Platform · Security & AI governance
AI governance for procurement that runs on your policy
What the system puts forward — which suppliers to invite, which bid to award, whether an invoice can clear — comes out of rules your organisation set, not out of a model’s opinion.
That is governed autonomy in practice: people set the rules, the system moves the work forward inside them, and anything that falls outside comes back to a person. The scoring is rules; the model writes the explanation on top; and whether it acts without a person is a setting you hold.
- Sourcing required above
- 25,000
- Minimum bids
- Three
- Supplier off the approved list
- Blocked from invitation
- Award mode
- Advisory
Thresholds, bid counts, supplier states, approval rules.
A recommendation you can argue with. Not one you have to trust.
The objection to autonomy in procurement is never that the software is slow. It is that nobody can say why it did what it did, or undo it afterwards. Governed autonomy is the answer to both.
Decision by decision
What the system puts forward. And what decided it.
Each one traces back to something your organisation set. That traceability is what makes governed autonomy defensible when somebody asks.
Which supplier states block a new sourcing invitation, contract or purchase order, and how a supplier off the approved list is handled.
The value above which a sourcing event becomes mandatory, and the minimum number of bids that event has to attract.
The scoring is rule-based. The model writes the narrative that explains it, over a result the rules produced.
Advisory or automatic, set on the sourcing policy alongside the threshold and the minimum bid count.
Not a sole source, not a new supplier, not a first order with that supplier, no gaps in the data, and under the value threshold you set.
Under your limit, the order not over-invoiced, a goods receipt present and quality-approved, no contract price variance, no duplicate flag, no tax mismatch, an exchange rate present. A rules check, with no model in it.
By amount, by category and by emergency, with the approver roles named, acting one-of or in sequence, with escalation timings and delegated authority.
An intake record holds whether the actor was a person, the assistant or the system, the reasoning behind what happened, and the link to the override.
Where this already runs
In operations like yours. Not vendor scenarios.
An oil and gas producer brings 4,000 suppliers onto one record
A supplier base above 4,000, with 50–80 onboardings a month 4,000+suppliers on one governed record50–80onboardings a month through one process3 → 1systems holding the supplier view Read the story
A chemicals manufacturer takes the spreadsheet out of the award
Raw materials, packaging and engineering goods across several business units 1supplier record behind every award0Oracle modules replaced3kinds of evidence in the decision: price, quality, contract Read the storyHow a decision is made
Six things that are true of a recommendation.
Governed autonomy in six steps. People set the rules, the system moves the work forward inside them, and anything outside them comes back to a person. The policy is the input, not a net thrown over the output.
-
01
Human
Your policy goes in first
The value above which something must go to market, the minimum bids, how an off-approved-list supplier is treated, which supplier states block an invitation, and the approval rule that will apply when it comes back. Written by your administrators, per organisation.
-
02
System
The score comes out of those rules
What a recommendation favours is produced by rule-based scoring against the policy above. It is not a model forming a view and a rule checking it afterwards — the rules are what produced the result.
-
03
System
The model writes the explanation
The narrative that accompanies a recommendation is model-written, over a score the rules produced. That is the division: the rules decide, the model puts it into words.
-
04
Human
You decide whether it acts
Advisory or automatic is a setting on your sourcing policy. Left advisory, a recommendation waits for a person. Set to automatic, it proceeds — and the guardrails below still hold.
-
05
Rule
The guardrails hold either way
An award does not proceed on its own where the source is sole-source, the supplier is new, it would be the first order with that supplier, the data has gaps, or the value is over the threshold you set. On the invoice side, eight conditions all have to pass.
-
06
Human
And the override is yours, with the reason kept
Overriding an auction AI recommendation is its own permission, assigned separately from creating, monitoring or awarding the event. An organisation can set that override to require a written reason, which is recorded with it.
Core capabilities
What AI governance for procurement has to answer for.
