Everything checked. Every finding cited. Every decision kept.
Give it your documents and the questions you have to answer. It reads all of them, shows you the passage behind every finding, and leaves the deciding to a named person.
Two jobs in regulated firms have the same shape — read a large pile of documents against a set of questions, and be able to show afterwards what you found and who decided. A new regulatory publication against your policies and controls. Your review checklist against every client file. Both are done by hand, on a sample, by people whose time is the constraint.
This is being built, and it is not finished. Everything below describes how it is designed to work, not what it does today. It is on this page because the firms who would use it should shape it before it is finished rather than after. If the problem is one you have, a conversation now is worth more to both of us than a demo later — and if it turns out you do not have it, that is the most useful thing you could tell us.
How it works
Five stages, and who does what at each. You set it up and you decide; the reading and the citing happen in between. The order is the design.
Set it up, once
YouLoad the filing cabinetYour policies, procedures, controls, customer letters — or client files. Each one is cut into paragraph-sized pieces, and every piece gets a permanent label so it can be pointed back to later.Load the checklistThe questions to ask. For a publication: what does this affect, what has to change, how serious is it. For a file: your own review checklist, word for word. It is held as data rather than written into the product — swap the checklist and the same engine does a different job.A subject turns up
Something arrivesA publication, or a fileA new regulator publication, or the next client file due for review. That is the subject: the thing everything else gets checked against.It does the reading
StratavueFind the likely places firstRather than re-reading five thousand documents for every question, it searches for the few hundred most relevant pieces. A dial, not a limit: as the cost of reading falls you turn it up, until it reads everything.Reason over each one, with the quoteFor each document it found, it asks the checklist question with the subject and that document’s paragraphs in front of it. Back comes what is affected, why, what should change — and a quote from the exact paragraph it relied on.Throw away what it cannot proveA finding whose quote does not actually appear in a real paragraph is dropped, and the number dropped is reported rather than hidden. It is never allowed to conclude that something is fine on its own authority — only that this passage is affected, and here is why.You decide
YouAccept, reject or amend — with a reasonA compliance lead or reviewer works the findings. Nothing is decided for you; the product records the decision, it does not make it.The decision is kept, permanentlyWritten down with a name and a date, and never edited or deleted afterwards. Every accept and reject records what a real reviewer at a real firm considered right — which is the part somebody with the same models cannot copy.It goes somewhere useful
StratavueInto the work, or into oversightAccepted findings become a change record in the Stratavue pipeline you already use, or roll up by member firm and adviser. A finding nobody can act on was not worth making.
Throughout: every run reports what it cost, and what the same run would have cost reading everything. That is how you know when to turn the dial up.
What using it is designed to feel like
Before it can check anything, someone has to confirm how your documents relate to each other. That confirmation is the first real screen, so it is designed as review work: one specific claim per screen, the evidence for it always visible, and a keyboard. Never “is this map right?” — always “does this document implement this rule, as this sentence suggests?”
The order is the point. It does not ask you about five thousand documents. It queues the rules that the most documents reference first, because confirming one of those settles several documents at once — then the policies linked to nothing, then the rules nothing implements, then owners. It can be paused, and the map is usable half-confirmed with unconfirmed links visibly marked as such. Rejecting a proposal suppresses it until the document itself changes, so you never see the same wrong guess twice. The first sitting is designed to be about forty minutes.
Then it stops being a queue and becomes an inbox. A policy is edited in your document store and one item appears: “v3.3 no longer references this rule — remove the link?” A new publication lands: “this introduces a rule nothing implements — is that right, or have we missed a policy?” An owner changes in the metadata: “confirm?” Two or three items a week, each one key, each recorded with a name and a date.
And not a diagram. A node-and-edge picture is the thing everyone builds and nobody uses to confirm anything — you cannot confirm three hundred links by dragging. The map exists for reading and for export; the confirming happens in the queue, because somebody checking claims about their own firm is doing review work, and review work is a queue with the evidence in front of you.
Two ways it gets used
What a pilot involves
A conversation, then a sample you choose and hand over deliberately — no connection to anything, no access to your systems. You see the output on your own material and tell us where it is wrong. There is no charge and no commitment at the end of it.
The thing worth agreeing up front is what would make it a failure. For regulatory impact that is usually a rule it should have flagged and did not. For file review it is a question where it and your reviewers disagree and your reviewers are right. Both are more useful than a demonstration that goes well.
What it is not
Where it stands
Stratavue is built and hosted in the UK by a financial services change specialist with fifteen years running transformation inside regulated firms. The existing product holds operating models and the change running against them; Assurance is the same idea pointed at documents and obligations.
How a pilot handles your data — where it is processed, what is sent where, what is retained — is a conversation to have properly rather than a badge to put on a web page, and it is one of the first things a pilot has to settle to your satisfaction.
The rest of Stratavue is here: what it does today.
Tell us whether this is your problem
Including — usefully — if it is not. Say which of the two it is closer to and roughly what you do about it today.