For a thirty-person winery, that usually looks like one of three things.
First, a small AI subscription budget the GM can authorize without asking anyone, capped at something like $200 a month for one-off experiments. The tasting room manager who wants to try Claude on a tasting note draft does not need to file a request. She uses the company card.
Second, a champion who has admin access to whatever the team's primary AI tool is, and who can spin up a project, a Custom GPT, or a shared workspace in twenty minutes when someone asks. The latency from request to working artifact stays under a day.
Third, a Friday afternoon window — even thirty minutes — where the champion is reachable in person or by phone for setup help. The Friday show-and-tell ritual is a separate piece of rhythm. This is the ad-hoc support window that catches ideas before they cool.
None of these require the engineering team Klarna has. They require a leader who has decided in advance that idea-to-artifact friction is a thing to remove, and who has authorized one or two small mechanisms to remove it.
Pattern 5: Peer modeling, not leader exhortation
The fifth piece is harder to describe and easier to recognize once a winery sees it working. Klarna describes the culture work around their AI rollout as a peer-modeling project. Senior leaders narrating their own AI use was part of it. The bigger lever was getting the first wave of users — the staff their peers already trusted — to be visibly using AI in front of those peers.
The Gartner number from 2025 has 37% of non-users citing "my coworkers don't use it" as their reason. That is the lever Klarna pulled. The retrospective does not talk about it in those terms, but the design choices are consistent: champions per function, and named prompt authors in a shared library where every team member can see who wrote what.
For a winery, the peer-modeling work is what happens between the first champion and the second employee. The leader's job is to set the conditions so that the first champion is visible to her peers — using the tool out loud at the Wednesday huddle, contributing entries to the prompt library, helping a coworker debug a stuck output. The second employee joins the play because she watched her trusted peer for four weeks, well before any announcement.
What does not transfer
Klarna also did things a winery should not copy. The most expensive one is the dedicated AI strategy team. At Klarna's scale that team has a job to do: cross-function coordination, vendor management, model evaluation, compliance handling. At thirty people, that role becomes a make-work assignment that pulls someone off real work to produce a strategy document that nobody reads.
The same goes for the formal AI governance committee, the cross-functional AI roadmap document, and the company-wide AI literacy training day. Each of those moves makes sense at five thousand employees and produces theater at thirty. A winery owner who reads the Klarna retrospective and pulls "we need an AI strategy team" off the page has imported the part that does not fit.
The right scale move is more boring. The GM owns the rollout part-time. A function lead is the champion. The shared prompt library is in a folder. The deep interview is fifteen minutes on the calendar. None of these need a strategy team to exist.
I think about this as the cost of importing the wrong pattern. A 30-person business that adopts a 5,000-person process gets the overhead without the benefits, and the rollout stalls under the weight of its own coordination cost. The pieces Klarna ran that produce the 90% daily-use number are the small repeatable ones. The big-org wrappers around them are noise at small scale.
What I would steal this quarter
If I were sitting with the GM of a thirty-person winery this week, the move I would push hardest is the quarterly deep interview. Block fifteen minutes per staffer in the next two weeks, ask two questions — how is the AI work going, and what would you build if you had time — and take notes. Run it again in ninety days.
The second move is the named-attribution prompt library. Stand up the folder this quarter. Seed it with five entries from the existing champion. Put her name on each one. The library grows from there, and the staff who contribute see their names compound.
Both of those moves are inside the budget and the staffing of a winery already doing the work. Neither needs the AI strategy team. The Klarna retrospective is most useful as a confirmation that the small operational rhythms are the load-bearing parts, not the company-wide programs that get the press releases.
A note on the webinar
The hour I am running is built for winery owners and GMs who are looking at Klarna-style numbers and trying to figure out which patterns fit a small team and which do not. The first half is a live demo of building a starter folder for one frontline winery role. The second half walks through the 60-day play and the operational rhythms — the deep interview, the shared library, the recognition patterns — that hold a rollout together past month two. Check the webinar schedule for the next date and to register.
Part of a 20-post series on employee AI adoption for wineries — see the full series under AI Adoption.