Jason Rivera, CTO, Verra Mobility, says companies rushing to adopt AI without identifying clear business problems risk creating an expensive innovation gap. He outlines a structured framework for evaluating AI applications to ensure real-world impact.
In boardrooms around the world, executives are demanding to know not only what their company’s AI strategy is but also how AI will actually make money. Despite significant global investment in AI technologies, McKinsey’s 2025 State of AI report reveals that “more than 80% of respondents say their organizations aren’t seeing a tangible impact on enterprise EBIT from their use of gen AI.”
This reality exists even as Beth Kindig, CEO and Lead Tech Analyst for the I/O Fund, projects that Big Tech could spend around US$240 billion on AI in 2025.
As technology leaders, we face mounting pressure to implement AI solutions regardless of whether they address actual business problems. This disconnect is creating an expensive innovation gap where companies invest significant resources in AI initiatives without clear returns.
After nearly a decade as CTO and Vice President of Technology Development at Verra Mobility and from my previous experience at companies like 3M, I’ve observed that successful technology implementations begin with understanding customer problems – not with the technology itself.
The problem with current AI approaches
Reversing the innovation process
When there’s such a big technology innovation and we’ve seen it with cloud computing, mobile devices and now AI, organizations can latch on to a new tool without delving first into what they’re trying to accomplish. Instead of asking ‘What problems need solving?’ they ask ‘How can we use AI to solve problems?’
This technology-first mindset leads to what I call ‘weaponizing’ AI – forcing Artificial Intelligence into business processes without clear purpose or value.
Learning from past technology cycles
This approach resembles past technology cycles. At 3M, we developed prismatic technologies, road paint and highway sheeting based on specific safety and visibility problems we identified. The innovations that succeeded weren’t just novel technologies; they addressed specific market needs with measurable outcomes.
Creating new paint technologies that improved visibility was the scientific breakthrough but finding how this technology could improve public safety became the business solution that delivered ROI.
The distinction between invention and innovation is crucial here. Inventions create new technologies, while innovations solve real problems. AI without clear application remains merely an invention – and often an expensive one. When AI exploration and deployment can cost millions, companies need to make sure these investments address genuine business challenges rather than just model ‘what if’ scenarios.
The discovery gap
Risk Tolerance and Success Metrics
The biggest challenge to effective AI implementation isn’t the technology itself but a failure to adequately define the problem. Many companies fail to properly identify and validate problems before investing in solutions. This discovery gap leads to expensive failures when sophisticated technology doesn’t align with actual business needs.
Risk tolerance for innovation varies significantly between organizations. Large enterprises can often afford a ‘hit-on-three-of-ten’ innovation approach, tolerating a 70% failure rate on new initiatives, because they have the customer base and scale that can make a 30% success into a win. But smaller companies need a success rate closer to 80-90% to justify their investments – they don’t have the budget to waste. Without a robust discovery process, hitting all of these goals is nearly impossible.
Rushing to implementation
The pressure from boards and executives to demonstrate AI capabilities often compresses the discovery timeline. This rushed approach leads to force-fitting AI into situations where simpler, less expensive alternatives might actually work better. As I often remind my team, sometimes hiring ‘a couple of kids out of college’ to solve a problem manually costs far less than a sophisticated AI implementation – and might deliver results faster with less risk.
A framework for AI evaluation
To bridge the discovery gap, technology leaders need a structured framework for evaluating potential AI applications:
1. Start with the problem
Document specific industry challenges and customer pain points before considering technological solutions. Consider what problems keep your customers up at night and which inefficiencies cost your organization the most.
2. Identify blind spots
Look beyond obvious challenges to discover problems your customers don’t yet recognize. As Steve Jobs famously noted, “a lot of times, people don’t know what they want until you show it to them.” So think about what customers will want next, not what they’re asking for today.
3. Assess disruption risk
Consider how competitors might use emerging technologies to gain advantages in your market. Identify which processes in your industry are most vulnerable to AI-driven disruption and how automation could streamline a process significantly. Often, this may be something ‘boring’ like invoice processing or managing international trade logistics, but automating these mundane processes can actually deliver significant savings.
4. Evaluate data assets
AI effectiveness depends heavily on data quality and quantity. Assess whether you have sufficient data for your proposed application and whether that data is clean enough to yield reliable insights. Often, we find organizations have abundant data but haven’t structured it in ways that make it usable for AI applications.
5. Match solutions to problems
Only after completing the previous steps should you assess whether AI is the right solution. Sometimes traditional software or even process changes deliver better results without the complexity and expense of AI.
6. Calculate ROI realistically
Include all costs – development, deployment, maintenance and organizational change management – when calculating potential returns from AI initiatives. This framework helps ensure that AI implementations solve real problems with appropriate technology and generate tangible business value rather than just creating technological showcases.
Real-World Applications Worth Exploring
At Verra Mobility, we’ve identified several high-potential AI applications by focusing on specific industry problems:
- Predictive intersection management: Using our extensive database of traffic violations and environmental factors, we’re exploring how AI can predict when violations are likely to occur. Instead of simply issuing citations after incidents happen, we aim to integrate with connected vehicles to provide warnings before dangerous situations develop, potentially saving lives through prevention rather than punishment.
- Distracted driving detection: In-vehicle cameras combined with AI can monitor driver attention and behaviour patterns. This technology could significantly reduce accidents while enabling insurance companies and fleet operators to offer incentives for safe driving behaviours, creating a positive feedback loop for road safety.
- Preventive vehicle maintenance: The wealth of data collected from connected vehicles provides early indicators of potential mechanical issues. AI can analyse this information to predict failures before they occur, reducing downtime and maintenance costs for fleet operators while extending vehicle lifespans.
- Virtual corridor management: Rather than investing hundreds of millions in physical toll infrastructure, AI-powered systems could enable dynamic road pricing using connected vehicle data. This approach could transform how cities manage traffic congestion without expensive hardware deployments, making smart transportation accessible to more communities.
Each of these applications addresses specific industry problems where the value proposition is clear and measurable, demonstrating the power of problem-first thinking.
Conclusion
Before investing in AI, ensure you’ve conducted thorough discovery work to identify and validate real business problems. The most successful AI implementations start with customer needs, not technology capabilities. By developing a robust process for problem identification and verification, you can significantly increase your success rate with AI initiatives and deliver meaningful value to your organization.
As AI capabilities continue to evolve at an unprecedented pace, the discipline of rigorous problem discovery will become even more critical – separating organizations that generate real value from those merely chasing the next shiny technology. The question shouldn’t be ‘How can we use AI?’ but rather ‘What problems do we need to solve, and is AI the right tool to address them?’