What produced the recommendation
The question that separates governed autonomy from automation nobody can account for
- Rule-based scoring, run against the policy your organisation set
- The sourcing threshold and minimum bid count as inputs to it
- Which supplier states block an invitation, and how an off-list supplier is handled
- The model writes the narrative that explains the result, not the result itself
- On the invoice side there is no model in the check at all — it is eight conditions
The policy that drives it, in your hands
Set per organisation, by the people accountable for it
- The value above which a sourcing event becomes mandatory, and the minimum bids
- How a supplier off the approved list is treated
- Which supplier states stop a new contract, purchase order or invitation
- Approval rules by amount, category and emergency, with named approver roles
- One approver is enough, or all of them in sequence, with escalation timings and delegation
- Eighteen contract actions and the full set of auction permissions, assigned to roles
Where it stops without a person
The limits that hold even when you have asked it to proceed
- A sole-source award does not proceed on its own
- Neither does one to a supplier you have not used before
- Nor a first order with that supplier, nor one with gaps in the data
- Nor anything above the value threshold you set
- An invoice clears only on all eight conditions, and the automatic limit cannot exceed the amount at which exceptions escalate
- Sourcing autopilot drafts only, is off until switched on, and works inside a value ceiling
What the record keeps
Named places, each one checkable afterwards
- An intake record: whether the actor was a person, the assistant or the system
- The reasoning behind it, and the link to the override, on that same record
- A written reason on an auction AI override, where that rule is switched on
- Before and after values on invoices, contracts, policy changes and supplier restrictions
- The same on certificate compliance and catalogue fields
- A recorded reason on a policy change and on a supplier restriction
Disagreeing with it
Override as a permission somebody holds, not a door left open
- Overriding an auction AI recommendation is assigned to roles on its own
- Separately from who may create, monitor or award the event
- That override can be set to require a written reason
- The reason is recorded with the override
- A recommendation left advisory simply waits for the person who holds the decision
Features
A decision you can stand behind when somebody asks.
Five groups of features, and not one of them depends on someone remembering the policy.
Rules produced the result, not a model
The scoring behind a recommendation runs against the policy your organisation set.
- Rule-based scoring, with your sourcing policy as the input
- The threshold above which something must go to market, and the minimum bids
- Which supplier states block an invitation, and how an off-list supplier is handled
- The model writes the narrative explaining the result, not the result
- On the invoice side there is no model in the check at all
Written where it applies, by the people accountable
Set per organisation, in the product rather than in a document.
- The sourcing threshold, the minimum bid count and the award mode
- How a supplier off the approved list is treated
- Which supplier states stop a contract, purchase order or invitation
- Approval rules by amount, category and emergency, with named approver roles
- Eighteen contract actions and the full set of auction permissions, assigned to roles
The limits that hold even when you have said proceed
Guardrails sit under an automatic award whether or not anybody remembered them.
- Not a sole source, and not a supplier new to you
- Not a first order with that supplier, and not one with gaps in the data
- Not above the value threshold you set
- An invoice clears only on all eight conditions
- Sourcing autopilot drafts only, is off until switched on, and works inside a ceiling
Named places, each checkable afterwards
Not a claim about everything. A list of the records that carry it.
- An intake record: person, assistant or system, with the reasoning behind it
- And the link to the override, on that same record
- A written reason on an auction AI override, where that rule is switched on
- Before and after values on invoices, contracts, policy changes and supplier restrictions
- The same on certificate compliance and catalogue fields
A permission somebody holds
The ability to disagree belongs to named roles rather than to whoever is in the screen.
- Overriding an auction AI recommendation is assigned to roles on its own
- Separately from who may create, monitor or award the event
- That override can be set to require a written reason
- The reason is recorded with the override
- A recommendation left advisory waits for the person who holds the decision
Where governed autonomy starts
Five things. True before anyone configures anything.
- Sourcing events and auctions drafted by the assistant are created as drafts.
- Nobody changes their own role.
- The last platform administrator cannot be demoted.
- A supplier user never reviews supplier onboarding.
- A supplier’s bank and tax changes are proposed by the supplier and approved on the buyer side.
Three questions a review asks about governed autonomy.
“Why did it recommend that supplier?”
Because of the rules the organisation wrote. The scoring behind a recommendation is rule-based, run against the sourcing policy — the threshold, the minimum bid count, how a supplier off the approved list is treated and which supplier states block an invitation. The narrative that comes with the recommendation is written by the model over that result, so what you are reading is an explanation of the rules, not a separate opinion that happens to agree with them.
“What stops it awarding something it should not?”
Five guardrails that hold whether or not the organisation has set awards to proceed automatically: not a sole source, not a supplier new to you, not a first order with that supplier, no gaps in the data, and under the value threshold you set. Anything that trips one of them waits for a person. Whether awards proceed at all without a person is itself a setting on the sourcing policy, and advisory is how it arrives.
