- H2O Driverless AI learns which journal entries and balances look unusual, from your own ledger history and your auditors' past findings.
- Every entry gets a risk score with reason codes, so your auditors see why an item was flagged before they open it.
- Scoring runs on every entry as it posts, moving your audit from a quarterly sample toward continuous monitoring.
- Auditor findings feed back into the model, so it keeps learning from what your team confirms.
Your auditors test samples, and errors hide in the rest
Your finance and internal audit teams test a sample of entries after the period closes. An error or a fraudulent entry outside the sample can stay hidden for months.
Much of the work is manual: pulling entries, comparing them with support, and reading documents to confirm them. That time comes out of the complex, high-risk cases where your auditors add the most value.
How AI accounting and auditing works
Your audit analytics team builds the models in H2O Driverless AI, and the scores feed the audit tools and workpapers you already use.
Step 1: start from the ledger
Load journal entries, account balances and posting metadata such as user, time, source system and approval path. Add the outcomes of past audits: which entries were questioned, adjusted or confirmed as errors.
Step 2: learn what unusual looks like
Anomaly detection and unsupervised machine learning find entries that do not match normal patterns, such as round amounts posted at period end, unusual account pairs or postings outside normal hours. Where you have past findings, H2O Driverless AI trains classification models on them, so your auditors' own judgment shapes the score.
Step 3: score every entry and rank the work
Each entry and balance gets a risk score with reason codes. Your auditors start with the highest-risk items, and the annotations they add become new training data.
Step 4: monitor continuously
Scoring runs as entries post, so a pattern that appears mid-quarter is flagged mid-quarter. Your team reviews a short, ranked list each week instead of a large sample each quarter.
Step 5: document the method
H2O Driverless AI writes model documentation while the model trains, and reason codes show which factors drove each score.
Where auditor judgment still decides
The model ranks the work and explains each score. Your auditors decide what to test, what to conclude and what to report, and their findings feed back into the model and improve it. The models run inside your own environment: on-premises, air-gapped, private cloud or cloud.