Products
ALEX Workers Order-to-Cash How it works Trust Company Start with one workflow
The operating layer for company AI

AI that takes responsibility for getting consequential work finished.

VERIDOM turns fragmented organizational knowledge into governed workflows, then coordinates people, systems and AI until every case reaches an evidence-backed conclusion.

Watch a case reach completion → Start with one workflow
What you buyA completed case
What is accountableThe decision
What you keepThe operating model
Default behaviourRespect the boundary
How work endsEvidence recorded
01 — Who this is for

She is not checking whether the AI worked overnight.

She is checking what changed. That is a different question, and almost no system currently answers it.

Claire Bramwell Finance Director
UK manufacturing group · 9 people in the AR function · ~4,000 invoices a month

It is 8:15 on a Tuesday. Claire opens ALEX before her first meeting and reads four lines. A payment matched itself at 06:42. Three accounts are waiting on evidence somebody is already chasing. A customer changed a portal rule, and everything decided under the old one has been found and re-opened.

And a £4,200 write-off has stopped, because the approval authority it was resting on expired yesterday afternoon. Nobody caught that. Nothing had to.

The line she trusts is the one that refused.
Claire Bramwell at her desk before her first meeting, seen from behind, reading her ALEX screen. Four lines: a payment of £12,480 matched automatically at 06:42 against invoice INV-77381 for Northbridge Components; three accounts waiting on evidence that is already being chased, at 07:12; a portal rule change at 07:46 that re-opened twelve cases decided under the old rule; and at 08:03 a £4,200 write-off stopped, because the approval authority expired yesterday.
Matched 06:42

Payment matched · £12,480 It ended. Applied against INV-77381 with no person in the path.

In progress 07:12

3 accounts waiting on evidence Someone owns it. Waiting is a state with a name attached, not silence.

Re-opened 07:46

Portal rule changed · 12 cases The old rule was superseded, so what depended on it was marked stale.

Blocked 08:03

£4,200 write-off stopped It refused. The approval authority had expired, so nothing executed.

02 — The operating failure

Work rarely fails at the moment of action.

It fails between intent and completion. Knowledge is fragmented, authority is unclear, exceptions are undocumented, and nobody persistently owns what happens next.

People close those gaps today because they have history with the company. They know which rule really applies and who genuinely decides. AI does not.

1
Days open
Nothing failed
Nobody was negligent
01 · Missing evidence

The document nobody followed up for.

A case looks active while the one thing required to move it forward remains unowned. Nothing flags it, because nothing failed.

02 · Authority

The approval with no legitimate owner.

Capability exists. Permission does not. And a person being present is not the same fact as a person being authorised.

03 · Decay

The expired rule still governing.

Retrieval finds it perfectly. Reality moved on eleven months ago, and nothing marked what depended on it.

04 · Waiting

The exception everyone assumed someone else was handling.

Most operational failure is unmanaged waiting. The shortage is not intelligence. It is persistent ownership.

Agents don't become trustworthy by knowing more. They become trustworthy when the organisation becomes operable under delegation.

03 — One workflow, made concrete

Watch one case reach completion.

A single invoice shows the whole model. One persistent operator keeps consequential work moving across systems and people until it reaches a legitimate ending.

Ask ALEX a question at any point.

Scroll to advance the case
Example case · Order-to-Cash O2C·4471 Establishing

      Ask ALEX

      An example case, using the states and refusal conditions the Order-to-Cash Worker runs on today.

      04 — The proof

      Watch it refuse to answer a question it cannot support.

      A fluent system finishes this sentence. It has a number, the number reads plausibly, and nobody finds out for eleven months. Here is what happens instead.

      Order-to-Cash · O2C·4471 Example account
      Asked

      What is Meridian Industrial’s current credit limit?

      The current limit for Meridian Industrial is £
      Stopped, mid-sentence
      What is missing
      No current source establishes a credit limit for this account.
      What was found
      £50,000, in Credit & collections policy v3, superseded 14 months ago. It is still searchable. It no longer governs anything.
      Who settles it
      R. Hale, Credit Control. The one person whose authority covers this account at this value.

