This is part 3 of Finance AI Field Notes. Part 2: How finance leaders actually plan to adopt AI.
Strategy decks tell you what a CFO hopes. Job descriptions tell you what a CFO will pay a salary to get. When a role says "you will be measured on X," X is the real goal.
I read about 105 live finance-AI job postings from July 2026. Here are the goals CFOs wrote down, in their own words, roughly in order of how often they appear.
1. Close speed — with integrity attached
The close is the universal target. Rohlik Group tells its new Head of Finance Automation to "track what matters and report it straight: automation rate, close time, cash allocation speed, cost and effort taken out." Shopfully's AI Architect exists to automate "close, consolidations, variance analysis, management reporting." Gartner's headline prediction — organizations with embedded AI assistants will close 30% faster by 2028 — is directionally what every one of these postings is chasing early.
But notice the second clause that almost always follows. Aon's mandate is agents that deliver "speed, integrity, and audit readiness" for the close. Not speed alone. A close that's two days faster and one reconciliation weaker is a failed project in every one of these JDs.
2. Touchless transaction processing
The volume goal: invoices captured, matched, coded, and posted without a human touching the routine cases. Cengage's AI Automation Engineer is scoped to "automate accounts payable and accounts receivable processing including invoice capture, matching, and approval" and to "build reconciliation automations across GL, bank, credit card, and intercompany accounts." CSC Generation wants agents for "A/R, A/P, and General Ledger reconciliations — eliminating manual touchpoints." The metric behind these sentences is percentage-touchless, and the honest targets I see practitioners set are workflow-by-workflow (90%+ on bank reconciliation, lower on judgment-heavy coding), not a blanket number.
3. Headcount leverage, not headcount replacement
Read closely: nobody in these postings is hired to fire the accounting team. The pattern is one builder who multiplies the existing team. Rohlik: a team of three, five markets, one Head to make it scale. Kraken: one architect for a global finance org, explicitly told to "create reusable frameworks… so that future Finance team members can safely build on top of the platform." The goal CFOs write is capacity — absorbing growth, new entities, an IPO-readiness workload — without linear hiring. "Do more with the same" appears in many forms; "do the same with fewer" almost never does.
4. Audit-readiness as a deliverable, not a tax
This one would not have appeared two years ago. AI governance is now written into the finance goals themselves: Kraken requires "human oversight where appropriate… design for auditability and control integrity, not just efficiency," and mandates that "no build goes to production without a named Process Owner, documented data flows, access controls, and audit logging confirmed." Aon wants "agent logging, execution traceability, and override controls." The sophisticated CFOs have realized that in a SOX or statutory-audit environment, ungoverned automation isn't a head start — it's rework. Evidence capture is being specified upfront, as a feature.
5. Multientity consistency
Gartner named "inconsistent multientity support" as a top blocker, and the postings confirm it from the demand side: Rohlik must roll out "across our five markets, balancing local reality against one Group standard"; Mews runs 32 legal entities; CSC Generation wants one standard "across 13 brands." The goal isn't an agent that works — it's an agent that works identically in every entity, currency, and jurisdiction, or at least fails loudly where it can't. Anyone who has done a consolidation knows this is where finance automation historically goes to die.
6. Forecast quality (the quieter goal)
FP&A goals appear less often than accounting goals, but they're there: Shopfully wants ML for "revenue, expense, cash flow, and driver-based forecasting… backtesting, model monitoring, drift detection." Scale AI wants agents that "automate forecasting inputs" and variance analysis. The measurable versions are forecast accuracy and cycle time. It's telling that most companies sequence this after the transactional goals — you forecast better on a ledger that closes clean.
What's missing from the goal-setting
Two absences worth noting. Almost no posting sets a goal for decision quality — "the AI recommended X and X was right" — which suggests CFOs are correctly keeping AI on prepare-and-propose duty while humans keep decision rights. And very few set adoption goals for the finance team itself; Rohlik ("raise AI fluency across finance") and Kraken ("train finance professionals… drive adoption and operational independence") are the exceptions. That's a gap: an agent nobody trusts gets bypassed, and a bypassed agent delivers none of the five goals above.
If you're setting AI goals for your own finance function this quarter, steal the pattern from the best of these postings: one speed metric, one touchless metric, one control metric, stated together — because any two without the third is how finance AI projects fail quietly.
Part of Finance AI Field Notes, a series grounded in a July 2026 scan of ~105 live job postings plus verified market research. I build Artifi — a platform where the control metric is the default, not the afterthought. Next: the build-vs-buy question, which the data says most teams are framing wrong.