author: Jonathan Stuckey
For organisations starting the journey with Copilot and SharePoint Agents the initial road-block is determining what are good candidates for tackling with this new suite of tools?
First you really need to determine suitable fit for each because Microsoft Copilot, Copilot chat and SharePoint Agents are targeted in entirely different ways. Ultimately they all use AI, Machine Learning and your content sources for generating new content, but you have to tackle their use-cases differently.
If you don't need the exposition, you can jump ahead to the examples and how I've collected (and then generated) them:
Targeting your approach to identify usecases
Microsoft's Copilot branding-stick has made a mess of everything but you can summarise the core Microsoft 365 Copilot suite broadly as:
Suite product | Scope | Focus | usage | Comments |
Microsoft Copilot for 365 | Embedded GenAI experience in Office applications | Context specific to the app, or type of content & information | Tuned on per app or type of content basis. Harder to learn to drive it well. |
Copilot Chat (formerly BizChat) | Centralised search-like UI experience. Grounded on all accessible content in your tenancy. | Question/Response driven. Using retrieval augmented generation (RAG) from your content. | Your orgs personal 'GPT' home experience. Broad-spectrum GenAI based on a search UX. |
SharePoint Agents (formerly Virtual Agents) | A site or content deployed Agent. Limited grounding scope. | Localised UX. Targeted to specific idea, role, process or outcome. | Pre-populated on sites, Wizard-driven setup makes this super simple RAG UI. |
It would waste my time and yours waxing-lyrical about Copilot for Office and what was called "BizChat", but is now "Copilot Chat". We are here to focus on getting going with SharePoint Agents.
Where to start
Copilot in the Office applications are tuned to get your content kick-started using format specific focus for the types of prompts you can create i.e.
in Outlook we might ask: "tell me what's happening over the next 3-days, summarise meeting objectives and tell me about associated information received previously", or
in Word you could ask Copilot "draft me a new <name_topic> policy for employees. Use <template_name> and this background document <document_name>"
...and you'll get a quick-summary draft to use.
Use-cases for SharePoint Agents though should be a lot more process or role specific, and can enable the concept of a 'virtual assistant' that is directly aligned to the content, role, function or task being undertaken.
Examples
Everybody doing research on the web will see the following tired, and largely irrelevant examples touted from Microsoft:
Streamlined employee onboarding: New hires often have dozens of questions about how to find documents or request access to teams. A SharePoint agent can answer these questions and guide them to the right information.
Internal help desk: Instead of emailing IT or HR for common queries like “How do I reset my password?” or “Where can I find the new expense policy?”, employees can ask the Agent and get immediate answers (or directions).
Budget planning: The SharePoint Agent can provide recommendations and prioritizations for the upcoming planning cycle, and reduces time intensive analysis and costly margins of error so the team can focus on strategic financial planning.
So what did I do? Well... isn't it obvious? I used 3 different Generative AI services to identify what, where and how I might target different business functions, roles and processes for use with a Copilot Agent.
I started with a simple prompt: "identify 30 business process or functions, where a SharePoint Agent can replace a specific area of expertise. List these functional areas in business, process name, brief description of the process or activity performed by specialist. format theses items as a table." and then I refined this by appending: "include additional information per item, which Identifies the Business Function normally associated to the "Business Process Name", and name 3 typical data-sources / documents / website URLs which supply the related grounding for the Agent" ...and so on until I had a set of useful prompts that generated a good set of responses.
The results were interesting, but ultimately too lite for actual use in generating an Agent, but does give you a good sample of starting points... e.g.
Business Process Name | Brief Description | Business Function |
Health and Safety Audits | Conducting audits to ensure compliance with health and safety regulations. | Health & Safety |
Legal Compliance | Ensuring adherence to legal requirements and standards. | Legal |
Training Needs Analysis | Identifying training needs within the organization. | Training & Development |
Internal Communications | Managing internal communication channels and strategies. | Communications |
Quality Control | Ensuring products or services meet quality standards. | Quality Assurance |
Risk Assessment | Identifying and evaluating potential risks to the organization. | Risk Management |
Marketing Strategy Development | Creating marketing strategies to achieve business goals. | Marketing |
Business Strategy Development | Developing strategic plans to guide business decisions. | Strategy |
Policy Development | Creating policies to guide organizational practices. | Policy Development |
Governance Frameworks | Establishing frameworks for organizational governance. | Governance |
Leadership Development | Developing leadership skills within the organization. | Leadership Development |
Talent Acquisition | Attracting and hiring top talent for the organization. | HR/Talent Acquisition |
Crisis Management | Managing responses to crises or major disruptions. | Operations/Risk Management |
Stakeholder Engagement | Engaging with stakeholders to inform business decisions. | Communications/Strategy |
Sustainability Reporting | Preparing reports on sustainability performance and initiatives. | Sustainability |
Compliance Auditing | Conducting audits to ensure compliance with regulations. | Compliance/Audit |
Data Protection | Protecting sensitive data from unauthorized access. | IT/Cybersecurity |
Business Continuity Planning for IT | Ensuring IT systems continue to operate during disruptions. | IT/Risk Management |
Innovation Management Frameworks | Establishing frameworks to encourage innovation. | Innovation/R&D |
Intellectual Property Management | Managing and protecting intellectual property rights. | Legal/IP |
Regulatory Affairs | Managing compliance with regulatory requirements. | Regulatory Affairs |
Now, what should scare you is that there is something in pretty much every major business function or process which can be helped, streamlined or improved - according to the Generative AI.
