# Table Stakes, AI Now and AI Wow: a buyer’s guide to the work behind the label

> Use evidence, scope, retained authority and verified outcomes to evaluate three levels of finance automation.

- **Source:** https://www.runproper.com/resources/ai-and-controls/ai-table-stakes-now-wow
- **Publisher:** Proper Technologies Inc.
- **Format:** Article · AI & controls
- **Reading time:** 3 min

Proper uses Table Stakes, AI Now and AI Wow as a way to examine the work a system carries. Table Stakes is strong routine automation; AI Now is a difficult bounded investigation; AI Wow is a proposed standard for sustained work across changing evidence and dependencies. The label alone does not establish availability or superiority.

## Table Stakes: expect a complete routine workflow

A credible baseline should carry useful routine work under approved rules and make exceptions actionable. In AP, that can mean capture, coding, duplicate checks, routing and acknowledged posting. In Payments, it includes controlled release and honest bank status. In Accounting, it includes supported schedules, reconciliations and period controls.

Some vendors use AI within those activities. That does not make the outcome extraordinary, and a native accounting or ERP system can provide a strong baseline. Evaluate the result, exception path and integration coverage rather than rewarding a label.

## AI Now: investigate a bounded problem with evidence

A useful Now story starts with a job a capable person could do, but would need time to reconstruct. The AP example links an invoice to amended terms and accepted delivery evidence. The AR example traces a rejected invoice through a changed order reference and supporting records.

The useful work is finding and interpreting the governing evidence, then preparing a reviewable resolution. Multiplication, ordinary matching, a generic reminder or a routine cash forecast does not alone demonstrate that burden. Ask to inspect the sources, uncertainty and proposed action before the system acts.

A conditional answer can be more valuable than an unsupported confident one. If an amendment is missing, the case should remain unresolved with a clear next step. If a source arrives later, the reasoning and affected proposal should be revisited.

## AI Wow: require sustained work that earns the description

Wow is an ambition for carrying a meaningful workload that a lean team could not practically maintain at the same breadth, cadence and duration. The proposed AP story follows supplier entitlements as terms and evidence change, with claims continuing through supplier response and credit application.

Adding more invoices, another alert or a larger savings number is not enough. The demonstration must explain the continuing reasoning and dependency burden, what remains running after the first answer, and how completion is verified. Compare that work with mature automation already available in the market.

The AI Wow films in this library present future concepts. Creative completion or website placement does not establish a released capability. Exact scope, failure handling and end-to-end performance still need product evidence.

## Apply the same control questions at every level

Separate access to information from authority to act. A system may be allowed to read a document, prepare a proposed correction and assemble supporting evidence while still requiring approval to send a message, change a record or release funds.

Approval should identify the action and version actually reviewed. New facts that materially change that action can require renewed authority. Likewise, a request is not an issued credit, a customer promise is not cash, and a draft journal is not a posted entry.

-   What sources can the system inspect, and what is missing?
-   Which work follows existing rules and which requires interpretation?
-   Which person or policy authorizes the proposed action?
-   What confirms completion, and who owns unresolved cases?

## Run one test that changes after the first answer

Bring an anonymized case with a clear objective and then introduce a realistic change: a new amendment, a short receipt or a rejected posting. Observe whether the system updates only the affected work, preserves history and keeps uncertainty visible.

Record the product edition, region, connectors and permissions used. Synthetic figures explain a workflow; they are not measured customer savings. The best buying evidence is a repeatable demonstration of useful work, retained control and an acknowledged outcome within your actual scope.

## Questions & answers

### Does AI Now mean that every shown workflow is released?

No. The tier describes the intended type of work. Confirm the exact available capability, configuration and evidence separately.

### Does AI Wow mean no competitor can do it?

No. It is a demanding proposed outcome standard, not a claim of market exclusivity. Compare it against current alternatives using the same cases.

### How should we evaluate an autonomous claim?

Ask what authority was delegated, what constraints apply, how changed facts affect approval, and what evidence confirms the result. Broad autonomy wording is not an adequate control description.
