Enterprise Agents Will Finally Deliver ROI For AI Investments

AI-based agents are very much having a moment right now. In recent months the topic has been pervasive across the events and marketing of many tech vendors, including Salesforce, Microsoft, ServiceNow, SAP, Amazon Web Services and Google. In many ways, the broad coverage and the promises about business impact are warranted: Agents represent the next evolution of generative AI capabilities and enable a more autonomous AI experience. Instead of just answering questions like AI chatbots do, agents can have tasks delegated to them by people or even other agents. It’s an exciting advance in AI technology, which is why I started writing about agents and their potential. And just like almost every aspect of AI, the technology is moving fast.

When I wrote my initial article on AI agents two months ago, the news cycle and product announcements were heavily focused on agents deployed within application platforms—with Salesforce serving as a key example. The purpose of these agents is to enable personal productivity and task automation within the context of the platform. So, for instance, consider a personal productivity agent that periodically scans multiple documents, summarizes the information in them and recommends actions to an end user. To stick with the Salesforce example, that might mean a quarterly scan of the prospect pipeline to see if there are new digital marketing opportunities and provide suggestions for tactics. These personal productivity agents provide some very helpful benefits to end users looking to get more done, more quickly.

The Trouble With Personal Productivity Agents

The problem is that while personal productivity agents are helpful, they are not always impactful. Personal productivity agents deployed in application platforms have been great in the sense that they have shown the overall potential of agents by leveling up an existing technology investment. And over time a percentage of these new agents are sure to deepen the connection between the users and the platform. But true business impact will be measured when a business sees a clear return on investment and spends net-new money on it, versus using something that’s included for free (or “free”) or that’s available for a slight incremental increase in price. And, sadly, I think personal productivity agents will not achieve that status, for three reasons.

1. Usage across teams drives business impact — It’s a good thing when a single employee finds a better way to do their job. But the biggest improvements come when that employee can reach out to other team members—and other functional groups—to improve an entire process. To put it another way, one or two developers writing code faster is nice, but it’s not likely to move up an application’s ship date.
2. Personal productivity improvements do not increase the number of power users or super users — While AI has great potential, we have seen other technical breakthroughs in personal productivity applications languish in the field. Everyone has had a spreadsheet application on their desktops for decades. Yet despite all of the improvements in spreadsheets over the years, has the ratio of power users to total users changed? The answer seems to be no. And I say “seems” since there is no clear way to define a power user and quantify their impact.
3. Personal tool-making technology is hard for IT to manage and govern — Over the years, increased personal productivity has been a benefit created by many tools. And vendors of technologies such as low- and no-code tools have touted the idea of individual empowerment by enabling people to make their own tools. While there are notable examples of this working for the individual, these tools also create many IT issues. Application sprawl, where many people create many apps that do sort of the same thing, is one key example. Another one is abandonment, where apps are just forgotten about and not used. The cost, governance and security issues created by these platforms can be a challenge for strapped IT teams.

Impactful Agents Will Be Enterprise-Focused

Again, personal productivity agents do have utility, but the generative AI ecosystems still need to provide more measurable business value to justify strategic investments. We are starting to see signs that this is a possibility. For example, I recently covered UiPath and how it is approaching enterprise agentic development to accompany its robotic process automation solution, ideally leading to more efficient business processes. By taking a different approach to governance and agentic development, UiPath provides more breadth in building cross-enterprise agentic applications.

In a slightly different context, consulting firms such as IBM are retooling how they deliver enterprise systems integration projects using their own enterprise-wide agent factories to reduce time-to-delivery and increase project quality. In researching these and other companies working on enterprise AI agents, I’ve found that, although the technology is different, the implementation still rests upon sound IT practices developed over the past 40 years. The key practices are:

1. Start with a meaningful cross-functional business problem — This means that the project needs to have at least one business metric that it impacts, for example deploying a technology that increases revenue yield without increasing the number of salespeople.
2. Place a high priority on change management — While there is no measure of the ratio of power users to all users, most IT executives say that fewer than 20% of users are power users. Therefore, introducing any technology advancement must either be invisible to average end users or have sufficient change management applied such that it is adopted widely. For example, it’s possible that agents may be very helpful in providing inline assistance for average users thanks to the core capabilities of large language models—and that they will be adopted widely for this reason.
3. Leverage a solid technology foundation — Achieving success requires a solid set of technologies to build upon. Agentic applications touch a lot of diverse data, so ensuring that the data is well secured is critical. Over time, infrastructure in the form of reliable compute and storage is also important, especially as you scale the application across the enterprise.
4. Deliver with a well-thought-out framework — For a project to be deemed successful, you need a set of clear goals to measure against throughout the project lifecycle. Those goals and measures should come from a project framework that is either internally developed or produced by a reputable third party. In addition to solutions such as Consulting Advantage from IBM, startups such as Zavvis are automating common functional roles (in Zavvis’s case, the CFO’s) as the framework.
5. Make it sticky and usable for years to come — Big-bang projects are a temporary salve for older harmful practices. Instead of focusing on solving for today’s issues, design the solution such that it can become a new operational process delivering value and preventing the risk of needing another big-bang project.

MORE FOR YOU
When Is Spotify Wrapped 2024 Coming Out? Here’s The Likely Release Date
Google Confirms New Gmail Security Surprise—And It’s So Simple
Samsung Warning—Do Not Install These Apps On Your Galaxy S24 Or S23

AI ROI = Enterprise Agents + Established IT Practices

Again, I am not trying to talk down about the first round of productivity agents. They do serve a purpose and honestly are a very good way to get stakeholders comfortable with AI. But when it comes to a leadership team making a real bet on a new technology I think the bar is—and should be—a bit higher.

The good news is that we are starting to see meaningful efforts from multiple cloud, enterprise and startup technology vendors. And in some cases we are also seeing these new AI agentic technologies being leveraged to accelerate aspects of the implementation. That said, while technology speeds some things up, we are also seeing that the process to get there remains very similar to the well-worn path of enterprise technology adoption.

In my next piece I’ll dig into some fresh examples of IT offerings coming into the marketplace that aim to achieve broader business results using agentic applications.