“Can somebody overrule it, and will anyone know?”
Overriding an auction AI recommendation is its own permission, held separately from creating, monitoring or awarding the event, so the ability to disagree belongs to a named set of roles rather than to whoever happens to be in the screen. An organisation can set that override to require a written reason, and that reason is recorded against the override.
Platform fit
The policy belongs to the organisation. So does the decision.
Written where it applies
The sourcing policy, the approval rules, the supplier states and the invoice thresholds are set per organisation, by its own administrators, rather than handed down.
A group writes each company’s separately
Governed autonomy is per organisation. In a group, each company sets the policy its own recommendations run on, rather than inheriting one central template it has to be an exception to.
Credentials for connected systems
Integration and email credentials are stored encrypted.
Sign-in and sessions
Email and password followed by a six-digit code sent to that person’s inbox, on by default in the product. A session lasts twelve hours at the outside, is re-checked every five minutes, and five wrong passwords lock that account for fifteen minutes.
Connects to How governed autonomy works · Autonomous agents · Intake & orchestration
Bring the approval matrix you run today.
The spreadsheet, however unwieldy. We will build it as rules in front of you, and open the screen that says who holds each permission.
Security, compliance and recognition
Checked by auditors. Judged by the industry.
Where each compliance programme stands, and what the procurement industry has given us so far.
Security and compliance
Awards
Five, presented on stage at three industry events.
- Autonomous Procurement Platform of the Year
- Procurement Technology Innovation Award
- Unified Procurement and Supplier Management Platform
- Procurement Startup of the Year
- AI in Procurement or Supply Chain Award — Solution Provider
Before you book anything. Answered honestly.
What is AI governance for procurement?
It is the difference between a system that produces an answer and one that produces an answer you can defend. In Proconomy, what a recommendation favours is produced by rule-based scoring against the policy your organisation set — the sourcing threshold, the minimum bid count, how an off-approved-list supplier is treated, which supplier states block an invitation. The model writes the narrative that explains that result rather than producing it. Whether anything proceeds on a recommendation without a person is a setting you hold, with guardrails underneath it either way.
What does governed autonomy mean?
People set the rules, the system moves the work forward inside them, and anything that falls outside them comes back to a person. In practice that means the policy your organisation wrote — the sourcing threshold, the minimum bid count, how an off-approved-list supplier is treated, the approval rules that apply when it comes back — is what produces a recommendation, rather than a model forming a view that a rule checks afterwards. Guardrails sit underneath it, and the ability to override belongs to named roles.
Does the model decide which supplier wins?
No. The scoring behind an award recommendation is rule-based, run against your sourcing policy. What the model does is write the narrative that explains the result. On the invoice side there is no model in the decision at all — automatic approval is a check of eight stated conditions.
What stops it acting on its own when it should not?
Five guardrails, which hold whether or not an organisation has set awards to proceed automatically: the source is not sole-source, the supplier is not new to you, it is not the first order with that supplier, the data has no gaps, and the value is under the threshold you set. Anything that trips one of them waits for a person. Sourcing autopilot is separate again: it drafts only, it is off until an administrator switches it on, and it works inside a value ceiling.
Can we see why something happened?
On an intake record, yes, and that is the fullest case: the record holds whether the actor was a person, the assistant or the system, the reasoning behind what happened, and the link to the override. Elsewhere, before-and-after values are kept on invoices, contracts, policy changes, supplier restrictions, certificate compliance and catalogue fields, and a recorded reason sits on a policy change, a supplier restriction and an auction AI override.
Can a person overrule a recommendation?
On an auction, yes, and the ability to do it is its own permission — assigned to roles separately from who may create, monitor or award the event. An organisation can set that override to require a written reason, which is then recorded with the override. A recommendation left advisory, which is how it arrives, simply waits for the person who holds the decision.
Who writes the policy the decisions run on?
Your administrators, per organisation. The sourcing policy, the approval rules, the supplier states that block work, the invoice thresholds, eighteen contract actions and the auction permissions are all configured inside the product by the people who run that organisation.
Your rules. In the settings.
No slides. We build one of your approval rules in the product, then open the permissions behind it and the invoice thresholds beside it.
Someone from client success replies, not a sales sequence. If we are not a fit we will say so on the first call.
Not ready to talk to anyone?
Fair enough. Both of these work without giving us your email.
Explore the next step