The beauty of AI is its ability to streamline processes and alleviate people from mundane, repetitive tasks. If done correctly, this can save time across your processes and allow people to self-serve the knowledge they need.
Copilot Agents particularly lends to this, automating manual tasks and removing productivity bottlenecks.
As more businesses begin to discover how they can leverage AI agents in their processes, we wanted to share our own way of using it. In this use case, we’ll be diving into how we turned an AI agent into a product knowledge resource, protecting consultant time.
The challenge_
As an IT and technology company, we provide a wide range of deeply technical products and services to our clients.
It’s unreasonable to expect every seller and consultant to memorise all the information and requirements across the products. However, they need this knowledge to answer client questions and ensure they are delivering the right message.
While there is technical documentation available, these are lengthy and complex. Few people have the time or desire to sift through them to find the answers, especially if they aren’t from a technical background.
Due to this, consultants are often asked questions about the products from other teams, who are looking to develop their understanding or check client relevancy. The questions commonly cover the same ground, meaning consultants must answer the same queries again and again.
Tom Lovell, Infrastructure and Modern Workplace Principle Consultant, faced this regularly and found it frustrating. He explains: “It’s one of those things that makes you curse when the phone rings. It takes up a lot of time, especially when it’s the same questions repeated. On top of this, they’re stuck until I have a chance to come back to them”.
How we use Copilot_
Facing endless questions about technical products, Tom saw a clear use case for a Copilot agent.
He states: “The objective for the agent was to make those repetitive questions go away. It can answer the questions for people, based on the technical documents which has the information they need. Rather than having to read the document in full, they can ask the agent the question and get an answer. They don’t need to wait for me to answer, and they can improve their product knowledge at the same time.”
For example, if a seller needs to know if their client meets the prerequisites for the product, they can ask the agent. It will then concisely list them in a user-friendly way, including things like device specifications, licenses and other important criteria.
The decision to create an internal AI agent was also crucial to streamline the information served.
Tom explains: “When you’re using open AI with access to the wide web, answers can easily become muddled based on disorganised information. There are multiple answers for any question, and it’s difficult for the AI to work its way through them rationally. But when we’re working internally with an agent, it only knows what I’ve told it, giving it a limited set of knowledge.”
This means that the agent will always serve relevant, accurate information, reducing the risk of imbalances and hallucinations. This is backed by a large-language model, which allows it to deliver a human response that breaks down complex information into understandable guidance.
Preparing the data_
The success of our agent lays in its ability to accurately answer product questions. The biggest challenge is ensuring it is pointed to the right data, and no more.
For us, this meant our agent to a centralised location storing technical documentation – in our case a SharePoint site.
But it isn’t as simple as just linking to files in a folder. You need to ensure that the data is up to date, and any information you do not want the AI to access is removed.
It means spending time cleaning up your data in advance and employing techniques like segregation and folder permissions to limit what the AI can search through. This will prevent it serving misinformation or sensitive data to people.
This is something you need to maintain long-term as your data evolves. Doing so will drive the accuracy of your AI output, while preventing breaches and stopping information getting into the wrong hands.
The results_
The product knowledge bot is still in early phase, but it is poised to save significant time. Tom was being asked product questions around four to five times a week, making it a notable drain on his capacity.
Tom summarises: “The biggest win for anyone using AI is automating the mundane. If it is something you are doing several times a week, which can be easily fixed, it’s a perfect use case for Copilot. I was acting as a knowledge base for people, when AI could be doing that instead, saving me time for other work”.
As we already had technical documentation in place for our products, we simply needed to connect the Copilot to that data. In all, creating the agent took Tom minutes – and he’ll save a significant amount more time. This offers a great return – and shows how easy it can be to leverage AI.
What can you achieve with Copilot?
As an organisation, we’re committed to experimenting with AI and using that to inspire and guide our clients. We’ve already found some great ways to use Copilot internally, and we continue to invest time in it every day.
If you’re looking to learn how to embed it in your business, we’ve got some resources to help you:
- How we began our organisational AI implementation
- 45 Copilot use cases to try out today
- AI agent examples for your business
We’re also always happy to talk to you about our AI usage and the lessons we’ve learnt so far. Just reach out below.
