Traceable from source to result
Every value keeps its link to the source document and to the checks that led to the result.
We build dependable AI systems for fund managers, lenders and asset managers in the Nordics.
45 min · a written assessment of what you can automate, within five working days of the call
See in the examples below what it can look like when we automate different workflows.
A drawdown request arrives with its attachments. Request 5 is for SEK 8,640,000. The checks run one at a time: the stage, the insurance, the loan to value and the accrued interest. Four checks pass without remarks. The fifth check deviates: the amount requested is more than the monitor has certified. The documents show a deviation of SEK 140,000. The invoice shows it is a change order outside the stage budget. A proposal of SEK 8,500,000 to approve, and SEK 140,000 left to judge. The case goes to the credit manager, who decides.
Example data. The companies in the documents are invented.
Two sources are read in: the fund’s books and the custodian’s holdings. Reconciliation as of 2026-09-17, nine funds, 412 holdings. Holdings are matched per custody account and ISIN, on quantity. Nobody needs to look at what agrees. 411 of 412 holdings agree. One difference remains. Booked quantity 61,500, quantity in custody 60,000. The difference equals a purchase that should have settled. Why it has not is not in the data. The case goes to the fund administrator, with the trade and the statement attached. The reconciliation is done. The team starts where there is something to investigate.
Example data. The fund company, the funds and the holdings are invented.
The documents for a credit case are read: the application, the annual and group accounts and the management accounts. The figures are pulled from the documents, each with a reference to its source. The ratio is calculated in code: net debt/EBITDA 3.2x against the credit policy’s maximum of 3.5x. The annual accounts show a shareholder loan of SEK 9.0m that is missing from the application. With the loan included the ratio is 3.6x. Whether the loan counts is handed to the credit analyst. The assessment is the analyst’s.
Example data. The group and the lender are invented.
The sources for the quarterly report are connected: the fund’s books, calls and distributions, the portfolio companies’ reporting and approved valuations. The report’s figures are filled in automatically. Three of five figures are locked. NAV and TVPI wait for one input. Company E has not delivered its quarterly reporting. The valuation as of 2026-06-30 stays, and NAV and TVPI are marked preliminary. The draft report is ready for review. Following up Company E goes to the investment manager.
Example data. The fund and the portfolio companies are invented.
This is an example. Every system we build is shaped around the client’s own documents, systems and routines.
Data from several systems is gathered and checked automatically. You get a compilation to review before reporting.
Figures from documents and spreadsheets are compiled and ratios calculated. The analyst gets one complete file to assess.
The system checks the documents against your terms and brings forward what needs review before the drawdown is approved.
Items from different systems are matched automatically. You see which breaks need investigating.
The system drafts answers from your documents, with references to the sources. You review the answers before they are sent.
Healthcare
Booking, payments and encrypted video sessions on web and mobile, built to the North American rules for health data. We have developed and run the platform since 2021.
Our own tool, HR
An AI tool in Slack that takes answers from existing systems and moves leave requests on to approval. From idea to working prototype in one week.
Education
Secure video delivery, automatic captions in several languages and a cloud setup that grows with the school, developed and run by the same team for several years.
We go through how you work today and assess where automation would help most. We will be in touch within one working day to book the review.
Financial workflows need traceability, reproducible calculations and clear approvals. We build that in from the start.
Every value keeps its link to the source document and to the checks that led to the result.
Language models can read and structure documents. Rules and calculations run in tested, version-controlled code.
When something is missing, contradicts another source or calls for a decision, it goes to the right person instead of the system guessing.
The code, the operating environment and the documentation are yours from day one.
Entra ID, OIDC/SAML and role-based access.
SAST, dependency scanning, secrets scanning and mandatory peer review.
Audit logs, TLS 1.2+ in transit and AES-256 for data at rest.
Azure EU regions by default, or deployment in the client’s own environment.
Tell us briefly which work you want to simplify. We will be in touch within one working day to discuss your needs and book a review.
Contact
