AI review engine
AI that proposes. Accountants who decide.
Sorobas uses machine learning to read, reconcile and categorize, but it never makes a judgment call. Every decision traces to a rule or a precedent, and people review every file.
Traceable decisions
Each line records the rule, precedent or model version behind it, its confidence and who confirmed it. Ask why any line was coded.
Checks before auto-posting
A line posts on its own only when it passes every check: source, track record, confidence, account risk, amount, statement and flags.
Two levels of review
A team member reviews every period, including a random sample of auto-posted lines. The partner reviews with the sales tax return and at year end.
Flags, not treatments
Possible capital items, shareholder balances, related-party mismatches and payments to family members are flagged for a person to decide.
Review-time checks
Figures are compared with the client's prior periods and the vendor's history, so unusual swings surface before partner review.
Learns from your firm
Every confirmation becomes precedent and every correction carries a reason, so the engine improves with each period.
Private AI, hosted in Canada.
The models are open-source and self-hosted on Canadian servers, on a private network with no internet access. Client files are never sent to an outside AI provider.
- No AI-written letters or emails to clients
- A precedent is retired the moment it proves wrong
- Tested against reference files before every model change
- Staff see their own coaching; partners see firm-wide trends
Now in beta. Accepting new firms.
Pricing is quoted firm by firm. Tell us about your practice and we'll arrange a walkthrough.