A Fictitious Company, the Real Method
Aurvik AS does not exist. The company, its founder, and every number on this page are invented. The method is not: the steps, the order, and the thresholds are the ones Attolloo uses.
Aurvik sells software that reads regulatory updates and maps them to a client's own routines, so that a compliance team sees what has changed and what to do about it. The founder has ordered a Readiness Assessment, and wants to know whether a longer partnership would make sense.
- Customers: four, in Nordic asset management and insurance, all paying
- Revenue: NOK 1.2 million a year in subscriptions, 14 months after launch
- Team: seven people. Three in engineering, two compliance experts, one in sales, and the founder
- Costs: NOK 7.45 million a year
Further down, the same method is applied to a second company, one that competes with the established players.
Step 1: The 6Sense Filter Opens With One Question
Before any scoring, one question: what happens if this ends up in the wrong hands?
Aurvik hits no red lines. The product reads public regulation and the client's own routines. It does not identify individuals, and it does not make decisions about them.
On the seven criteria for misuse and dual-use risk, two score low. Both become work for the first sprint.
The seven scores
Each criterion is scored from 1 (high risk) to 5 (low risk).
- Misuse potential: 4. The output is advice to a compliance team, with little use outside it.
- Access control: 3. Access is by role, but changes to the rule library are not logged.
- Customer control: 4. Four known customers, with terms of use.
- Data and privacy: 4. Little personal data, and none of it sensitive.
- Supply chain: 2. The product depends on one external language model. Sub-suppliers are not mapped, and there is no plan for replacing the model.
- Regulatory maturity: 2. The product has not been classified under the EU AI Act, and nobody is named as owner.
- Governance and culture: 3. The founder takes misuse seriously, but there are no written principles or escalation routines.
Step 2: The Six Dimensions
Four dimensions pass. Two pass on conditions, which is the most the filter allows. One more, and the answer would have been no.
The six dimensions, with the numbers
① DIRECTION: PASSES ON A CONDITION
Four customers pay annual subscriptions on signed contracts. That is demand shown by payment, not by market size. But all four came through the founder's own network. Condition: within 30 days, two qualified buyers from outside that network.
② ENERGY: PASSES
Each customer pays NOK 300,000 a year at a gross margin of 78 per cent, and costs NOK 85,000 to win. Assuming a customer stays five years, that is NOK 1.17 million in lifetime value, about 14 times the cost of winning the customer. The threshold is 3. The cost is earned back in 4.4 months. The five years are an assumption: no customer has left yet, but 14 months is a short history.
③ CAPACITY: PASSES
Customers pay within 28 days and suppliers are paid after 45, so growth does not eat working capital. The company loses NOK 6.25 million a year and breaks even at about 32 customers. The real limit is people: every regulatory change is reviewed by hand by two compliance experts.
④ PRIORITISATION: PASSES ON A CONDITION
AI Compass places the company at 2.5 stars. The minimum for an early-stage venture is 2. Condition: a full baseline in the first two weeks, and a plan for 3 stars within six months.
⑤ RISK: PASSES
The pre-mortem asks: assume Aurvik has failed in 24 months, what happened? The most expensive answer is that the model gave a client wrong compliance advice. The two low scores from the first step belong here as well.
⑥ DECISION: PASSES
Three alternatives are compared: an executive partnership, a board or advisory role, and no further engagement. The comparison is in the next section.
Where the Filter Corrected the Numbers
Most of the value of a screening lies in the numbers it changes. In this case there were three, and one of them was Attolloo's own.
THE VALUATION
The founder opened with a valuation of NOK 50 million. Built from the bottom, yearly subscription revenue of 1.2 million times a multiple of six gives 7.2 million. The multiple is an assumption, but the distance is not a rounding error. The first number had anchored the conversation.
THE FORECAST
The founder expected to go from 4 to 50 customers in a year. Without a sales team, the base case was set at 13 to 20 customers, to be checked against the pipeline every month.
ATTOLLOO'S OWN ESTIMATE
Attolloo's first estimate gave an executive partnership a 65 per cent chance of success. That is the same optimism the founder was corrected for. The decision was therefore tested at 30 per cent.
Should this become an executive partnership?
The assessment was a fixed-price job. The question below is what comes next.
- An executive partnership over 24 months: 4 per cent equity and a fee of NOK 20,000 a month, NOK 0.48 million in all. If it succeeds, the stake is worth NOK 2.4 million, assuming a sale at 60 million. The time spent is valued at 1.2 million, so the fee covers part of it and the equity has to cover the rest.
- A board or advisory role: a fee agreed in advance and no equity. The fee pays for the time, so there is little to lose and no share in the upside.
- No further engagement.
If the partnership succeeds, Attolloo has the stake and the fees: NOK 2.88 million. If it fails, the fees cover 0.48 million of time worth 1.2 million, a loss of 0.72 million. At a 30 per cent chance of success, that gives an expected value of NOK 0.36 million. The advisory role roughly pays for itself, and stopping after the assessment comes to zero. The partnership is worth more than stopping as long as the chance of success is above 20 per cent. Below that, the answer is no.
