Buying an AI tool is not the same as being accountable for the call.

AI is showing up in remittance matching, variance drafts, planning scenarios, and exception queues — and not only inside finance. Sales, ops, and HR run their own tools while you still own the risk story and the reporting pack. Accountability can’t live only in IT.

Why it matters for finance teams: when audit, the board, or another department asks what happened, you need inputs, model or version, proposal, downstream action, and who approved — in under five minutes. Soft on vendors. Sequencing and controls first.

Soft on providers. Hard on whether you can defend the walkthrough.

Wooden card-catalogue drawers, with a gold AI scan frame labelled “AI scan · audit trail”
Accountability is a trail, not a SKU.

What accountability means (five things you can defend)

  1. Audit trail — reconstruct inputs, model/version, proposal, downstream action, and who approved.
  2. Human-in-the-loop (HITL) — named humans at critical points (cash, recognition, material disclosures, anything relied on externally).
  3. Policy — written rules that match how work actually runs.
  4. Model / use-case inventory — what's in production, owner, risk tier, and what's shadow.
  5. Cross-dept ownership — RACI across finance, IT/data, risk, and the business owner.

If you can’t answer those five, the pilot isn’t accountable yet — it’s just faster.

Workday ANZ (Matt Lovell, Dec 2025) puts the bar plainly for Australian CFOs: agents need to sit inside audit, compliance and security policies — with a human in the loop at critical points — plus credentials, orchestration, and explainability. That framing travels beyond one ERP.

Categories that help (not a #1 list)

Think in layers. Most mature shops mix them.

  1. AI governance / registry — inventory agents, models, apps, vendors; risk tiers; policy packs; evidence. Illustrative: Credo AI (registry, risk, compliance; Mastercard GenAI governance case study); Holistic AI (shadow AI discovery, inventory, HITL approvals, framework mapping).
  2. Enterprise AI control towers — cross-estate visibility, ownership, risk workflows, runtime oversight. Illustrative: ServiceNow AI Control Tower (discover; govern risk/compliance; observe runtime; measure value).
  3. Model risk / MLOps governance — factsheets, evaluation, drift/bias/safety monitoring, approvals. Illustrative: IBM watsonx.governance; DataRobot AI Governance (lineage, policy enforcement, compliance docs, production guards).
  4. Finance platforms with embedded AI + trail — close, AR/O2C, planning where controls already live. Illustrative: BlackLine (Verity recon/match logs); HighRadius (agentic AP/AR with approvals); Anaplan Intelligence (AI Gateway + Agent Studio guardrails); Workday (HITL and audit-oriented governance messaging).

None replace chart-of-accounts discipline or month-end ownership. They encode it — if you design them that way.

Multi-department reality

Sales assistants, ops bots, HR screening tools — residual risk can land in disclosures, provisions, or reputation. Finance doesn’t need every prompt. It needs a seat on inventory, risk tier, and who signs when numbers move. Cross-dept RACI beats a lonely IT register nobody updates.

Buyer’s checklist (finance team)

  1. Reconstruct the decision — inputs → version → proposal → approval in under five minutes?
  2. HITL gates — named roles for cash, master data, externally relied-on outputs.
  3. Cross-dept RACI — business owner, finance control owner, IT/data steward, risk reviewer.
  4. Data lineage — sources, cut-offs, what the model could read.
  5. Change notice — vendor model updates logged with materiality before ICFR-adjacent reliance.
  6. Shadow AI scan — what's already outside the approved list.

Sequence: inventory + RACI + trail → then widen autonomy. Tools amplify what you already have — including gaps.

AI accountability gets sold as a product category. In practice it is decision design: know what is in production, name who signs, and keep a trail you would defend in a walkthrough.

Full categories, illustrative shortlist, and finance team checklist stay on FinanceSignal.ai — peer notes for finance people who have to live with the call.