Accelerate Your CoE with AI

Learning Objectives

After completing this unit, you’ll be able to:

• Use AI tools and agents to establish and manage your Center of Excellence.
  
• Use AI tools and agents to assist in the research needed to launch a CoE.
  
• Use AI tools and agents to efficiently govern requests and solutions for a CoE.
  
• Verify the output of your AI tools and agents with a human in the loop.
  

A New Era

It’s easy to think of a CoE as a “manual” group, a group that relies on spreadsheets, long meetings, audits, and hands-on testing. But as we enter the era of Agentic AI, CoEs don’t have to be fully manual any more. Agents are powerful tools that make your CoE faster, smarter, and more scalable.

For any work that the CoE must perform, you can categorize it in one of three ways.

• Agent-executed work: These are tasks that can be performed almost entirely by AI agents. A human should still check and confirm each agent’s work, but very little human-in-the-loop interaction should be necessary for an agent to complete that work. Examples include data categorization, checking work against policies, and taking notes at meetings. For complex tasks across multiple systems, you might need an additional orchestrator agent to coordinate and wrangle multiple specialist agents.
  
• AI-augmented work: These are tasks that, while primarily performed by your team members, get a big boost from AI agents and tools. The most common use case is for research or analysis that must be performed by humans, who then use AI to assemble, model, or search through the results. Examples include use case analysis, scenario modeling, and risk identification.
  
• Human-led work: The work of the CoE supports your human employees, so your CoE’s decisions should be made by humans. Decisions about strategy, funding, risk, and complex exceptions need that final human input. And many CoE tasks need that face-to-face interaction to make them work.
  

Let’s talk about the types of tasks that AI can enhance.

Using AI to Launch Your CoE

When starting a new Salesforce CoE, your team has quite a few tasks on its plate. In addition to following the seven steps detailed in the previous unit, you must define the current state of your organization, its processes, and its needs. AI tools can help with many of these tasks.

AI tools are only as good as their grounding. Make sure you ground your tools in your organization’s business, IT, security, and AI policies.

When Establishing a CoE

Though most of the tasks in the seven steps of creating a CoE require human decisions and face-to-face conversations, agents and AI tools can assist with the administrative overhead and other cognitive tasks.

• Take notes. It takes a lot of meetings to set up a CoE. Take advantage of your meeting software’s AI note-taking capabilities and let it take notes for you. If the meeting software’s AI doesn’t provide summaries, ask an LLM to summarize those notes for those who couldn’t attend the meeting.
  
• Aggregate decisions. All those decisions must be tracked somewhere. Instead of assuming that everyone in the CoE has perfect memories, task an agent with going through your meeting notes and logging every decision made in a single, organized reference sheet.
  
• Draft your charter, documents, and communications. Though it’s handy to have quick summaries and reference sheets, you sometimes need longer-form content. Use an LLM to turn your notes and bullet points into a mission statement and scope. An LLM can also draft communications such as Slack posts and emails that announce the CoE and everything it intends to accomplish.
  
• Ask for suggestions. If the CoE has trouble making decisions or coming up with ideas, try asking an AI. The AI might give an unconventional answer, but its alternative recommendations can help you look at problems from a different angle.
  
• Ask Slackbot. Because Slackbot has access to all of your organization’s Slack communications, it has additional insight into how your organization works and who’s doing what. This knowledge can be extremely useful when trying to identify stakeholders and define KPIs.
  

When you finish following the seven steps and have established the CoE, you still need to take notes, collect data, write email, and get help. Keep using AI to help with all of these tasks throughout the life of your CoE.

When Doing Your Initial CoE Research

When you first set up your CoE, your core team members have a lot of tasks to complete before the CoE can be effective. You can’t govern anything if you haven’t documented what there is to govern, and how exactly you’re going to govern it. Use AI tools to assist with these tasks as well.

• Organize your roadmap. Organizing a long list of initial backlogged requests into a prioritized roadmap can be a daunting task. AI can analyze historical project data, resource availability, and business impact scores to suggest your most effective starting point and a ranked order for your subsequent roadmap.
  