Step 3: AI Compass
After the filter comes the baseline: where does the company stand today, and what should be different in 90 days?
Aurvik is aimed at a real problem and creates value a client can measure. It is weaker on the two dimensions that institutional buyers ask about.
The four dimensions
DIRECTION: 3 STARS
The product is aimed at a specific, testable problem: the 15 to 40 hours a week a compliance team spends following regulation by hand.
VALUE: 3 STARS
Clients can measure hours saved. But Aurvik can only see whether an analysis was read, not whether the client acted on it.
READINESS: 2 STARS
Engineering, compliance and sales work in three separate groups. The link between the model and the rule library is undocumented and breaks at every major update.
INTEGRITY: 2 STARS
Reports are reviewed by a person before they go to a client, but the review is not logged. There is no EU AI Act classification, and no plan for replacing the external model.
Step 4: The First 90 Days
The first Polaris sprint is built from what the screening found, not from a standard plan.
THE CONSTRAINT
Kando names one constraint: the two compliance experts. Before anyone is hired, routine classification of regulatory changes is taken off their desk, so their time goes to the cases that need judgement.
THE WORK
- An EU AI Act classification, with a named owner
- A map of suppliers, and a plan for replacing the external model
- A log of human review for every report sent to a client
- Two teams built around the work, each with compliance expertise, instead of three separate groups
- A monthly forecast model in place of the founder's estimate
HOW THE GAIN IS COUNTED
As capacity: from four to seven customers with the same seven people.
THE EVIDENCE AT DAY 90
AI Compass from 2.5 to 3 stars, and both conditions from the filter closed. If a condition is not met in time, the engagement pauses.
A Second Company: The Challenger
Tidevik AS does not exist either. Where Aurvik sells to the established players, Tidevik competes with them.
Tidevik lets small businesses take payments and have the money the same day. An AI model screens every payment for fraud and decides, in real time, whether to let it through. The company runs under a partner bank's licence, and plans to apply for its own.
- Customers: 1,200 small businesses, all paying
- Revenue: NOK 5.76 million a year in subscriptions and transaction fees, 20 months after launch
- Team: 14 people
Tidevik has no procurement function to get through. Its customers sign up online in ten minutes. The ones who have to say yes are the regulator, the partner bank, and the investor in the next round.
What the Assessment Found
No red lines. Five of the six dimensions pass, and one passes on a condition. AI Compass places the company at 2 stars, the minimum for an early-stage venture.
The numbers were not the problem. The company rested on one agreement and one person.
The findings in detail
THE FIRST STEP: RISK OF MISUSE
Two of the seven criteria score low.
- Data and privacy: 2. Payment data on 1,200 businesses and their customers, and no impact assessment.
- Regulatory maturity: 2. The fraud model has not been classified under the EU AI Act, and nobody is named as its owner.
② ENERGY: PASSES, NARROWLY
Each customer brings in NOK 4,800 a year at a gross margin of 55 per cent, and costs NOK 2,900 to win. Assuming a customer stays four years, that is NOK 10,560 in lifetime value, 3.6 times the cost of winning the customer. The threshold is 3. The cost is earned back in 13 months.
⑤ RISK: PASSES ON A CONDITION
All revenue runs through one agreement with one partner bank, which can end it at 90 days' notice. Condition: within 60 days, either a longer notice period or a second partner in dialogue.
⑥ DECISION: PASSES, WITH ONE FINDING
The model stops about 2 per cent of payments. Nobody can say who set that threshold, or when it was last reviewed. Decision cannot pass on a condition, so the finding goes to the top of the 90-day plan: a named owner and a written decision within 30 days.
AI COMPASS: 2 STARS
The model's decisions are logged in a form only engineering can read. A customer whose payment is stopped has nobody to complain to. The partner bank's last questionnaire took five weeks to answer.
THE FIRST 90 DAYS
Kando names one constraint: the CTO, the only person who can answer the partner bank and the regulator. Before anyone is hired, the routine answers are taken off her desk and put in a standing evidence pack.
- A named owner and a written decision for the threshold that stops payments, within 30 days
- An EU AI Act classification of the fraud model, with a named owner
- A readable log of stopped payments, and a person who handles complaints
- An impact assessment for the payment data
- An evidence pack that answers the partner bank's standard questions
- New terms with the partner bank, or a second partner
THE EVIDENCE AT DAY 90
AI Compass from 2 to 3 stars, the partner bank's questions answered in days instead of weeks, and the condition closed.
What the Examples Show
A screening that only confirms what the founder hoped is not worth much. Here it changed a valuation, a forecast and Attolloo's own estimate, set two conditions, and turned the two weakest points into the first 90 days of work. For Tidevik it showed that a growing company rested on one agreement and one person.
Take the self-test to see where your own company stands, or send a message.