Review season arrives, and someone spends a fortnight pulling numbers out of the operational systems by hand, into a spreadsheet, to answer a question the business already has all the data for. How did Sarah do against her targets?
Your HR system knows what Sarah agreed to. Her role, her KPIs, the weightings, the bonus structure, the review cycle. It has no idea what she actually did. That lives in the property system, or the sales system, or the job system. The two have never been joined without a person and a spreadsheet, and that is why performance reviews are dreaded by the people running them as much as by the people receiving them.
This post is about what changes when they are joined. It is about Employment Hero reporting that reaches into the systems where the work happens, and about the access model that makes it acceptable to connect AI to the most sensitive data in the business. We will start with the second, because it is the part people worry about first.
“We connected AI to your HR system” is a sentence that makes people tense before they hear the second half. It should. HR data is the most sensitive category in any business: pay, performance, health, grievances. A careless approach here does real damage, and we would rather be visibly the careful ones than the fast ones.
So before any capability, the model.
People see what their role lets them see. A manager sees their team. An employee sees themselves. HR sees the organisation. The connection carries the person’s Employment Hero permissions, and nothing about asking in plain English changes them.
Every question is logged. Who asked, what was asked, what came back. When someone in HR wants to know how AI has touched employee data this quarter, the answer is a report, not a shrug.
Nothing changes without someone approving it. Reviews are drafted and scored. A person signs them off.
Proposed changes are logged as proposals, not applied. If the review conversation concludes that Sarah’s KPIs should change next period, that is recorded as a suggestion awaiting sign-off. The agreed KPIs stay as they are until someone approves the revision. This is exactly the behaviour an HR manager needs before letting anything near the process, and we built it in because they told us so.
Handled properly, the nervousness is the opportunity. This is the vertical where being careful matters most, and where a careless competitor will make a mess.
We connect Claude, Anthropic’s AI model, to your Employment Hero environment through an open standard called the Model Context Protocol. You ask in plain English, in Microsoft Teams or the Claude app, and Claude answers from the live system as you. With your approval it drafts, schedules and records.
It reads the live system, not a copy. It acts as the person asking. It runs where people already are. The managed AI service around it handles the connector, the access model, the logging and the tuning against your data.
Employment Hero has one of the largest SME install bases in Australia for HR, payroll and performance. The reader we have in mind is the business owner, the operations manager, or whoever ended up owning HR in a company of 20 to 200 people. Usually that person did not ask for the job.
This is the demonstration we would build the entire pitch around if we could only pick one.
A performance review needs two things: what someone was supposed to achieve, and what they actually achieved. Those live in different systems and always have.
“Run Sarah’s half-year review.”
Here is what happens.
First, the HR system is asked what Sarah agreed to. Her KPIs, each with a target and a weighting, and, critically, a note of which system holds the evidence for each one. Arrears and vacancy for her portfolio live in the property management system. Settlements and conversion live in the development sales system.
Second, each of those systems is asked for the actual number. Not a figure someone typed into a spreadsheet in March. The live number, from the system that owns it, as at today.
Third, the review is scored against the agreed rubric and weightings, with every number traceable back to the system it came from.
Fourth, the meeting is scheduled, the invite issued, and the review questions set.
Fifth, after the conversation, the notes go back onto the review, and any proposed changes to next period’s KPIs or to her job description are logged for sign-off.
A fortnight of spreadsheet work, in a conversation. The part that makes it defensible in the room is the traceability: when Sarah asks where the arrears number came from, the answer is a link, not a memory.
The detail that makes this work rather than being a party trick is that the KPIs themselves name where their evidence lives. That means the process cannot quietly score someone against a number nobody can find again. If you are designing KPIs for next year, that one idea is worth more than the rest of this post.
“Who is in the Commercial PM team?”
Filterable by department, or by whether they sit on a bonus structure. Only people on a bonus carry agreed KPIs, which is a useful filter in itself.
“Tell me about Sarah: role, reporting line, bonus structure.”
“What is the current job description on file for her?”
“What are Sarah’s agreed KPIs?”
Each with its target, its weighting, and which system holds the actual number.
“How is Sarah tracking against her targets this half?”
“Score this review against the actuals.”
“What review cycles are open?”
“Where is Sarah’s review up to?”
The questions asked, the meeting, the notes recorded, the score.
“Who in the Developments team has not had their review meeting yet?”
That is the one an operations manager asks in week three of review season, and it currently means opening every record.
“Open the FY26 H1 cycle for everyone on a bonus structure.”
A cycle, and a review for each person named. If any name cannot be placed, nothing is created. You fix it and run the whole batch again. No half-finished cycles where eleven reviews exist and three do not and nobody is sure which.
“Set the review questions for this cycle.”
“Book Sarah’s review for 2 October at 10am and send the invite.”
“Remind me a week before each of these.”
“Record the notes from that conversation onto her review.”
“Log a proposed change to her KPIs for next period.”
That last one is deliberately a suggestion log, not a change. We keep saying this because it is the point. Automating the paperwork is easy and a hundred tools do it. Automating the paperwork while keeping every decision with a person is what makes an HR manager willing to switch it on.
The review is the flagship, but it is one example of a pattern. The HR system sits next to everything, and almost every valuable question involves both.
