Most organisations don’t have a policy for AI-related grievances. They have a general grievance procedure, written before AI played any role in workplace decisions or complaints, and an assumption that it will stretch to cover whatever comes next.
That gap is more urgent than it might look. Acas opened a consultation on 30 July 2026 to update its statutory Code of Practice, and it explicitly asks the public whether AI-related issues in disciplinary and grievance processes need addressing at all. The body that sets the national standard for handling grievances hasn’t decided yet. Waiting for that answer before building a policy of your own means waiting on a timeline nobody controls.
This piece covers what an actual AI grievance policy needs to contain, covering both the decision-origin and AI-assisted grievances already established in this series, how to build human review into the process rather than reconstructing it afterward, and how to roll a policy out so it actually changes what happens on the ground.
Explore: AI Grievances and Employment Tribunals: A Guide for HR Teams
Why a General Grievance Policy Isn’t Enough
A general grievance procedure is built around a single assumption: a human made the decision being complained about, and a human can be asked to explain it. That assumption breaks down twice over with AI:
- Decision-origin grievances (where an algorithm helped produce the outcome): the gap is ownership and documentation. There’s often no one assigned to explain what a system actually did, and no record of the human review that happened at the time, if it happened at all. ICO’s Recruitment Rewired findings confirm this isn’t rare: many employers believe a human is meaningfully involved when, in practice, the system made the decision alone.
- AI-assisted grievances (where the complaint itself was drafted or built with AI): the gap is readiness. Most HR teams have had no training on how these differ from an ordinary complaint, even though Irwin Mitchell’s survey found over half report they’re harder to resolve.
Neither gap gets closed by a policy that doesn’t name either problem.
Explore: Anatomy of an AI-Related Grievance
Explore: Tribunal Rulings, Regulatory Action and Claim Trends on AI in the Workplace
What a Real AI Grievance Policy Actually Needs to Cover
A policy that just says “AI-related grievances will be handled fairly” doesn’t actually change anything. Four specific elements do:
- Assigned ownership. A named role or function responsible for explaining any AI-influenced decision, not whoever happens to be free when a grievance lands.
- A documented human review point. Captured at the time the original decision is made, not written up afterward to justify it. ICO’s own standard requires this to be genuinely meaningful, not a rubber stamp, more on exactly what that means below.
- Transparency requirements. Employees should be told when and how AI played a role in a decision affecting them, before they have to ask.
- A dedicated escalation path. AI-influenced disputes shouldn’t get stuck in a generic process that was never built to handle them.
None of this needs to be complicated. It needs to be specific enough that someone can actually follow it when a grievance arrives, rather than improvising.
Building the Human Review Point Into the Process, Not After It
This is the single most repeated failure across everything we’ve covered in this series: human review reconstructed after a complaint, rather than documented at the time the decision was made.
ICO’s own standard is specific about what “meaningful” review actually requires. It has to involve someone with genuine authority, discretion, and competence to change the outcome, not a person glancing at an AI-generated recommendation and approving it by default. A human who technically looked at the decision but couldn’t realistically have overturned it doesn’t meet that bar, even if a box was ticked somewhere.
The practical fix is to build the record into the decision workflow itself, not to reconstruct it later. Whoever makes or approves an AI-influenced decision should note, at that moment, what they reviewed and why they agreed with it. That single habit is what turns “we’re confident a human was involved” into something an employer can actually demonstrate if a grievance or a tribunal claim asks the question.
Handling AI-Assisted Grievances Within the Same Policy
An AI-assisted grievance doesn’t need its own separate track. It needs the same investigation standard applied with a bit more discipline in how it’s read.
Kingsley Napley’s guidance is useful here: start with an initial meeting to scope out what’s actually being raised before launching a full investigation. Where a grievance bundles multiple allegations together, as AI-drafted ones often do, categorise them separately rather than treating the document as one undifferentiated complaint. Neither step requires new policy language, just a consistent first move.
What the policy shouldn’t do is let length or formality change the standard. The Acas Code does not permit dismissing or truncating a grievance because it’s poorly evidenced or unusually broad. Every allegation that’s reasonably capable of investigation still has to be investigated, whether it arrived in three sentences or three pages.
Explore: Anatomy of an AI-Related Grievance
Rolling the Policy Out: Training and Communication
A policy that exists only in a document nobody’s read doesn’t change what happens when a grievance actually lands.
The cost case for getting this right is stark. Acas’s own research found:
- Formal disciplinary and grievance procedures cost UK employers an estimated £2.36 billion a year
- Informal resolution costs roughly £250 million, about a tenth as much
- 44% of working-age adults in Great Britain reported experiencing conflict at work in the past year
Training that helps a grievance get resolved earlier, or handled correctly the first time, isn’t a soft investment, it’s the cheaper path by a wide margin.
Acas’s own draft update to the Code reflects this, encouraging employers to provide training as good practice specifically to build the skills and confidence needed for early, effective resolution. For AI-related grievances specifically, that training needs to cover two distinct things: how to explain an AI-influenced decision when asked, and how to read past an AI-assisted grievance’s formatting to find the actual complaint underneath.
None of this needs to be elaborate. It needs every manager who might receive a grievance to know the policy exists and what it specifically asks of them.
How Avado Can Help
Building a policy is only half the work. Making sure managers can actually apply it, recognising when a decision needs a documented human review, knowing how to read past an AI-assisted grievance’s formatting, is what determines whether the policy holds up when it’s tested. Avado’s HR Compliance for Managers course, presented by employment law specialist Amanda Chadwick, builds exactly that judgement, covering disciplinary and grievance essentials alongside the wider legal landscape managers now operate in.
Explore HR Compliance for Managers and make sure your policy holds up the first time it’s actually tested!
