AI As A Value Creator: How To Tell The ROI Story

Last week, Google announced slowing cloud revenue growth compared to Microsoft, which is realizing continued momentum. What’s the difference? Google saw customers reducing spend by optimizing their workloads. Microsoft saw AI workloads contribute significantly to Azure’s overall growth.

This week, I was part of an event and shared a story from my first client. How did a tiny, less than a-year-old consulting company win our first client? I discussed the business’s challenges and how solutions would deliver value by addressing them. I initially targeted marketing use cases, but supply chain leaders were more receptive. I pivoted and presented options aligned with supplier discovery use cases because that’s what leadership focused on.

Companies that are succeeding with AI don’t tell a technology story. They talk about how AI will create and deliver value. Look at these two pages:

Microsoft Copilot.

Look at the cheek and swagger on them. Microsoft is so confident that customers are buying Copilot that the page dives straight into resources to help prepare for implementation and adoption. I aspire to be that confident someday. Further down the page, Microsoft gets into use cases and capabilities.

Microsoft is still very much a technology-first business with many customers on the leading edge of the adoption curve. Businesses show up prequalified and ready to buy, but Microsoft still begins with those customers’ needs (accelerating and supporting adoption). The company continues with explanations of capabilities and use cases to educate and bring new customers into the early adopter category.

SAP Joule.

SAP knows its customers are working to identify use cases for AI. Most are looking for efficient growth and learning how AI can deliver it. The breakdown is by functional area and covers solutions to everyday business challenges. It’s a high-level explanation of the types of problems AI solves for businesses and the value it can deliver. Right behind it, SAP backs it up with case studies and results.

Contrast those with Google’s Generative AI landing page. Google leaves the use cases, ROI calculation, and adoption up to customers. It’s an example of technology-first thinking vs. business-first.

For data teams working to establish themselves as value creators vs. cost centers, creating a landing page like SAP’s is a good idea. Break out use cases by functional area and discuss how technology creates and delivers value. Start with core strategic objectives and list a few opportunities for data and AI to support them. List early successes and the returns or impacts they delivered.

It’s an effective approach to telling the ROI story instead of the technology story. Here’s how I would frame it for the Procurement and Supply Chain domains.

Business First AI

Procurement is forward-looking by nature. Data, analytics, and more advanced models can deliver visibility into what’s happening now and what’s likely to happen next. There’s a strong alignment between how procurement creates value and the kind of value creation data and AI supports.

Procurement has complex conditional workflows with dependencies and the need for adherence to standards. There are opportunities to build on the value that digital solutions are already delivering. Some conditional workflows can be handled by straightforward logic, and those play to digital technology’s strengths.

Could we replace those with data and AI-supported solutions? Sure, but just because we could implement the technology doesn’t mean there’s value created by doing it. I like Peter Thiel’s rule for replacing an existing solution with a new technology. It must be 10 times better than what is in the market now. New technology can provide incremental improvements, but those aren’t the best use cases to target.

We should target workflows that are too complex to be supported by digital solutions. Data and AI’s core strengths are their ability to manage complexity and reduce uncertainty better than other technologies. Processes are often too complex for an individual to have complete visibility into the execution. It isn’t easy to:

• See inefficiencies.
• Discover root causes.
• Enumerate all optimization options.
• Understand the upstream and downstream impacts of each potential resolution.

These are the best use cases for data and AI because the value creation potential is the highest. Procurement workflows are highly collaborative and can touch multiple business units. Giving each team more complete solutions (solutions that handle more of the workflow than before) creates value across the business.

Procurement workflows generate a significant amount of data and require high degrees of traceability. Again, these can exceed an individual’s ability to track progress, identify bottlenecks, and see the need for a human’s attention. In my last article, I explained a granular example from the healthcare domain. Procurement has dozens of workflows that follow the same paradigm.

When I ask, “What data do you need access to?” supply chain and procurement domain experts come back with answers like, “All of it. And not just the business’s internal data. I need partners’ and vendors’ data, too.” People in complex roles are used to doing long-chain workflows independently and need to track all the moving parts without support. Data and AI-supported tools can change that and provide assistance that digital tools can’t.

