The marketing agency, rebuilt as intelligence.
Built in Denmark for e-commerce and SMB teams who don't need another retainer. They need to know the one thing to do next, and the confidence that it'll actually get done.
Agencies are expensive to be told the obvious.
Most small and mid-sized e-commerce and service businesses can't afford a full-time growth team, and a traditional agency retainer often means paying five figures a month to be told things a good analyst could surface in an afternoon, if you had one on staff.
Kompena connects the same data sources an agency would ask for, runs the same kind of diagnostic thinking through a deterministic scoring engine, and turns it into a prioritized, explainable plan, for a fraction of the cost and rerun whenever you want it instead of once a quarter.
We don't think software should replace judgment. It should remove the grunt work of finding where to apply it.
Three things we won't compromise on.
Explainable over impressive
Every score and every value estimate traces back to a real number. If we can't show our work, we don't ship the claim.
Prescriptive over descriptive
A dashboard that tells you what happened is easy. Telling you the one thing to do next, and what it's worth, is the actual job.
Follow-through is the product
Insight nobody acts on isn't worth much. The roadmap and execution layer exist because analysis alone doesn't move revenue.
Your growth, measured all the way round.
We read your numbers where they already live: your website, your ads, your search results and your visitor numbers. We never change anything of yours. You get the next move and the method for it, and you do the work. It is remembered twice: what happened at your place, which nobody outside your organisation can read apart from our own support staff, and what measurably worked, rolled up with businesses like yours. The loop that corrects your next estimate runs every night and has not yet had a finished piece of work to measure. The goal never moves: your growth.
What we know about your business
We read your numbers where they live, and change nothing
Analyses and the audit turn numbers into findings
You put the work live, a human all the way
Measurements and your corrections are kept: with you, and rolled up across businesses like yours
Every new reading is held up against your previous one
The ring closes on your own numbers today: every new reading is held up against your previous one. The night loop that corrects the next estimate is wired and running, and it is waiting for the first completed work to measure.
Every status is counted from the parts of that layer, in the code. Nothing here is promised.
Your business
Holds what your business is: the trade you picked, the nine things your front page says about your brand, the competitors we found or you named, and the numbers we can measure.
Your CRM has no connector yet. All three prose fields are read by the prompts only; no engine computes on them, and the budget plan still optimises measured spend, never your written budget.
Agents
A nightly agent runs your analysis and your audit and reports back in the bell, and you can see every run it made. You can also start everything yourself: an analysis is data a deterministic evaluator runs, a skill is a file you run in your own AI, and the audit is a single endpoint.
The nightly agent runs the analysis and the audit and reports back in the bell, but it takes no instruction yet: you cannot tell it what to keep an eye on, and it cannot run a skill for you. It produces material and never publishes it, which is why every skill we hand out is read-only.
Workflows
Carries one piece of work from recommendation to done: you accept it, it lands in a 30, 60 or 90 day bucket, and it gets a status you move as the work happens. A task can only move between the states our table allows.
There is exactly one workflow, and it is ours: you cannot build another, add an approval step, or branch it. Nobody is nudged unless they open the page.
Data sources
Reads your numbers where they already live, and never changes anything. One Google login covers Analytics and Google Ads today, and ten sources need no login at all and read from your first day: your own site, Core Web Vitals, search results, Domain Rating and both ad archives. A number that was not measured is never written.
Nineteen of the nineteen connectors fetch real numbers today. What remains is not code: Meta's reading waits on Meta's app review, and Trustpilot runs only for a workspace that switched it on. Klaviyo and HubSpot take a pasted key on the data sources screen. Salesforce and Pipedrive are next in the CRM queue and not in the product yet.
Memory
Two memories, kept apart on purpose. Your history remembers what happened at your place: every audit with its score and how much of the picture it could see, your corrections in your own words, never rewritten, and your numbers over time. Nobody outside your organisation can read it, apart from our own support staff. The shared experience remembers what tends to work for businesses like yours: measured outcomes rolled up per play across customers, in open tables that can answer why a number is what it is.
Your notes, strategy and notifications are kept with us. The shared experience is grouped on three coarse trade roots so no single business can be read out of it, and it has learned nothing real yet: no customer has completed a play and been measured. There is no vector database and there will not be one: memory you cannot read cannot answer why you scored 46.
Automation
Two loops run by themselves. Every time your numbers are read anew, they are compared against your own previous reading, and a drop or a jump is raised together with the recommendation that answers it. The second loop runs every night and is wired to correct your next estimate with what completed work actually delivered, from the day completed work is written down.
Nothing reaches you unless you open the page: we send no email and no push. The night loop has measured nothing yet: completed work is written down now, but no customer has completed a play and sat through the settle window. Neither loop reads the memory yet: they read your numbers, not what you have told us. And nothing yet ranks the likely causes when something falls.
The difference between Kompena and a tool that guesses is not that we do more. It is that we tell you what we could not see.
That is why every layer above says out loud where it stops today. A claim never travels without its limit, and a score never travels without how much of the picture it saw.
Plain words, on purpose.
Every part is named after what it does for you. This is the whole list, and there is nothing more to explain.
A question we answer with your own numbers, against a threshold you can see.
A method you take with you as a file and run in your own AI, again and again. We never run it, and it never publishes anything.
Something that keeps an eye on your numbers on a fixed rhythm and speaks up when something moves. You do not have one yet.
A place your numbers already live, that we may read and never change.
Everything we have measured at your place and everything you have corrected, kept so you never explain yourself twice. Nobody outside your organisation can read it, apart from our own support staff.
What has measurably worked for businesses like yours, rolled up across customers so no single business can be read out of it.
One piece of work, from the recommendation to done.
Work that happens without anyone starting it.