The AI Wave: Upcoming AI features in EXPs

My previous article showed some of the AI features that are available within intranet and employee experience platforms (EXPs). This is impressive when you consider that ChatGPT only launched in November 2022, which precipitated the current boom.

Microsoft Copilot developments are likely to dominate the conversation in the next six months, but I’ll first share a few upcoming features from other vendors.

Staffbase Companion search

Staffbase Companion’s search will provide a digital assistant chatbot-esque approach to returning answers. Results can be returned from integrated systems, including the intranet, to answer “how much vacation time do I have left?” and such questions. If someone asked “how do I dispute how many vacation days I have left?” they could be given the link to a policy, as well as a relevant answer formed from that policy. Unlike chatbots that have to be taught the right answers, this AI will learn each time it’s used – drawing only on internally facing content to do so.

Inaccurate responses are still a risk – or, as a previous example put it “the generated answers may be wrong”. The AI also needs access to the right content to generate answers, so having a strong content strategy and governance is still a necessity. However, this could deliver a really nice employee experience, reduce the number of systems they have to access and reduce calls to help desks as employees can self-serve more easily.

The Staffbase Companion search will return results in a chatbot-esque style, but will draw on available internal content to form the response rather than having to be taught.

Interact content context for FAQs and quizzes

In this example from Interact, the AI will take what’s written and generate an FAQ based on the content and anticipated questions. The initial draft can then be further manipulated using AI, and can be manually edited.

I’d advise caution here, as there might be a temptation to create an FAQ about everything because it’s so easy. The point of an FAQ is that there really are questions that are asked frequently, so publishers should consider whether this is the most appropriate way to present information. I know communicators will have different feelings about the appropriateness of the format at all, as some just don’t like using FAQs to share information.

A nice addition is that the same technology can be used to create quiz blocks to test employee’s understanding of what they’ve read. This is a helpful use-case for the organisation (for compliance purposes), the publisher (who might not be skilled at writing simple quiz prompts) and the reader (to help encourage understanding).

Video does not have sound.

I want to highlight the output of one of the AI actions here, where the publisher wants to change the tone to ‘confident’. The AI adds the word “confidently,” arguably an unnecessary bit of text at the end of the sentence. This highlights the need for editing mentioned in my previous post. So while people may save time re-formatting content, they’ll still need to check the accuracy, tone, and phrasing before it can be published.

The original text

The AI amendment

Interact image search or generation

In the coming months employee experience platforms are going to introduce generative AI for image creation as well as for words. This is an example from Interact, where publishers will enter a written prompt and indicate a style for the image such as “abstract”. This could be a time saver for publishers and help make text-heavy intranets more attractive to employees.

The process for generating images in Interact is simple.

Human intervention is needed here though to ensure the accuracy of images as well as words the AI generates. The below example was created from the prompt “cats in bow ties” in an “oil painting” style. Note there are more bow ties and bodies than heads in the final image, which is an issue with AI, not the way Interact has implemented it.

Organisations will have to consider a number of issues here, including whether they’re comfortable using AI to create images, in spite of concerns around authenticity and ethical implications for photographers or other graphic artists.

Additionally, is AI an acceptable alternative to stock images, or is it just a slightly different version of stock imagery? There’s a lot to think about. 

Microsoft Copilot in SharePoint for creating sites

Copilot will use large language models (LLMs), company data in the Microsoft Graph, and Microsoft best practices to turn written prompts into SharePoint sites. The publisher describes the site or page, Copilot in SharePoint will begin to create it, and then work with the publisher to edit and refine it using conversational prompts. The site will collate content from across the intranet and align with existing branding.

This looks great, but the site will still only be as good as the content that goes into it. From experience, it can be tricky to get publishers to consistently use templates, so this may give people too much freedom and result in even more inconsistency. A potential solution is to have a central admin team handle the initial creation then hand to departmental colleagues to refine This could change the way intranet teams are currently structured though, so organisations will need to consider the impact.

Video does not have sound.

