Using AI in Work Investigations: What UK Employers Need to Know 

The greater focus on culture and governance in the workplace in recent years, along with enhanced regulatory scrutiny around whistleblowing, has led to a sharp increase in investigations into employee complaints and conduct. 

At the same time, artificial intelligence has started to infiltrate the world of work, due to its ability to increase productivity. 

Employers are therefore naturally starting to deploy AI and automation when conducting workplace investigations, capitalizing on their ability to streamline administrative tasks and provide decision-making support. But, as with any use of AI, the potential benefits also carry legal risk, which is amplified in the context of a workplace investigation, as it is often a precursor to employment claims. The key ways in which AI can be used to assist with a workplace investigation are as follows. 

Automated Notetaking 

Documenting interviews is central to the evidence-gathering process in an investigation. While investigation meetings will typically be attended by a notetaker, it is not a legal requirement to have a notetaker present, and increasingly, employers are using automated notetaking technology for these meetings. The potential benefits of this include: 

• Freeing up resources and administrative time, as only one individual needs to attend the meeting, and the note is automatically transcribed. 
• The note can be distributed automatically, removing the need for the ancillary process of agreeing to the content of the note and inviting the interviewee’s comments. This can be very time-consuming, particularly on large investigations involving numerous interviews. 
• An automated note is perceived as more objective than a human notetaker, lowering the level of antagonism in the investigation process. As a “verbatim” note, it is likely to reduce the likelihood of discussion and/or disputes over the content or allegations of bias by the notetaker. It should also remove the need to deal with employee requests to record the meeting. 

However, it is important to bear in mind the following: 

• Automated notetakers may misinterpret speech, particularly if a speaker has a strong accent, there is background noise, or technical jargon has been used. 
• Depending on the quality of the AI technology, the automated note will always require a degree of human review to amend any errors. In addition to spelling errors, fundamental details such as names are often mistranscribed, which significantly reduces the use of the note as an accurate record. 
• As with all workplace recording, employers will need to establish a legal basis for the processing of personal data. Although it might be possible to argue that the processing is “necessary” for the performance of the employment contract, employers may need to consider and agree to objections from those being recorded. The usual considerations around data protection and confidentiality will also apply to any transcription. 

Document Review 

Workplace investigations tend to be document-heavy, requiring an employer to collate, review, and manage a large volume of documents in a short timeframe. AI and automation technologies offer unparalleled efficiency in being able to sift through large volumes of data, including: 

• Reviewing documents for extracting information including keywords, date ranges, or any combination of factors, to identify relevant data quickly. 
• On large investigations, an employer may want to conduct data and trend analysis, as a way of investigating particular themes or issues (although employers will need to be aware that this could infringe the data protection law concept of “data minimization,” and they will need to be prepared for objections and challenges along those lines — as far as possible, they should seek to process the minimum amount of personal data required for the purpose of conducting the investigation). 

However, using AI to locate relevant documents has inherent limitations. We have found that when an investigation is complex and involves multiple different issues, a review platform tends to be less successful at identifying relevant documents automatically through trends or in a nuanced way. Employers should, therefore, consider how and when review platforms can be used effectively at the outset of an investigation, in conjunction with the terms of reference, and should adapt their approach as there will not be a one-size-fits-all approach. 

Employers should also avoid processing special category data (for example, personal data revealing racial origins, political, religious or philosophical belief, health, etc.) when using AI, unless they can show that it is strictly necessary for the purpose of the investigation. 

Generative AI 

Increasingly, employers have their own in-house generative AI (GenAI) tools. 

While all investigations will be fact-specific, GenAI can be used effectively to provide template documents (such as invitations to investigatory meetings) or to summarize information, such as condensing an employee’s grievance within the investigation report, or to assist with formulating the terms of reference. 

However, employers should also note that: 

• Personal data and confidential information should never be entered into a publicly available GenAI tool, as this could lead to possible exposure of confidential or proprietary information into the public domain. 
• Any organization that uses (or proposes to allow the use of) GenAI tools should have an appropriate policy in place for staff, so that they are not misused. 
• While AI can be used to assist with the investigative process, it cannot substitute for human decision-making when reaching an outcome. The investigator/chair must be able to show that they have considered the evidence and reached a clear and objective outcome based on their assessment of the facts, so that they can defend their decision-making process, if challenged. 
• Due to the societal and cultural factors embedded within GenAI’s training data, the systems are likely to be inherently biased. Users should therefore be aware that an AI model may create output containing biased language or reflecting societal stereotypes. 
• GenAI is known at times to produce inaccurate or even made-up information. This is also why it is crucial to ensure that there is a process of human review and intervention. 

With no sign of investigations slowing down in the workplace, AI is undeniably a valuable tool for managing investigations efficiently and avoiding unnecessary duplication. However, as with all AI and automation, it needs to be deployed carefully and human oversight remains critical, to ensure fair, accurate and robust outcomes. Employers should also be mindful of their data protection obligations, and recognize that employees have a general right to know that AI is being used to pursue or inform an investigation. 

Emily Knight and Daniel Gray are attorneys with Mishcon de Reya, a member firm of IR Global, in London. © 2025 Mishcon de Reya. All rights reserved. Reposted with permission of Lexology.