The use-case landscape
The five families of finance AI — document intelligence, close acceleration, forecasting, anomaly detection, narrative reporting — and an honest maturity grade for each.
Vendors pitch finance AI as one glowing category. It's five distinct families with very different maturity levels, and knowing which is which protects both your budget and your credibility:
- Document intelligence (mature — deploy with confidence). Invoice extraction, receipt matching, contract-terms pulling, statement parsing. The technology works at 95%+ on decent documents; the craft is the exception workflow for the rest (Module 2). If your AP team still keys invoices by hand, this is the first dollar you spend, and the payback is measured in weeks.
- Close & reconciliation acceleration (mature-ish — deploy with review). Auto-matching transactions, drafting journal-entry support, flagging unreconciled items, assembling close checklists. Strong assist; the matching thresholds and the review gate are where judgment lives.
- Forecasting & planning (useful — with discipline). Driver-based models drafted faster, scenario branches generated, seasonality surfaced. The honest framing: AI accelerates the analyst's forecast; it doesn't replace the analyst's ownership of assumptions (Module 3 drills this). Fully-automated black-box forecasts answering to nobody are how FP&A loses the CFO's trust.
- Anomaly & fraud detection (powerful — high false-positive tax if lazy). Duplicate payments, policy violations, odd vendor patterns, expense outliers. Works beautifully when tuned to your baselines and thresholds; drowns the team in noise when bought as magic (Module 4 teaches the tuning).
- Narrative & reporting (convenient — watch the numbers). Variance commentary drafts, board-pack narratives, policy summaries. Real time-saver with one iron rule: every number in AI-drafted narrative gets verified against the source — models fluently 'improve' figures while polishing prose, which in a finance document is a career event.
Where the value concentrates at Alder (and probably at yours)
Priya's team mapped their month against the five families: invoice processing eats 340 hours/month (document intelligence — mature, huge volume, clear rules); carrier-invoice audit against contracted rates is sampling-only today (anomaly detection — the misbillings they don't catch are real money); the 13-week cash forecast takes two days of assembly before an hour of thinking (forecasting — the assembly is the automatable part); and month-end variance commentary is a day of writing (narrative — draftable, verifiable). Notice the pattern the whole course follows: the wins attack assembly, extraction, and flagging — never judgment, approval, or sign-off.
For any pitched finance-AI capability, ask: 'What error rate do your current customers see on documents/data like ours, and what does the exception workflow look like?' Mature categories have crisp answers with numbers. Immature ones answer with roadmap. You now know enough to hear the difference — and Module 5's evidence requirements give you the follow-up questions.