AI and Agentforce: Governance Strategies for Your First AI Project

There is a lot of excitement around artificial intelligence (AI) right now. New uses and new models are changing the way we think about how AI can support and improve the ways we work.

At Dreamforce ’24 one of the big announcements was Agentforce, a powerful companion to help Salesforce users resolve cases, answer questions, and drive productivity. But a lot of companies and organizations are struggling with how to even start using AI in a safe, trusted, and responsible way.

Like any digital transformation, a good place to start is with planning and preparation. During our Dreamforce session, “Transparency in AI,” Hayley Tuller and I discussed several steps to help admins prepare their teams, and their data, to support launching their first AI project.

Thank you to Hayley Tuller, Navigators co-founder, for co-authoring this article.

Data governance

When you think about the steps for designing a successful AI project, it’s really not that different from other technology projects. Preparation is key, and the first thing to consider is governance. Work with all of your stakeholders, internal and external, to either create or revise both a data governance and an AI governance plan. Generally speaking, a governance plan is a framework used to define your organization’s policies, procedures, roles, and tools for a specific aspect of your business. It outlines the stakeholders involved and can help ensure compliance and provide a structure for decision-making criteria.

If you don’t already have a clearly defined and documented data governance plan, that should be your first action. AI models are only as successful as the data they use, so if you have incomplete, inaccurate, duplicate, or otherwise “dirty” data, your AI outputs are likely to be less than useful and could cause more harm than good.

For example, if you have test data in your system but don’t have a clear, consistent way to identify and exclude those test records from your AI processing, you could end up with skewed results.

Another benefit of starting the process by defining a Data Governance plan and cleaning up your data is that most aspects of your organization can immediately benefit from the process, whether you proceed to launching AI or not!

AI governance

After you’ve created or updated your data governance plan, it’s time to address AI governance. Think of this as the guardrails and roadmap for how your organization will decide which use cases are appropriate, assign responsibilities, and ultimately execute, monitor, and adjust your AI tools.

While there are a lot of things that AI and Agentforce can do, it’s up to your governance team to decide what AI should do for you. Some things may be best left to a manual or human-driven process, but if you don’t have the governance structure in place, it will be a challenge to consistently evaluate uses.

If you have leaders pushing to launch a first AI project, but you don’t already have an AI governance plan, creating that plan IS your first AI project!

Steps for your first AI project

Now that you’ve made sure that the governance structures are in place to support data quality and appropriate use and processes for managing AI, it’s time to dive into your first use case. Like many other types of technology projects, there are some key tasks to consider.

Identify all stakeholders.
Explain AI governance structure.
Collect use cases and evaluate against AI decision matrix.
Complete a risk assessment.
Define a communication and monitoring plan.

Start by identifying your stakeholders and ensuring that everyone is familiar with the data and AI governance plans. It’s important to not only identify stakeholders but also assign responsibilities so everyone clearly knows their roles and what is expected of them. Then, following the steps you’ve defined in the AI governance plan, review several possible use cases and evaluate them based on your decision matrix and risk assessment process.

While it may be tempting to start with something big and flashy, it’s often more realistic to start with a smaller, contained, manageable project. This gives you a chance to test out the AI governance structure, get hands-on practice with the new tools, and validate the procedures in a lower-risk manner.

Risk analysis has two major considerations: severity and likelihood. A risk may be minor in scope, but if it’s very likely to happen or happen often, the overall impact could be great. Similarly, if a risk is unlikely to happen, but could be catastrophic if it did, overall impact would again be great. Once you’ve identified and evaluated the risks, define a mitigation plan to manage those risks.

Ideally, communication should occur throughout the project. Keeping your stakeholders informed goes a long way toward improving buy-in and ensuring better adoption. Define a regular cadence of communication and consider what messages should be shared through which channels. If obstacles or challenges arise, communicate early and often so no one is caught by surprise.

Once you’re ready to start building or configuring your AI tools, always work in a sandbox! Make sure you test for both positive and negative cases to ensure your AI action is doing what is expected and NOT doing what it shouldn’t.

Monitor and adjust

AI tools are not “set and forget” systems. Have a plan, tasks, and timelines for regularly monitoring the outputs of your AI models to ensure they continue to function as expected. Collect feedback from users to further refine your models. And be prepared to adjust or update your AI tools as data sources or business processes change.

Try this sample first project

One way you could launch your first AI project while also executing a data quality strategy is to use AI to identify data quality issues and work to improve the quality of your data. As you execute this project, collect feedback and suggestions for improvement that you can test out on your next project or subsequent iterations.

Final thoughts

Every age of new technical innovation starts with a phase of lofty aspirations and hype. Yet there is rarely a case where the new thing lives up to the imagined promises. AI is likely no different, but that doesn’t mean you shouldn’t adopt it. Instead of hoping for a miracle cure-all, use the buzz around AI to create real improvements in the core foundations of your organization: data, user engagement, and processes. Then, layer AI on top of those successes with confidence, trust, and transparency.