Xalios
Book a discovery call
Solutions·HR

Your HR team answers for policy, leave and benefits.
Here is AI it can answer for.

Policy, leave and benefits questions get answers from your own documents with a source for each one, and conflicting policy versions go to a person.

The work the HR team answers for

Staff ask how much leave they have left, what a policy says about their situation and whether a benefit covers them. Leave and benefits requests follow. They arrive by email, in Teams or Slack, as files, or on a schedule.

The HR operations lead answers for what staff are told. That person is named as responsible for the agent, and the HR team works its review queue.

Example questions

“How many days of leave do I have left?”Teams
“Can I come back from parental leave part-time?”Email
“Does the dental plan cover my partner?”Slack
“I’d like to book the last week of August off.”Teams
“Can you send me the salary bands for my team?”Email

For example only. The last one draws on a document most staff may not see.

Answers you can stand behind

What the assistant answers, and when it refuses

The assistant answers from your HR policies and handbooks in SharePoint, OneDrive or Google Drive, and cites every claim. It follows the permissions on each document, so an answer draws only on sources the person asking may see.

When it can’t confirm the asker may see a source, or can’t back an answer, it refuses and says why. When two versions of a policy disagree, it sends the question to a person to decide which one applies.

Answers →
It answers

For example, “You have nine days of leave left this year.” The leave record and the leave policy are cited.

It refuses

For example, “I can’t confirm you may see the salary bands, so I can’t answer from them.”

What the agent does

The agent handles the routine follow-up. It drafts the leave booking, sends the benefits form and replies to the person who asked. A request with several steps, such as a return from parental leave, is tracked as a case until it closes.

It holds a role in the HR team, with the HR operations lead named as responsible. It can carry its own Microsoft Entra identity. You set it up from a plain-language job description and plain-English rules.

What it can see and change

HR policies and handbooksRead
Leave balancesRead
Leave requestsWrite
Replies and case notesWrite

Your administrator grants each connection in your own systems. The agent reaches only what was granted, across about 200 business apps.

Integrations →

How autonomy moves at your volume

Autonomy is set per task by your team’s run of approvals. Approvals with edits count as neutral and 5% of work is spot-checked. Two mistakes in its last 20 and a task steps back. The steps and timings below are the defaults. Your organisation can set its own steps when it sets Xalios up.

LEVEL 1Every item approvedEach reply and booking waits in the review queue for a person to approve, edit or decline. Every task starts here.
LEVEL 230 in a rowAfter 30 approvals in a row, the task does more of its work alone. About three days at 300 items a month, about a month at 30.
LEVEL 3150 in a rowAbout two weeks at 300 items a month, about five months at 30.
LEVEL 4750 in a rowAbout two and a half months at 300 items a month, about 25 months at 30.

The timings are for one request type. Leave balance questions and benefits questions each move on their own count.

Autonomy →

What never runs alone

People decide. The agent proposes until your team’s approvals say otherwise, and some work stays with a person whatever the count.

Conflicting policy versions

When two versions of a policy disagree, the question goes to a person to decide which one applies.

Actions you mark as serious

For example, leave outside policy or a change to someone’s benefits. You choose the list, and those actions wait for a person at every level.

For staff and works councils

Approvers can be pseudonymised in the record. We bring a works-council pack, and we make no promise of fewer colleagues.

On the record

An example record. One question from a member of staff, traced from what they asked to the person who decided.

Example · return from parental leaveSent to a person
ASKED“Can I come back from parental leave on three days a week, and how much notice do I give?”
FOUNDTwo versions of the flexible working policy that give different notice periods.
CHECKThe asker is a member of staff, so may see both versions.
DIDAnswered the leave dates from the leave policy, with its source. Sent the notice question to a person with both versions attached.
DECIDED BYThe HR operations lead, who confirmed which version applies.

Every step sits in one record with a tamper-evident history: what was asked, the sources, the check that let it run and who decided. The answer can be traced end to end months later.

The record →
Paid 30-day pilot

What the pilot measures

A paid 30-day pilot on the HR and policy inbox. It starts with every item approved by a person and includes a seeded test of restricted documents.

Book a discovery call→
SOURCEDThe share of questions answered with a source.
REFUSALSHow often it refused when it should have.
LEAKSRestricted documents shown to the wrong person in the seeded test. The target is zero.
STARTEvery item approved by a person in the review queue.
PACKA pre-filled data protection impact assessment, a SOC 2 audit date and a works-council pack.
Next teamYour team answers for what partners and customers hear.An embedded assistant that answers with sources and shows each partner tier only its own documents.Partner and customer service →

Bring the policy questions your team answered last month. We’ll start with the ones that come up most.