Here are some practical steps to roll out Copilot and set your team up for success. Feel free to pick and choose, based on your industry, your team size, and the amount of capacity you have to dedicate to this.
You can also switch Copilot on straight away. It's not immediately customer facing, so it won't have a negative impact on your customer experience and you'll quickly start to learn how Copilot works with your content.
Step 1: Make your content available to Copilot
Content section is where you store and manage all sources to power AI, agents and self-serve support on your workspace. From here, you can start importing or creating new content and decide which content you want to enable for Copilot.
Step 2: Test it out yourself
Now that you have a resourceful AI Copilot trained on all your knowledge, it's time to test it out! You need to use Copilot to get a sense of how it works with your content, and then use these learnings to enable your own team. That way when you introduce it to your team, you'll be coming from an informed place.
You can test it right here in the right side Preview panel: select Copilot from the menu and start asking it some questions.
Step 3: Enable your team
Introduce Copilot to your support team and ensure they know how to access and get the most out of Copilot. From onboarding queries to technical troubleshooting, Copilot is designed to assist all team members.
Including your more experienced support agents in the initial rollout allows for a thorough evaluation of Copilot's capabilities across a variety of queries and complex topics. This also helps transform these agents into advocates for Copilot when you implement it across your broader team.
Here’s how we introduced Copilot at Intercom:
We created a tiger team of trusted and tenured agents who were involved in the test. We did a quick kick off meeting to walk them through the tool and created a dedicated Slack channel for them.
Over the course of a few days, we prompted them to share their Copilot experiences here:
Success stories of when Copilot worked well
Examples of when it didn’t
Content gaps they identified
Suggestions for new content
Our team leads also sought more anecdotal feedback in daily stand ups and 1:1 sessions with the tiger team.
Step 4: Turn it on for your team
Copilot can be assigned to any teammate and they will be given unlimited usage by default as part of a flexible per seat Copilot add-on. From Settings > Workspace > Teammates you can choose which teammates to assign a Copilot sea, or switch it off for a teammate.
Step 5: Ongoing optimization
Ask content owners to review their content
If you’ve got existing support content that hasn’t been reviewed for a while, you should dedicate some resources to auditing and ‘cleansing’ that content. AI is only as good as the knowledge you give it, so if your content is out of date, the answers Copilot provides your teammates will be less accurate too.
Check for accuracy and update or archive content where necessary. From sales and engineering, to legal and security - knowledge can be created by multiple teams and subject matter experts (SMEs). So divide your content by owner and ask each team or SME to review their knowledge area.
Reviewing all support content at once can be a daunting task! Focus on the most used content first i.e. the content with the most views. Employ the 80/20 rule here and ensure that your more popular content is up to date first.
Optimize your content for Copilot
Once all your content has been fact-checked by experts, you should start looking at how each article is structured. When creating content for AI, there are some formatting best practices that can make a big difference between making the answer available, and allowing Copilot to easily find and use that answer in the right context:
Use headers
Use rich formatting (tables, bulleted lists, etc.)
Avoid ambiguity
Restate questions
Give context to multimedia
Create a style guide and reusable templates for different types of content, then share them with knowledge managers to ensure all new content follows this best practice for AI.