your people know what is right, pyxta captures it, and AI puts it to work
In the example on our home page, the first answer had an incorrect line: an installation fee spread over twelve months, like a service plan. The finance lead spotted it at once. Installation is a one-off job, and this one was finished in March.
That is how it is meant to work. A newly hired analyst gets things wrong too, until someone shows them how the business does it. The difference is that Pyxta does not need telling twice: in April, two more installation fees were counted in full without her.
what your people do
Day to day, they check the answers in their reports and dashboards the way they would check a colleague's work. They confirm what is right and flag what is wrong, and Pyxta proposes the fix. Each figure opens to the lines behind it. No rules to write, no formulas to read.
That checking is also how Pyxta learns. Nobody has to write down how they work an answer out: Pyxta picks it up from what they confirm and what they correct.
A new answer starts with a request in their own words, such as "revenue earned each month, based on each customer's contract". Pyxta drafts a way to work it out, and they check the first results.
what AI does
It reads the documents and works out the figures, applying what Pyxta has learned from your people. In our example, it read each contract for when the service started and how long it ran.
It also drafts how each new answer should be worked out. Whether a draft is right is never its decision.
what keeps it in check
- your way, every time. Each time AI reads a document or works out a figure, it is given what Pyxta has learned from your people. Where a contract could be read two ways, it applies your reading, not one of its own.
- your team signs off every change. A new way of working out an answer, or a fix to one, is used only once someone on your team has seen its results and approved it. Any fix can be undone.
- every answer shows its working. Any figure opens to the lines it used and, for anything read from a document, what was read and where. Lines that no figure uses go on a separate list, so you can see what was left out. Work March out again months later, on the same records, and you get the same figure: Pyxta re-runs the calculation it stored, instead of asking AI again.
when something changes
Contracts are renewed, rates go up, a policy is rewritten. Each record is worked out under the terms in force at its date: last year's shifts at last year's rates, this year's at this year's.
In our example, the care plan with three months free was new to Pyxta. It was held out of April's revenue and went on the same list, for the finance lead to decide.
what happens to your systems
Nothing. Pyxta works from the files your systems can already export and the contracts and policies in your folders, or connects to a system directly, read-only unless you ask otherwise.
It is not tied to one AI vendor either. It works with any capable AI model, and what it has learned from your people carries over when the model changes.