Representative engagements across our four practices — what the problem was, what we built, and what it actually changed.
Start a conversationThese are representative engagements — anonymised composites that convey the shape and quality of the work, not claims about a specific named client. We name a client, their logo, or their numbers only with their permission, so you won’t find a borrowed logo wall here. Every outcome below is something we shipped or made possible; we don’t quote ROI percentages we can’t stand behind.
Each of these sits in one of four practices — what we do, and which one you need.
The habit of not inflating numbers came from working at a scale where they couldn’t be fudged — what I learned from one million accounts.
Fintech · payments · ~120 staff · Nordics/EU
Risk analysts were reviewing every flagged transaction from scratch — cross-referencing scattered policy PDFs, sanctions notes, and prior decisions by hand. Backlogs grew faster than headcount, and no one could later prove why a case was approved or declined.
We built an agentic review workflow that drafts a recommendation for each flagged case — gathering the signals, retrieving the governing policy, and proposing approve, decline, or escalate with its reasoning — while a human analyst stays the decision-maker. A separate copilot answers policy questions grounded only in the firm’s own documents. We worked evaluation-first: a graded eval set and guardrails came before the agent, not after.
Analysts moved from reviewing every case cold to confirming or correcting a cited draft — clearing the standing backlog and leaving each decision with a logged, source-linked rationale a reviewer or auditor can reconstruct. Policy answers now arrive with the binding clause attached, or with silence instead of a confident guess.
Veterinary care · 9 clinics · ~70 staff · DK/SE/NO
A growing veterinary group ran nine clinics across three countries on a generic practice tool plus a sprawl of spreadsheets. Patient histories were siloed per site, medicine stock was counted by hand, and each country billed in its own currency — so month-end consolidation took days and never quite reconciled.
We built a bespoke multi-tenant platform where each clinic is its own tenant, but a patient’s record follows them to any branch. Booking, clinical notes, prescriptions, and stock live in one system; invoicing is multi-currency with each country’s VAT and rounding handled correctly; and a group dashboard rolls every site up to a single, reconcilable view.
Month-end group consolidation went from a multi-day spreadsheet reconciliation to a same-day close. A patient arriving at any of the nine clinics now has their full history on screen, rather than a phone call to another site.
HR-tech · ~70 staff · Series A · Nordics + DACH
An HR-tech scale-up whose product screens and ranks candidates had quietly become a high-risk AI deployer under the EU AI Act — and a procuring enterprise’s security questionnaire had just asked them to prove it. They needed to know exactly where they stood, what to fix in-house versus buy in, and who owned what, before the obligations bit.
Over a fixed-scope engagement we mapped their actual AI surface against the Act’s obligations, separated genuine high-risk uses from lower-risk ones, and turned the gaps into a sequenced, owner-assigned plan. Counsel from an operator who still ships — so every recommendation was build-versus-buy honest and costed against their real roadmap.
They walked away with a defensible, evidenced answer for the enterprise procurement questionnaire and an owned remediation backlog already in their sprint board. The biggest single change was operational: AI governance moved from an anxious open question to a named owner with a dated plan.
B2B SaaS · ~25 staff · Nordics/EU
The product was genuinely good and the team knew their domain cold — but the market couldn’t tell. Growth leaned entirely on founder networks and paid ads that stopped the moment spend paused, and the site read as a feature list rather than a point of view.
We started with positioning — naming the specific problem the product solves and the buyer it serves — then turned that narrative into an SEO and content system built as infrastructure, not a campaign. Editorial, technical SEO, internal linking, and conversion paths were treated as one connected machine the team could run without us.
The company shipped a coherent narrative and a content engine it now runs itself, publishing on a steady cadence against a clear topic map — with a growing library of pages ranking for buyer-intent terms, a measurable share of demo requests arriving through organic search rather than paid, and positioning the sales and product teams reuse verbatim.
Charity / non-profit · ~25 staff + volunteers · Nordics
Donations ran through a dated hosted form that dropped roughly one in seven gifts at the payment step. And every quarter, two people lost most of a week hand-assembling the board and funder impact pack from four exported spreadsheets.
We rebuilt the giving flow as a fast, accessible, multi-currency checkout on Stripe, then put a small reporting pipeline behind it that turns the same donation and programme data into a board-ready pack — with a light AI assist that drafts the narrative and always cites the figure it is summarising. The point was leverage, not headcount.
The new checkout recovers gifts the old form was silently dropping at the payment step, and the quarterly board and funder pack now arrives as an approved first draft the same day the quarter closes — turning close to a person-week of spreadsheet assembly into a review-and-sign task.
Not a client’s — so we can show you the whole thing.
The system this very site runs on
Private client portals, time-tracked to the hour, multi-currency invoicing, threaded conversations and day-accurate subscription billing — the same platform every engagement above is delivered through.
EU compliance cockpit · AI Act · GDPR · NIS2
Know where you stand across the regulations that touch you, with an advisor that cites the binding article and abstains when it can't. Where the design language you're reading began.
The cleanest signal a vendor can send is restraint about other people’s names.