At Microsoft Build, and more recently at Microsoft Inspire 2023, Satya Nadella (Microsoft CEO), and Kevin Scott (Microsoft CTO and EVP of AI) made some key announcements about Copilot, a new generative AI tool that is going to revolutionise the work and personal productivity.
But how does Copilot integrate in the overall Modern Workplace architecture, what are the implications for company data and security, and how can you tailor Copilot to meet your organisation's business and technical requirements? Let's find out!
What is Copilot?
Copilot is a virtual assistant powered by modern AI to assist users with complex cognitive tasks in the workplace. Copilot combines the power of LLMs such as OpenAI's GPT, Codex and DALL-E, with the user and company data from their Microsoft tenancy.
Where is Copilot being rolled out to?
Copilot is being rolled out for the entire suite of Microsoft products. It's currently in private preview, and a GA release date hasn't been communicated yet.
Benefits of Copilot
- Microsoft 365: better reading and writing experience, better e-mail management, better presentations.
- Microsoft Teams: better meetings, better knowledge management through the new Business Chat chatbot.
- SharePoint Online: faster SharePoint page authoring using natural language.
- Microsoft Viva: better Answers experience in Viva Engage, insightful conversation starters based on sentiments and trending topics.
- Microsoft Dynamics: better sales and customer support.
- Power Platform: better creation of apps, flows, and agents, automate repetitive tasks.
- Fabric: advanced AI capabilities for Power BI.
- Microsoft Security: better threat detection, identification and mitigation.
- Windows: better interaction with the apps, files, the operating system.
Security, Data and Privacy aspects
Copilot's data sources are:
- Organisation's Microsoft Graph
- Organisation's Dataverse
- Azure Cognitive Search
- Microsoft Search
- Other Microsoft data sources (Microsoft Bing, GitHub)
Copilot unlocks business value by connecting LLMs to organisation's business data in a secure, compliant, privacy-preserving way.
Copilot has got real-time access to the organisation's Graph and Dataverse. However, this consent can be revoked at any time. Furthermore, Copilot accesses organisation data impersonating the user's context, that means Copilot cannot access data that a user itself would not have access to.
Microsoft states that they don't use customers' data to train LLMs. OpenAI has got no access to the organisation's data and model.
How Copilot works
Copilot is powered by Azure OpenAI Service, which is an enterprise-ready version of the OpenAI core services that ChatGPT is based on, and it's maintained by Microsoft. Azure OpenAI leverages OpenAI's GPT (GPT-3, GPT-3.5Turbo, and now also GPT-4). However, there are some important differences between Copilot and OpenAI's ChatGPT. Some of these differences are evident, such as ChatGPT having no access to Microsoft customers' data, while others may not be as apparent:
- Microsoft conducts their own ethical, security, compliance and privacy checks at various stages of Copilot actions' lifecycle.
- Copilot applies grounding to prompts in order to enhance the context of the prompt and mitigate the "hallucination" issues commonly associated with LLMs.
Let's clarify it with an example. Let's see what happens when a user enters a prompt into Microsoft 365 Copilot:
- The user enters a prompt into Copilot.
- (pre-processing step) The user's prompt is grounded on Microsoft Graph data, and a modified prompt is created. This step is called pre-processing.
- Instead of directly sending the user's query to the LLM, Copilot calls the Microsoft Graph to retrieve the user's business context and content.
- The prompt is then modified to make it more relevant or to prevent "hallucination" problems (...imagine what would happen if Copilot provided a plausible sales quote from a sales meeting that never occurred, and as a Sales director, you forwarded that quote to your CEO...).
- The modified user's prompt is sent to the LLM.
- Responsible AI checks are performed to detect hate, violent, sexual, or self-harm content in images and text.
- (post-processing step) Copilot takes the response from the LLM and post-processes it.
- It makes additional grounding calls to Microsoft Graph.
- It makes additional responsible AI checks.
- It makes security, compliance and privacy checks.
- The response and the app commands are then returned to the user.
Microsoft Copilot will be extensible. This will allow third-party providers and software developers to integrate their apps and services into Copilot, further expanding its capabilities. There will be three possible types of plugins available: ChatGPT Plugins, Power Platform Connectors, and Teams Message Extensions.
Interoperability with ChatGPT Plugins
Microsoft is going to adopt the open standard introduced by Open AI for ChatGPT plugins. This besides will allow also for interoperability between ChatGPT's and Copilot's plugins.
How to get your Modern Workplace ready for Copilot
Licensing and technical requirements
In order to roll out Copilot, organisations must make sure they have the proper licensing and technical prerequisites in place.
Users must have an Enterprise license (E3 or E5), or (not confirmed yet) a Business Premium/Standard license. Users must also have a Microsoft Entra ID (an Azure AD account).
Copilot will come with its own license, and it's going to cost 30USD/user/month.
Permissions and content management
Copilot will access organisation data impersonating the user's context, therefore whatever content the user will be given access to, will be reachable by Copilot too. That means, bad practices such as oversharing and lack of data governance will be amplifying their negative effects, with Copilot.
Therefore, organisations that seize this opportunity to establish a structured and compliant environment, ensuring data is appropriately managed, secured, and leveraged to drive informed decision-making, will be the ones that get the best out of Copilot once it's rolled out.
Adoption is crucial in organisations when a new technology is introduced because it ensures that the potential benefits of the innovation are fully realised. Copilot makes no exception to that. Creating training programmes, establishing a Centre of Excellence, nominating Champions, building an internal Community around Copilot are all initiatives that organisations can take to get ready for this transformation in how we work with Microsoft Copilot.
Copilot introduces a new whole way of working, and it's set to have a huge positive impact across every aspect of your Modern Workplace. It complies with the security and privacy regulations and can be extended to cover virtually any user and app scenario imaginable. It's certainly not cheap, so organisations will need to evaluate the ROI on Copilot with a "pilot" (excuse my wordplay) once the tool is generally available. This will involve a restricted group of users before rolling out the tool to the wider organisation. There will also be a need to assess and adapt the governance and control frameworks.
Without a doubt, Copilot will help organisations to streamline operations, increase productivity, and improve decision-making processes. By automating repetitive tasks, Copilot will allow employees to concentrate on more creative and strategic aspects of their work, fostering innovation and driving overall organisational progress. Moreover, Copilot-driven insights will assist businesses in making data-led decisions, resulting in enhanced efficiency and improved customer experiences.
To find out more or to speak to a member of the pre-sales team, please email firstname.lastname@example.org or call us on +353 1 603 4300