Copilot is a powerful workspace tool – if you have your data house in order

After several months of working with Microsoft Copilot, I agree with the 70% of users Microsoft surveyed late last year who said they were more productive after using it. 

With Copilot's help, I've cut down time spent on mundane tasks, can now summarise email threads for quicker reading, and draft quality documents in much less time. 

I recently needed to put together an internal memo for senior managers. In the past I would have put aside three or four days to complete that task but with Copilot, I had a document ready to review in 45 minutes and my colleagues were happy with the results.

Some have pointed to the cost of Copilot for Microsoft 365 - NZ$48.50 a month on top of a standard Microsoft 365 license, as being on the expensive side. But I recoup that investment on the first Monday of each month. After a few weeks of use, I had already received a year's worth of value, taking into account the time it had saved me.

What we have found with Copilot is that it is excellent at content-related tasks such as generating documents, drafting and summarising emails, and making sense of Teams chat and video call transcripts. It's not as proficient yet at putting Powerpoint presentations together, and it has some way to go before being useful in Excel, which still sits at the heart of many organisations due to its versatility with large amounts of information.

Making strategic use of Copilot

Rather than looking at blanket deployment, we've found that for Copilot, the return on investment comes when you target areas of the business for its use, buying licenses for individuals or particular groups of users. For example, it can instantly add value to aspects of marketing function – marketers can use prompts to build content that better targets their GTM audience, and graduate-level marketers can use Copilot to step up their content creation.

It can be a bit hit-and-miss with business administration tasks, but that will likely improve dramatically in the next 12 - 18 months. For decades, we've been promised greater productivity through the uptake of technology. But reality hasn't always lived up to the promise. Smartphones, apps and the advent of cloud computing haven't necessarily saved us time and hassle. However, my conversations with C-suite executives who have adopted Copilot suggest this application of artificial intelligence is different.

It signifies a step change in how people will do their jobs in the future. Some types of jobs won't exist because of Copilot-like services being widely deployed. But new, higher-value roles will be created, as we saw with the internet revolution.

Tidy up your data domain

Still, Copilot currently has its limitations. It can generate a well-written, eloquent, and grammatically correct email for me. But as a 50-year-old dyslexic Scotsman, anyone who receives that email will know that it can't have been written by me. AI assistants have some way to go to really learn and embody our personal style and reflect the context that we work in.

The large language models (LLMs) that underpin them are very much "garbage in, garbage out". The quality of the data they are trained on determines the quality of the answers they will give you. You might have 25 years of accumulated human resources policy documents in a Sharepoint folder. But chances are that if you ask Copilot to give you the leave or vacation policy, it won't be 100% accurate. 

You need to get your information in a state that the language model can consume and come up with sensible, accurate, and relevant answers. Keep the relevant, accurate documents and archive the rest. Every business looking for the productivity benefits of Copilot now has the task of tidying up their information domain to make good use of it. This doesn't have to be tackled in one go but can be broken down into manageable pieces of work as you deploy Copilot department by department or group by group.

The next steps - multimodal and context-aware

For a glimpse of how AI assistants will evolve this year, we need only look at the world of software development and the rise of AI agents like AutoGen and Devin. AutoGen is a framework that enables the development of LLM applications using multiple agents that can converse with each other to solve tasks.

Similarly, Devin is an autonomous model that can plan, analyse, and execute complex code and software engineering tasks with a single prompt. You now have agents for coding, reviewing, testing, and publishing software, all working together.

If you apply that approach to the marketing department, we'll soon have agents for graphic design, wordsmithing copy, web development, and content reviewer, all operating in unison. The multimodal nature of AI assistants will see text, video, audio, and code generated via one user interface. In the executive assistant role, we'll have agents for travel planning, diary scheduling, and email management. 

We'll also see the increasing capacity for AI to be applied to unstructured data. This is why the big software as a service (SaaS) providers have moved quickly to infuse AI into their products. They realise that if they don't offer a compelling experience with AI, they may not continue to be the gatekeepers of information for your business in a world of AI agents and multimodal chatbots cutting out the middlemen.

Security and privacy remain top priorities

Copilot runs on your Microsoft tenancy, so your data remains as secure and private as the documents held in Sharepoint and the chats in your Teams account. However, you may look to use other, more specialised third-party AI models for different purposes, such as legal research or graphic design. 

You need to be really careful that information shared with models beyond the realm of your Microsoft tenancy is secure and kept private. When it comes to factual inaccuracies, bias and use of copyrighted information, the major LLM developers, Microsoft included, have put a lot of guardrails in place.

We don't see a lot of hallucinating chatbots any more, and bias has also been reduced. Copilot is very good at referencing information sources, so you can see exactly where its outputs have come from. 

The value of AI assistants as workspace tools will increase exponentially in the coming years. Already, they have the potential to lift the collective intelligence of your organisation by giving your people more time to focus on the things that matter to the business. 

You don't need to pursue a big-bang approach with Copilot. Rolling it out gradually to address the most compelling use cases will improve prospects of a decent return on investment and build the case for its wider adoption.