Generating potential targets for agents
From working with various customers we've come up with a catalogue of options, as part of your adoption I would suggest using some of the global services (or possibly even Microsoft 365 Copilot) to help with 'bootstrapping' where you start.
Actually what you need to do is take the ideas captured for your target options and run them through a few reality checks for your organisation and people before trialing some.
For example ask questions like:
Q: do you have this process / function in your organisation?
R: lots of small organisations don't have a lot of the roles or processes, so there's no point in creating something you don't want or need. Only focus on ones which apply to your situation
Q: is there a lot of paper, forms or repetitious actions involved in your version of the activity?
R: if the proposed target is something you do in structured system (CRM), or embedded in core platform which is automated already there's no reason to look at it. ...yet. If not, if it's paper / document heavy it should be reviewed for potential
Q: does the activity have a lot of overhead, or required significant effort (mental, not physical) to get the output?
R: these could be good candidate if its document oriented, manually driven, requires a specialist knowledge-set which is rules based... Put them on the list
Q: is the target something that you maybe don't retain in-house and you need to got an external professional? not because you couldn't do the process, but because you don't do it often enough to warrant hiring someone?
R: specialist roles, that are based on procedural, documented regulations or governance activities and that usually require steep investment in time, effort and experience (which is documented)... chuck them in to be evaluated too!
Q: Is the output structured content / document / form / media etc? Would it have checks, reviews or other process steps you may/may not have documented involved?
R: These are where you should concentrating!
In fact, any role or task associated with heavy rule-based process, and / or a standardised format for output is a prime candidate; Activities that you may have previously gone out for 3rd party to do a lot of leg-work, but ultimately they just "process" your info and give you some tailored (but relatively generic or repeatable) advice - these are in the crosshairs too.
I've built up catalogue of roles and tasks over the last 2-years, and better rules-of-thumb and guidance on refinement for agents to start and there's some really obvious ones, which interestingly get a lot of push-back until people see how this supports them.
Creating a SharePoint Agent
Once you've identified the right business areas, roles and tasks you want to support with an agent, you need to identify how it is going to work for you people because the agent is not replacing your people, its providing assistance to them.
Rather than starting with the "research-led" approach for the agents, select the task its supporting and then apply the 'Flipped conversation' approach or similar inversion to the end-user beginners-guide.
e.g. Creating the "Frank" the Financial Advisor agent for responding to budgeting, expenditure and reporting needs in the organisation
So to create an agent...
identify your source content and documents to ground the agent in Financial process documents, relevant external sites or documents which identify financial rules, and or legal precedents
identify constraints like the need to limit access to some information or data e.g. enforcing access restrictions based on applied Sensitivity Label or permissions for user (based on their security privileges and role-based access to source content)
define the way in which the agent should interact, by framing the tone, types of questions, scope of answers to supply, if there should links for reference data for the users, examples which may be available for Budgets, or internal management DFA etc
run the agent wizard, bung-in the definition, publish to the site and test
Just for fun here's a simple example based on the 'Flipped conversation' approach:
"You will be acting as a very experienced Financial controller in a New Zealand business. Your task is to engage the user in a dynamic conversation about their business challenges and provide tailored advice.
Start by introducing yourself: "Hello, I'm Frank the financial advisory assistant! I'm have a vast amount of experience and financial expertise available and I'm here to help you optimize your business's financial operations. What's the primary financial issue you're facing right now?"
After the user responds, generate 3-5 relevant follow-up questions based on their answer.
These questions should:
Dig deeper into the financial problem
Explore potential causes of financial inefficiencies
Identify affected areas of the business's financial structure
Uncover attempted financial solutions
Ask these questions one at a time, waiting for the user's response before proceeding. Use the information gathered to formulate your advice.
Once you have a comprehensive understanding, provide your insights:
Summarize the financial situation
Offer 2-3 potential financial solutions
Discuss pros and cons of each solution
Recommend a course of action for financial improvement
After giving advice, ask: "How does this sound? Would you like to explore any other aspects further? Or expand on the question(s) you asked?"
Throughout the conversation:
Adapt your language to the user's level of financial expertise
Share relevant financial anecdotes or case studies if appropriate
Encourage the user to think critically about their financial situation
Be open to clarifying or expanding on your financial advice
If the user introduces new financial information or challenges your advice, adjust your approach accordingly. If there is little or no background information available to use in response, then direct the person to people who can help saying: "Hmmmmmm - I'm sorry either I dont have access to the right documents, or its just available to you. Please take the information provided so far and contact someone direct on 0800 Finance or email finance@organisation.com"
Remember, your goal is to guide the user towards effective financial solutions while fostering their own financial management skills. Begin by introducing yourself and asking about their primary financial challenge.
Refining the experience and agent updates
Now obviously you need someone who knows and understands the content being used, to run their frequently asked questions, and be available to validate the responses, and provide feedback on what works, doesn't work, is missing (content), shouldn't be accessible (permissions), etc
You will also need that person, and someone capable of helping to (re)frame the agent definition(s), like the example above, to focus the agents response and scope over time to improve the outcomes.
I would recommend that you have someone with Copilot Studio handy, because there's a few precautions and controls which Microsoft thoughtfully leave open and hooked-up to their models - which is potentially a massive privacy &/or data security issue if you don't address them before publishing org-wide.
...What, you thought this was magic and didn't need you involved afterwards?
Don't you worry, there's a whole host of learning and management todo along the way.
Want to know more?
Give me a call and I'll walk you through how to take the next steps. While the above is a good starting point - it's only that.
Resources
About the author: Jonathan Stuckey
Comments