A new starter exists in Employment Hero days before they exist anywhere else. Everything that has to happen next, the account, the licence, the groups, the mailbox, the induction, is manual, repetitive, and occasionally forgotten.
“Priya starts Monday. Set her up: account, licence, the right groups and distribution lists, a mailbox, and book her induction.”
“Marcus finishes Friday. Revoke his access, convert the mailbox, forward his mail to his manager, and list anything still assigned to him.”
“Who has started in the last month who is missing a licence or is not in the right groups?”
We are a managed IT provider, so we will admit an interest here: this is HR data driving exactly the IT work we already do for clients every week. It is also the offboarding gap that turns into a security incident about once a year at every company we have ever audited. The person who left in March whose account is still active in September. Having the HR system be the trigger, rather than an email someone forgot to send, closes that gap for good.
“Does the payroll journal in Xero match the bonus we calculated from the KPI numbers?”
“Who has had a pay change approved in HR that has not reached payroll?”
“What did we actually spend on wages by department this quarter versus budget?”
We covered the Xero side of this in our post on a Xero AI assistant for cross-system questions.
Beyond the review cycle itself:
“Who is on leave next week, and what jobs are scheduled to them?”
“Which team has the highest overtime, and what were they working on?”
The first of those, for a contractor, is the difference between finding out on Monday morning and finding out the Thursday before.
“File the signed review against Sarah’s record and schedule the follow-up.”
“Set up the new starter’s first-week induction meetings with the right people.”
Employment Hero alongside Microsoft 365 (accounts, email, Teams, SharePoint), Xero, and the property and development sales systems in the review demonstration all run today. If the system that holds your KPI evidence is something else, it is a build rather than a blocker, and the KPI-names-its-source pattern is what makes it a short one.
Connecting an AI model to Employment Hero is not the hard part. The hard part is that HR data has places where it quietly misleads, and the consequences of a wrong answer here are worse than in most systems. A job title updated but a reporting line not. A KPI target entered as a percentage in one record and a number in another. A terminated employee whose record was closed but whose review was left open.
Point a generic AI at that and ask how Sarah is tracking, and you will get a confident, specific, wrong score. In a performance review, a confident wrong score is not an inconvenience. It is a grievance.
Making KPI reporting across systems trustworthy means measuring the data against itself, writing down where it misleads, and building those traps into how the review is scored. When ours scores a KPI, it shows the source system, the query it ran and the date. When a target and an actual are in different units, it stops and asks. That is what an integration specialist is for, and it is the part no self-serve connector provides, because it does not know your business or your people.
Businesses of 20 to 200 people running Employment Hero, with staff on bonus structures, KPIs that reference operational numbers, and a review season that currently means spreadsheets. Property agencies, development sales teams, contractors, professional services, anywhere targets are set in one system and achieved in another.
We are a Perth managed IT provider and we deliver this hands-on, so WA businesses are who we work with most. Employment Hero is national, and so is the connector. If you are interstate and this is your review season, talk to us and we will say plainly whether we can deliver it well.
Look at your KPIs and ask where the evidence lives. For each agreed KPI, can you name the system and the report that produces the actual number? If the answer is “a spreadsheet Dave maintains”, that KPI cannot be scored defensibly, with or without AI. Fixing that is worth doing regardless.
Audit who can see what in Employment Hero. Check permissions by role. Any AI connection inherits them, and HR data is where a loose setup does the most harm.
Watch a review run end to end. We have a working demonstration that runs the whole cycle across three systems, from agreed KPIs to scored review to scheduled meeting. If you are drowning in review-season spreadsheets, watching it once is the entire sales pitch. Twenty minutes, no commitment. Contact us on 1300 EPIC IT or through the contact page.
The paperwork, yes, and many tools do that. The harder part is fetching the actual performance numbers from the operational systems and scoring against agreed KPIs, which is what an Employment Hero AI assistant connected to those systems does. The review is drafted and scored automatically; a person conducts the conversation and signs it off.
Each KPI needs a target, a weighting, and a named source system for the actual figure. The assistant pulls the live number from that system, scores it against the target, applies the weighting, and shows the source for every figure. If a KPI has no named source, it cannot be scored defensibly and you should fix that first.
It depends on how the connection is built. Ours acts as the person asking, so a manager sees only their team. Every question is logged with who asked and what came back. Nothing is changed without approval, and proposed KPI or job description changes are recorded as suggestions awaiting sign-off, not applied.
Yes. A new starter in Employment Hero can trigger the account, licence, groups, mailbox and induction booking in Microsoft 365, prepared for approval. A leaver can trigger access revocation, mailbox conversion and mail forwarding, with a list of anything still assigned to them. That closes the offboarding gap that most often becomes a security incident.
It means answering questions that need Employment Hero and another system at once: how someone tracked against targets held in HR using actuals from the operational system, whether the payroll journal matches approved pay changes, or who is on leave next week and what is scheduled to them. No single system can answer those alone.
Perth and WA are where we deliver most of our work, because we set this up hands-on. Employment Hero is used nationally and the connector works anywhere. If you are interstate, contact us and we will tell you plainly whether we can deliver it well for your business.