      It had a number. It had a source. It refused to present a superseded figure as current, and it named who can settle it instead.

      0
      Fabricated claims
      Nothing asserted that no source supported.
      100%
      Traced to source
      Every statement carries where it came from.
      0
      Conflicts collapsed
      No contradiction quietly resolved to sound confident.
      42
      Graded runs
      It stopped every time it should have.

      Measured across 42 graded runs on three independent document sets, scored against answer keys written before the runs. We used two language models roughly thirty times apart in capability: the stronger one produced more unsupportable material, and the checks caught all of it. That is the point of putting the checks outside the model. How this was measured →

      05 — The difference

      Not another agent platform.

      VERIDOM begins where the conversation ends, with the conditions an organisation needs before execution can be accountable. Four of them, and what each one changes on Monday.

      Cases, not conversations

      The work stays open until it reaches a legitimate ending.

      Starting an action is not the same as owning its completion. A queue is not an ending, and neither is a good answer.

      What changes: every open case has a next action, and it ends as completed, resolved or honestly blocked. Missing evidence and approaching deadlines stay actively owned instead of becoming invisible backlog.

      Authority, not capability

      Being able to act is not the same as being allowed to act.

      Routing, authority and permission stay three separate facts, each inspectable on its own. Collapsing them is how a company loses the ability to prove which one actually happened.

      What changes: exceptions arrive ready for a decision. A person receives the issue, the evidence, the authority required and a recommended next step, not another context-hunting task.

      Operational truth, not recall

      Expired knowledge cannot quietly keep governing.

      ALEX keeps the history and withdraws the outdated belief from current use. Search will return both versions. Only one of them decides anything.

      What changes: a policy that has moved stops being quoted back to you as current, and every decision resting on the old version is found by name rather than by search.

      Credit & collections policy — v3 Superseded
      Credit & collections policy — v4 Governing

      Controlled learning, not drift

      Reality can challenge the model without rewriting it in secret.

      A change propagates, whatever depended on it is marked stale, and a person with the authority republishes the truth. The model never quietly edits itself.

      What changes: the case moves with its context attached. Evidence, decisions, approvals, communications and system state travel together across every handoff.

      06 — ALEX

      Ask your company a simple question.

      You will get several answers, all sincere, all from real places. That is not a documentation problem. It is the reason you cannot hand this work to an AI yet.

      What your company says today 4 answers
        What a search tool returns

        No date checked · no authority checked
        What ALEX returns

        Not resolved

        Example questions and sources. One governing answer, where it came from, and anything still contested surfaced rather than averaged into it.

        Something has to know how the work actually runs.

        Before AI can be trusted with consequential work, it needs a current, evidence-backed account of how the organisation genuinely operates, not how the handbook says it does. That is what ALEX builds, and keeps current as the company changes.

        The part that matters is what it refuses. Where two sources genuinely disagree, it will not pick one to sound confident. It tells you, and it says who has to settle it.

        Meet ALEX →
        07 — Where to start

        Start with one consequential workflow.

        Before automating the work, we establish how it actually operates, what counts as done, whose authority applies, and where cases are currently being lost.

        The first operating model

        • The real workflow, and its legitimate endings
        • Authoritative sources, and where they contradict each other
        • Decision owners and the limits of their authority
        • Exceptions, waits, and evidence that is missing
        • Which work is human, which is AI-assisted, which is AI-executable
        • The first governed execution design
        Week 1–2
        Understand
        Map where work stops, what evidence is missing, who holds real authority, and what currently governs.
        Week 3–4
        Authorize
        Agree the boundary: what the Worker may do alone, and what always routes to a named person. You set the line.
        Week 5–8
        Execute
        Live cases in a bounded slice of the ledger, alongside your team, not instead of it.
        Close
        Prove
        One meaningful contradiction, one correct refusal, one completed case, with the evidence behind each. Then you decide.
        Start with one workflow →

        Still deciding which way in? See what Order-to-Cash does → See what ALEX establishes →