Microsoft Copilot in SharePoint for creating pages

Copilot will help page authors turn their existing content into SharePoint pages. Microsoft promotes Copilot as a “web design partner,” where it takes an existing document or presentation and turns it into a page that uses all the available web design visuals within SharePoint. Copilot can also help publishers rewrite page text, adjusting tone and length.

Again, this relies on the quality of the initial content. If a publisher has created an ugly PowerPoint then the resulting SharePoint page won’t look great either. Also, have you ever used PowerPoint where it suggests slide layouts? In my experience they aren’t always the nicest looking suggestions, so I wonder whether that will be a familiar experience when using Copilot here.. However, for those with a good starting point this will save time and potentially improve the resulting SharePoint page.

But again, the fact that this process is so easy creates the risk of loads of pages being created instead of alternative and more suitable content approaches or solutions. Getting governance right therefore remains important for intranets and for intranet teams.

Video has no sound.

Microsoft Copilot in Viva Engage for post creation

Copilot is also going to be available in Viva Engage and this first example is designed to encourage leaders to get involved in conversations.

When the CEO or other leader selects a campaign or topic, Copilot will offer a summary of sentiment on that topic, including prompts to help jumpstart communications by providing a first draft to edit. Using these prompts, the CEO can drill into what other leaders or employees are talking about, add details to complete the post, and choose the tone of the message before personalising with their own voice. Copilot then creates a draft post and will even add relevant attachments automatically. The CEO can then revise the message as needed before publishing.

For communicators who struggle to encourage leaders to participate in spaces that feel social, like Viva Engage, Copilot might encourage usage as it reduces some workload. There’s quite a lot being presented on this screen though, so the leader would need to read, absorb, use the AI to generate post, then edit – so it might not be that time saving after all. They will definitely need to check and edit the post before it’s published, to ensure it accurately reflects their personal voice.

There is a ethical aspect to consider here of course, around the authenticity of any message from a leader that isn’t written by them. Yes, it happens now with ghost writing, but this is going an extra step to remove the human connection altogether. This should be an area of discussion within organisations before it’s rolled out.

Copilot in Viva Engage aims to help people, particularly leaders, get involved in conversations.

Microsoft Copilot in Viva Engage for anticipating replies

Copilot will also predict potential questions readers may have about a post, and prompt the author to answer those questions before they are even asked. As comments and reactions come in, Copilot suggests replies and prompts the publisher to refine the content over time.

In this example, Copilot is helping the author write the most helpful possible post for their readers. While it might feel longer for a publisher to read and absorb the suggestions, it could save them time in the long run as the post anticipates and answers questions. The publisher will need to consider whether the suggested additions are beneficial and whether the additions are accurate or not, so editing skills are needed again here.

However, answering questions before something is shared isn’t always helpful. Sometimes publishers may want people to ask questions, even obvious ones, to generate conversation and encourage a community feel, or even to show that the organisation is listening by publishers openly responding to comments.

Copilot in Viva Engage will offer prompts for people to consider what questions might be asked, before they’re even asked.

Microsoft Copilot in Viva Engage for ‘Answers in Viva’

Copilot is also coming to ‘Answers in Viva’, which is a Q&A space that encourages voting on questions and answers to crowdsource information. When an employee asks a question, Copilot can help suggest additional details to improve the clarity of the question. Copilot can also help people provide answers to questions, by recommending existing resources and potential answers.

For questions that have many answers or replies, Copilot will create a summary of the best answer, as well as an overview of all of the contributions. Copilot will also automatically aggregate information about each knowledge area into a curated view complete with copy, FAQs, and resources so people can dive deeper into the topics they are curious about.

For me this seems a bit superfluous – if the AI can help ask the right question, write the right answer, and summarise comprehensive results, why can’t it jump to the final step and comprehensively answer the question in the first place? It feels like a long-winded approach that’s purpose should either be to encourage colleague collaboration (which should rely on knowledge that might not be captured elsewhere in writing) or help people find ‘official’ content more quickly, and it’s missing the mark in both places just now.

While it does look easy to use and could be helpful for some, I think it also risks creating more content that a search engine would need to filter through or highlight the need for governance to tidy any outdated content.

Suggested answers and resources in ‘Answers in Viva.’

A version of this article was first published by Reworked.