Solutions Require More Than 1 Type Of AI

Business leaders need to be reminded that Generative AI isn’t ‘The AI.’ Leveraging Generative AI for everything doesn’t play to its strengths. Generative AI can handle anomaly detection in contracts. Great. There’s value in automating that piece of the workflow, but the workflow doesn’t stop once the issue is discovered. What do you do once the anomaly is detected?

Generative AI alone can’t address the complete workflow. Prescriptive models can recommend resolution paths. Integrating Generative Interfaces with traditional applications leverages its orchestration capabilities. Once the user decides which resolution path is the best, the Generative Interface can orchestrate the workflow across multiple apps.

We worry about model reliability. It’s less of an issue when people maintain autonomy over their workflows, with models augmenting them rather than taking over. In the contract use case, the models support a user, but the person maintains full autonomy over decision-making. The productivity gains are delivered by the models working behind the scenes to accelerate steps. Reliability requirements are much lower in this paradigm than in the full-automation approach.

The value creation challenge for many solutions is a disconnect between Generative AI (or any AI) and everything else. For Generative AI to deliver application orchestration, it must be integrated into those apps, or that integration becomes a barrier to adoption. It must be integrated with data sources, or that integration becomes a barrier to adoption.

That’s the Microsoft Copilot and SAP Joule paradigm. Copilot is integrated into applications like 365 and Bing Enterprise Search. Joule is integrated into SAP’s suite of applications and has access to all their capabilities. SAP has a portfolio of data sources and models that Joule can collaborate with, opening up new ways to create value.

Today, we should be focused on incrementally delivering value in the short term but also setting up for more extensive, long-term opportunities as technologies mature. Significant new functionality will be supported by multi-agent systems built on several Generative AI models and predictive, prescriptive, and diagnostic machine learning models. It will feel like a single interface to users because the orchestration will be managed for them.

How will models work together to manage more complex workflows?

Generative AI auto-creates personalized workflow templates based on the business’s existing processes.

Predictive models can call out potential exceptions, issues, or negative downstream impacts that the new workflow creates.

Prescriptive models can recommend fixes and optimizations to address those challenges when solutions aren’t obvious.

Diagnostic models can quickly identify root causes of exceptions, issues, and negative up or downstream impacts. The business addresses the cause vs. symptoms and spends less time iterating through solutions.

Generative Interfaces manage application orchestration so fixes, optimizations, and improvements can be deployed faster. Accessing the right apps is as easy as telling the intelligent assistant what you want to do.

Creating A Value-Centric AI Strategy And Vision

One of the main reasons I’m partnered with SAP is the value and business first story it is telling about AI. The business has a short-term roadmap that is being delivered over the next 12 months and a long-term vision. The features SAP is shipping today set up for multi-model, multi-agent implementations tomorrow. Instead of talking about the roadmap that way, SAP has chosen to tell the story of how AI meets business needs.

Hype-cycles are driven by new technology and its potential. Adoption is driven by value creation and delivery. We need both sides, but hype must quickly be replaced by value and impact as the bubble bursts. This year, we had the opportunity to watch the early successes and failures of companies working to pivot from hype to value.

Data teams and business leaders should also think about how to make the pivot. It won’t happen all at once. Transformation is continuous, so it must also be incremental. It must deliver value quarterly with a vision for long-term growth that aligns the incremental initiatives.

Data teams should start conversations with business needs and current strategic goals. We need to discuss value externally and technology internally. We should be focused on educating and helping the business select high-value opportunities over incremental ones.

Business leaders should consider how data and AI will change the business and operating models. Transformation is enterprise-wide, and adoption requires prepared, solutions-literate users. As overconfident as Microsoft’s presentation feels, they support one of the biggest challenges with Generative AI. Users must upskill to get the full value from Copilot or any other Generative Interface.

It's a collaborative process with opportunities for developers and non-developers, data scientists and non-data scientists, to build solutions collaboratively. That’s another piece that Generative AI will enable, and platforms must support a new type of technical solutions development environment.

It’s short-sighted to think about the future of technical solutions development belonging to the technology organization alone. Operating models must be ready to adapt to a time when anyone in the business can deliver technology to meet their own or customer needs. A siloed approach won’t survive larger disruptions to come. Strategy and value are the common languages that can align the business.