Bankruptcy within a year
The company goes bankrupt within the next 12 months.
Finds 3 out of 4 bankruptcies a year ahead of time.
The TIC Risk Open Model is an open-source AI model with 232 million parameters, trained and tested on the public records of 1.5 million Swedish limited companies. The same principle can be applied in other countries. It's trained on events occurring in a company and estimates the chance that something goes wrong in the coming year. We have made a sample of four Swedish companies, with figures for October 2026.
The model is not trained on any personal data. It only contains limited companies and their data.
What Swedish authorities have recorded about the company: annual reports, board changes, late fees, unpaid debts.
It has learned from the records of 1.5 million Swedish companies which patterns put a company at risk.
For each question it gives a chance, such as 2 %: of 100 companies with that chance, about 2 are affected.
Based on real Swedish limited companies, with their details slightly changed.
Wholesale trade · started 2021
Grocery retail with a broad range · started 2022
Retail trade · started 2021
Construction of residential and non-residential buildings · started 2021
The model answers five example questions about each company.
The company goes bankrupt within the next 12 months.
The company goes bankrupt within the next 24 months.
Within a year, the Swedish Enforcement Authority (Kronofogden) is asked to collect a new unpaid debt from the company, or an old one grows.
Within a year, the Swedish Tax Agency revokes the company’s F-tax approval, which it needs to invoice as a business, because it broke the rules, for example by not filing or not paying tax.
Within a year, the Swedish Companies Registration Office (Bolagsverket) charges the company a fee for filing its annual report late.
A chance of 2 % means that of 100 companies with that chance, about 2 go bankrupt within a year and 98 do not.
2 in 100
To make companies easy to compare, the chance of bankruptcy within a year also gets a risk class from 1 to 5.
The model uses 69 million records of 54 kinds from 1.5 million companies.
No personal data: the model sees that a board seat changed, but nothing about the person.