New: what governed autonomous procurement actually means
Explainer

NOVA intelligence: how it works and what to require

In short

NOVA intelligence answers procurement questions across connected spend, supplier, contract and workflow data, summarises the history behind a decision, and explains why an agent acted, paused or escalated. From an answer, an authorised user can launch the governed workflow that acts on it, so intelligence converts into completed work.

See NOVA intelligence

Questions that wait for an analyst. Answered while you ask them.

Twenty-four questions currently answered by building a spreadsheet first.

Spend and commercial

  • What did we spend with this supplier across every entity?
  • Which plants buy this material and at what delivered price?
  • Where is spend running outside an agreement?
  • What is our exposure if this supplier fails?
  • Which categories have not been sourced in two years?
  • What did we agree on price review for this contract?

Supplier and quality

  • What is this supplier's open quality position?
  • Which suppliers have certificates expiring this quarter?
  • What corrective actions are overdue and with whom?
  • Is this supplier approved for this entity and category?
  • What changed at this supplier in the last year?
  • Which suppliers are single-source for a critical part?

Operational status

  • Where has this request reached and what is it waiting on?
  • Which approvals are overdue across the group?
  • What did the agents do overnight?
  • Which sourcing events are open and at what stage?
  • What was written to the ERP today, and did anything fail?
  • Which exceptions are waiting for a person?

Evidence and audit

  • Why did this purchase follow this route?
  • Who approved this exception and on what basis?
  • Show every action taken on this transaction
  • What rule blocked this order?
  • Which overrides happened this quarter and by whom?
  • Assemble the qualification file for this supplier

An answer you can check. Not one you have to believe.

Six properties, and the last one disqualifies most assistants.

MechanismWhat it has to do
Plain-language questionsAsked the way a person would ask a colleague, across spend, supplier, contract, purchase order, invoice, quality and agent activity together.
Answers with the records attachedEvery answer shows the underlying records, so it can be checked rather than believed. An assistant that cannot be verified will not be used for anything that matters.
Long histories condensedThe relevant part of a two-year supplier file assembled before a call, rather than read through during it.
Scoped by role and entityThe same question from a group leader and a plant buyer returns the answer each is permitted to see.
Answer to actionAn identified issue can start the work that addresses it — open an event, raise a corrective action, escalate an approval — without changing screen.
No answer without a sourceWhere the data does not support an answer, it says so rather than producing a plausible one. In procurement a confident wrong number is worse than no number.

A description of what the practice requires, not a feature list.

Same question format. Different question by sector.

What a CPO actually asks, by industry.

Automotive

Launch readiness by programme and supplier, and the open corrective-action position before a new award.

Consumer electronics

Every product and entity affected by a component change or end-of-life notice.

Aerospace and defence

The complete decision history for a purchase an auditor has questioned, years after the fact.

Building materials

Delivered cost for the same material across every plant, and where the spread is widest.

Five questions. Starting with one it cannot answer.

A system that invents a plausible number is disqualifying.

  1. Ask a question the data cannot answer. A system that invents a plausible number is disqualifying.
  2. Ask the same question as two different roles and confirm the answers respect permission.
  3. Ask for the records behind an answer and check they actually support it.
  4. Ask a question that spans spend, contract and quality at once — that is where connected data shows.
  5. Ask what the reader can do next from the answer, without leaving the screen.

Definitions. Asked and answered.

NOVA answers questions, but the point is what happens next. An authorised user can launch the governed workflow directly from the answer, which is the difference between an assistant and an operating platform.

From your connected procurement, supplier, contract and workflow data, with the underlying records shown so the answer can be verified rather than trusted blindly.

No. Answers respect the same role- and entity-based access that governs the rest of the platform. A plant user does not see group data by asking a question differently.

What the agent observed, what it was authorised to do, the rule that applied, the action it executed and where it escalated. That record is what security and audit teams review.

No. It prepares the decision and can launch authorised work once a person has decided. Consequential judgement remains with the roles your policy names.

Conversational analytics lets a person ask a question in plain language and receive an answer assembled from underlying procurement data, rather than requesting a report or building a query. Its value depends entirely on whether the answer shows its sources.

A dashboard answers questions someone anticipated when they built it. An assistant answers the question actually being asked, including ones nobody designed for. The trade-off is verification: a dashboard's logic is fixed and reviewable, so an assistant has to show its records to earn the same trust.

It can be trusted for retrieval and summarisation where every answer is traceable to source records and access is scoped by role and entity. It should not be trusted to make consequential decisions, which is why those remain with named people under threshold and approval control.

See it on your own workflow. Not a prepared scenario.

Reference material explains the model. A demonstration on one of your own processes is what settles the internal argument.

Someone from client success replies, not a sales sequence. If we are not a fit we will say so on the first call.