• Identify similarities. One of the initial tasks of the CoE is to identify similar use cases and processes across the Salesforce environment so they can potentially be standardized. After documenting those use cases and processes, you’ll probably have a lot of information to sort through, so give those documents to an AI and ask it to find those similarities for you.
  
• Research best practices. The Design Authority (DA) must create a series of guidelines and best practices for your Salesforce team to follow, but Salesforce is always changing and growing, making it difficult to stay on top of current recommendations. The Salesforce community is a great place to find real Salesforce workers sharing their problems, answers, and best practices. If the DA gets stuck when defining your architecture standards, use AI to search through public sites like the Trailblazer Community, the Salesforce Well-Architected Guides, and Reddit for best practices and solutions. Make sure you ground your AI with Salesforce’s Well-Architected Framework
  
• Review initial data. If your Salesforce environment has data prior to the CoE getting off the ground, that data should be considered effectively ungoverned. So when the CoE decides on its initial data compliance guidelines, your existing data needs to be reviewed and corrected. Give an agent your new data compliance guidelines and have it review and suggest corrections for your existing data. Consider giving it rules for identifying duplicates as well.
  
• Manage your to-do list. When the CoE gets started, the above bullet points cover just a few of the items that the CoE will have on its to-do list. If you’re not using a ticketing system for this to-do list, feed your meeting notes into an agent to keep track of that to-do list. An LLM can also analyze everything that the CoE is working on and talking about, look for possible gaps that the CoE might have missed, and suggest additions to the to-do list.
  

Again, many of these initial tasks still need to be done on a continual basis once the CoE is up and running. Keep using AI to help with all of these tasks throughout the life of your CoE.

Using AI to Run Your CoE

When the CoE is up and running, one of the biggest threats to its continued success is not keeping it streamlined. If your team members feel like the time and effort put into the CoE isn’t worth its gains, they may abandon it. Here are some ways you can use AI tools and agents to keep your CoE running at peak efficiency.

• Report to leadership. LLMs are great at rewriting your words for a different audience, and let’s face it: many technical folks aren’t great at writing with executive leadership in mind. Use an LLM to stay aligned with your organization’s leadership by formatting your progress reports into executive summaries.
  
• Check proposed solutions. When someone proposes a solution to the DA, it’s the DA’s responsibility to validate that solution against guidelines and best practices. But sometimes details get overlooked. Reduce the number of unforeseen problems and long-term tech debt by asking an agent to verify the proposal against your Salesforce environment’s metadata. You can also have the agent check any proposed changes to data structures to make sure they comply with policies and regulations.
  
• Automate Salesforce audits. Don’t stop checking solutions after the proposal stage! Set up an autonomous agent to periodically check your development sandboxes, looking for bugs and checking for compliance with DA guidelines regarding build methodologies, naming conventions, and security standards. The agent can catch issues before the work even reaches a human reviewer, saving everyone time and effort.
  
• Generate documentation. While LLMs struggle with generating accurate step-by-step documentation for end users, it can still produce other documentation. Image generation AI can create visual process maps, and LLMs can assemble technical summaries and dev docs from your solution proposals and other documents.
  
• Support your end users. Create an agent that functions as a first tier of support for your users when they have troubles or requests. When empowered by documentation that covers how your Salesforce environment works, this on-demand service can save your Salesforce team from dealing with a lot of low-level help requests. You can also configure the agent to take feature requests and automatically submit them to your backlog system.
  

Stay in the Loop

Remember, AI is a tool and it can’t replace human thought and ingenuity. Even autonomous agents need to be monitored. Make sure that someone on the CoE reviews everything generated by AI or agents. Feel free to override any decisions AI makes by applying your own judgment.

Don’t forget: The path to a high-performing Salesforce implementation isn’t found in a single feature, tool, or piece of code. It’s found in the excellence of your people and your processes. Happy governing!

Resources

• Salesforce Architecture Center: Well-Architected Framework Guides