The popular options are:

  • Go direct to vendor and invest in ChatGPT Team (made by OpenAI) or Claude Team (made by Anthropic) accounts for your users
  • Go the Microsoft way and invest in Copilot accounts for your users
  • Build your own AI environment and connect it to the AI models you’re keen on (full disclosure: this is what we do for companies)

There are pros and cons to each.

However, Microsoft has aggressively marketed Copilot from late last year and we’re increasingly finding companies who don’t understand what they’ve signed up for and are severely limited as a result.

First off – the advantages

Copilot is designed to bring AI to the user i.e. it appears in Microsoft 365 apps and is very visible and easy to use as a result. It also (if you’ve got some internal technical nous) connects to Microsoft Graph including your calendar, emails, chats, documents and meetings). Very useful if you use Microsoft Azure for document storage.

However, there are definite drawbacks.

Limited functionality

Copilot puts frustrating limits on your team’s productivity.

According to Microsoft you can only type 2,000 characters at a time when asking it to do stuff.

In addition, you can only give it PDF files that have fewer than 20 pages. And, chats should not be longer than 15,000 words.

In future these limits will increase, but at the moment you’re severely restricted.

That’s pretty outrageous really. Copilot uses OpenAI’s GPT-4 Turbo, a model that supports chats up to 96,000 words long.

Also, OpenAI doesn’t put limits on PDF length if dealing with them directly.

Paying for users not usage

Copilot licenses cost £19 per user, per month. For 50 users, that’s £950 a month.

Two things:

  1. You pay this whether these users use the product or not. Forever.
  2. If you want access to AI models beyond the OpenAI one Microsoft gives you (Microsoft is currently using OpenAI’s AI, essentially reselling it – more on that later), you have to create new accounts. Say your marketing team needs Anthropic’s latest LLM (Claude 3 is – at the time of writing – better at ‘writing’ than the Open AI models). You’d need separate Anthropic Pro accounts at £18 per user, per month. Another £900 on your monthly bill. Same goes for any other model outside Copilot. Like Google’s Imagen. You’d need extra Gemini licenses at £20 per user, per month.

Old AI models

Copilot is powered by OpenAI’s AI models.

Microsoft is reselling them to you.

This means you have to rely on Microsoft to update the models when new ones become available if you want the most powerful AI that is the best at following instructions and outputting content.

At the moment, for example, Copilot ships with OpenAI’s older GPT-4 Turbo model. This was superseded by GPT4o in May.

You also can’t access other AI models in Copilot. Anthropic’s latest model, Claude 3.5 Sonnet, is great at writing marketing content. Google’s Imagen 2 model is great at outputting photorealistic images. You can’t access either via Copilot.

And this is really tip of the iceberg stuff. There are lots of other great AI models (video, audio, translation models etc.) too.

No customisation

Copilot doesn’t have a central prompt library tailored to your organisation’s tone of voice and specific needs: somewhere you can store all of your prompts your users have lovingly crafted.

It does have a GitHub based prompt library, but none of the prompts in there will write content in your organisation’s tone of voice. Nor are they easy for your users to get to.

Likewise, if some game changing AI tech is released, you won’t be able to benefit from it unless Microsoft decides to integrate it. For example, in April, OpenAI released something called File Search. This allows users to augment its AI models with their own data. We integrated the tool in six weeks and are making full use of it already.

And what if you want to harness the power of AI from a customer/client service perspective and do some bespoke development? Nope. That’s definitely got to happen outside of Copilot.

Final thoughts

The AI world is moving quickly: Anthropic released its large language model Claude 3 in March and then Claude 3.5 in June. Claude 3.5 now outperforms everything else out there.

We are about to see a load of new AI video models flood the market. Google has an AI that specialises in structuring unstructured data. You get the idea. There are lots of new models, that do lots of things that are very useful for your business, being released all the time.

You need to stay agile and give your employees the best tools for the job in the most cost-effective way. You may decide that’s Copilot. You may not.

If you decide to explore other avenues, then drop us a line – we’ll happily advise you on the best options available for your particular use case.

Hopefully that’s food for thought.

Luke Budka Screen

Written by Luke Budka, AI Director at Definition.