InsightsMarket notes

Market notes

4 / 35

profiled systems document an AI module

The AI layer: what is generally available, and what is on stage.

Four of the systems Stackmark profiles document an AI module. In demonstrations the layer looks finished. In contracts it is often a roadmap. Here is how to tell the two apart before the shortlist is set.

· 2 min read

Every finance-system demonstration this year has had an AI segment. Invoice capture that codes the line, a journal anomaly flag, a cash forecast that narrates itself, a chat box over the reports. The segment is polished, and it is usually the part the room remembers.

Stackmark records an AI module for a vendor only when the vendor documents it as offered. Four of the systems profiled meet that bar today. That is not a small number for a module that did not exist in most product catalogues three years ago, but it is a long way from the impression a season of demonstrations leaves.

Three states a feature can be in

A useful habit is to ask, for each AI capability shown, which of three states it is in. The vendor knows; the demonstration rarely says.

  • Generally available: every customer on the current release can switch it on, and the price list says what it costs.
  • Limited availability: a named set of customers has it, often under a design-partner agreement, and the vendor decides who joins.
  • Roadmap: it exists as a build, a prototype or a slide, and the contract will not commit to a date.

The module page for AI features carries four demonstration questions written to force that answer. The first is the blunt one: which of these are generally available to all customers today? Ask it before the segment starts, not after.

Where the data goes

The second question buyers under-ask is about data residency for the model. A capture feature that sends an invoice image to a third-party model is a data-processing arrangement, whether or not the vendor calls it one. It belongs in the security review, not in the demonstration debrief.

If the vendor cannot say where the model runs, the feature is not ready to be scored.
Stackmark demonstration guidance

What this means for the shortlist

Do not remove a system from a shortlist because its AI layer is thin. The transactional core is what runs the close; the AI layer is what will change fastest over the contract term. Do remove a system that cannot answer the three-state question, because a vendor that will not say which state a feature is in is asking to be scored on a roadmap.

On this site

More notes

Keep reading.

All insights