Salesforce CIO: How agents change internal workflows

Salesforce CIO and executive vice president Juan Perez has been with Salesforce for three years, following a 32-year stint at UPS as CIO and chief of engineering.

We talked about this and much more in this Q&A -- such as how does a low-code platform vendor like Salesforce approach DevOps governance, given that it's committed to using its own products, such as generative AI agents?

The previous CIO of Salesforce was responsible for keeping trains on time in her prior position. You were responsible for keeping packages on time. What experience from UPS did you translate to your job at Salesforce?

Perez: A couple of things come to mind. The first is the importance of having an IT organization that can support [a large organization] at scale. That's one of the primary reasons why Salesforce was interested in me. At UPS, I had to manage a large IT organization that supported technology for the entire company all around the world -- a very large Fortune 50 company -- with lots of complexity, multiple business units from running an airline all the way to making sure that packages get delivered on time.

Second, Salesforce very intelligently realized that as we grow the business, we need to have state-of-the art technologies -- Salesforce technologies first, and then everything else in position to be able to drive appropriate support for growth in the company. That includes having a solid data strategy, that includes the modernization of a number of technologies in the enterprise; that includes having the appropriate IT operating model to support the company's needs. And with AI, that now includes making sure that we have an IT organization that can support the implementation of it.

You've said that generative AI looks to be as big as the internet was when it first emerged in the 1990s on a large scale -- you're not the only one saying this, other vendors and tech leaders such as journalist Kara Swisher at The New York Times are, too. Why?

Perez: I have been involved in so many different products where we wanted to do a number of things with technology, and it was not really there for us to be able to generate value for the organization. I believe generative AI has the ability to reduce a lot of the time that people are spending looking at all reports and data to create insights and take action. What that means is, in the end, we will have more time to take appropriate action on improving customer service operations, improving productivity, improving efficiency.

Gen AI is not going to do all those things. Gen AI will be an enabler to allow the [people] to do those things more effectively, spend more time on the things that will make the business better.

You said earlier Salesforce tech takes the front seat in your IT stack -- that's a given. How do you evaluate and implement things like LLMs and other non-Salesforce generative AI tools in your tech stack?

Perez: Whatever you try, whatever technologies you develop, whatever technologies you ultimately implement at scale have to be aligned with the business strategy and where the business wants to go. Think of it as a filtering mechanism. My first filter when it comes to making these types of decisions is: Is this aligned with the company's strategy? Then the next question is: Is this something that can generate significant value for the organization? Can it improve service or quality? that passes to the next filter. The next filter is, do we have the technology capable of doing what we want to accomplish? If the answer is yes, then it goes to the last filter, and that is: Do we have either the partner that can support us with this, or do we have the skills ourselves to be able to do this? And if the answer to that is yes, then then we move on.

If the answer to that last question is no but yet the other three are yesses, then it's my responsibility to take action with my team and develop their skills or bring in the right partners to help us.

How does Salesforce measure return on investment for AI tools?

Perez: When GenAI came out, CEOs started putting a lot of pressure on CIOs to go and deploy AI. We all started putting AI solutions in place, and we didn't really go through the normal rigor that we normally go through to implement technology in the business.

I think that all is coming back to a good and appropriate state of implementing technology in the enterprise. I trust the AI Salesforce builds into its products, and I will take advantage of that trusted AI in my business processes. For other technologies we're starting to bring them back into a normal assessment process to ensure that what we're implementing is safe, works and can generate value.

We're also getting better at measuring the value of GenAI solutions. Just recently, we measured the numbers of lines of code that we moved to production, and in the last run that we went through, out of 60,000 lines of code that we promoted to production, roughly 26% of those lines were generated through our AI code generator. We measure that in my own CIO dashboards; I have now a view as to how we are adopting AI solutions in my group and in the enterprise. We're also starting to measure time saved and reallocated to other tasks and other work. I cannot use "26%" for everything -- that is not true, and it's not right, but I can use 26% for certain work that we've got to do next year. So, as we build our plans, our budgets for next year, I am counting on a certain amount of savings going into our budget associated with the use of AI that then I can allocate to other initiatives in the company.

You've warned about the "shadow AI," a variant of shadow IT. It's a fascinating idea, especially coming from you, the CIO of a company that makes low-code tools for line-of-business users to build generative AI agents. What does your GenAI governance look like?

Perez: For decades we have been dealing with the notion of "should we centralize all IT for an organization or not?" Should you allow for some decentralized IT out there? Should you manage all aspects of it in just one group, or should there be some democratization of it so that different groups can build some level of IT under the supervision of IT? Or do you let it all loose and see what happens?

When you have everything set loose, it becomes a major problem to manage data, manage security and organize your IT investments. You end up spending a ton of money on stuff that is not going to really provide huge value to the company.

I truly believe that with AI, it is no different. If we don't have some level of visibility as to what is going on with the implementation of AI in the enterprise, whether it's agents, whether it's just AI solutions like what you mentioned before, establishment of LLMs, contracts with third parties that provide AI, you can end up in a world where you have so many disparate and non-organized AI solutions. [That world is] a shadow AI world where the IT organization has no visibility as to how customer data is being used, how your company's data is being used, and what's in these models. How are you managing the privacy and the copyrights in these models? When that happens, this shadow AI can ultimately cause the organization significant pain in terms of extra cost, legal exposure, exposure to investors, exposure to customers, exposure to employees. And that's why I think it's important that there's an organized approach to managing and supporting AI in the enterprise.

The way that we have addressed that at Salesforce is we have an AI council. It is expanding into an Agentforce center of excellence, and that AI Council works with all the appropriate groups to ensure that there's visibility into what we're working on, the needs of the organization, support for the projects that matter most, and at the same time that we do it responsibly.

Don Fluckinger is a senior news writer for TechTarget Editorial. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.