AI With a Purpose: Are We Using the Right AI for the Right Problem?

Start With the Problem, Not the AI

In recent years, most discussions have centered on what AI can do. That approach made sense during the early stages of experimentation. Now, businesses need to prove that AI makes a difference, whether by reducing customer effort, improving first-contact resolution, boosting employee productivity, increasing revenue, raising retention, or lowering service costs.

This shift changes how we begin. Rather than asking, “Where can we use AI?” it’s better to ask, “What problem are we trying to solve?”

MIT Sloan highlighted this again in May 2026, cautioning against treating AI as just another tech project instead of starting with a clear business problem and the changes needed to solve it. Current CX research shows the same pattern: companies invest in AI but still struggle to show ROI, get their data ready, and connect the systems AI needs.

If you don’t define success before starting, it’s much harder to prove value later.

The Market Is Moving from AI Features to AI Outcomes

The CX market is changing. Genesys now focuses on contextual intelligence and orchestration. NiCE is blending conversational AI with its CXone platform. Salesforce is integrating Agentforce more deeply into customer and service workflows. ServiceNow is using AI for enterprise workflow automation, and companies like Microsoft, Google, and other AI vendors are adding agents across their platforms.

While products and systems vary, the trend is the same. AI is moving past just answering questions or summarizing conversations. Now, it’s about understanding context, coordinating work, and acting across different systems.

This shift makes the business problem even more important. Before adding another agent or automation, companies need to know what outcome they want to improve, what information is needed, what authority the AI should have, and how to measure success.

Not Every Problem Needs the Same AI

One challenge is that we often talk about “AI” as if it’s a single technology, but it’s not. Predictive AI can forecast contact volumes or suggest the next best action. Conversational AI handles routine questions. Generative AI can summarize interactions or help employees write responses. Agent assist brings up useful information while keeping people in control. Agentic AI can solve problems, use tools, and take actions across systems.

Each of these AI types solves different problems and comes with its own risks. For example, checking a balance doesn’t need as much autonomy or oversight as handling an insurance dispute, changing a financial account, or solving a complex healthcare issue.

Choosing the right AI also means thinking about what happens after deployment. When a policy, product, or process changes, who updates the AI? Who tests and approves the change, how quickly can it reach production, and how easily can it be rolled back if something goes wrong? A solution that looks impressive in a demonstration may be a poor operational fit if every small business change requires a lengthy engineering or vendor cycle.

The most advanced AI isn’t always the right answer. Sometimes a simple, rule-based process is best. In other cases, AI should help a person rather than take over. And sometimes, it’s better to fix the current process before adding any AI.

The Industry and Interaction Matter

The right amount of AI depends on the setting. In retail, it’s easy to automate inventory checks, order status, or store info. Banks might use AI to collect information and suggest actions but still require people to approve certain transactions. In healthcare, AI can summarize information or support staff, but people should remain involved in decisions that require clinical judgment.

A simple rule helps automate tasks that are predictable, repeatable, and low risk; use AI to assist when it can make someone better or faster; keep people in charge when judgment, accountability, empathy, or complexity matter.

These boundaries will shift as technology, regulations, and trust develop, but changes should be intentional. Transparency is also key. Customers want to know when they’re dealing with AI, and rules like the EU AI Act are making disclosure and accountability clearer.

AI Cannot Fix the Foundation

Sometimes, AI isn’t the issue that needs fixing. Many organizations still face fragmented data, disconnected apps, inconsistent processes, and legacy systems. Adding a smarter AI model or agent won’t make these problems go away.

Genesys’ 2026 State of Customer Experience research offers a useful example: 48% of companies surveyed still do not pass data from a virtual agent to a human agent. If customers give information to an automated system and then must repeat it after a transfer, that is fundamentally an integration and journey-design problem.

A more advanced language model won’t solve this. Using agentic AI on top of disconnected systems can make the tech look better without improving the customer experience.

This idea applies outside the contact centre too. AI needs context, good data, accessible systems, clear workflows, and strong governance. If these basics are weak, giving AI more freedom can make the problems worse instead of fixing them.

There is also a practical reason to look at the technology already in place before adding another AI platform. Organizations may already have years of integration work connecting their contact centre, CRM, data, and systems of record. A specialist AI agent may appear more capable on its own, but if it cannot easily reach the information and workflows needed to resolve the customer’s problem, that sophistication may not translate into a better outcome. Before buying another platform, the better question may be: what specific problem can our existing environment not solve?

Automation Is Not the Outcome

We might also need to change how we measure AI success. Metrics like containment, automation rates, average handle time, and agent numbers matter, but they don’t always show if the customer’s problem was solved.

An interaction might be “contained” but still leave the customer unhappy. Handle time can drop even as repeat contacts go up. An AI agent might finish a workflow successfully, but the result may not be what the customer wanted.

As AI starts to act instead of just giving answers, we should measure results more closely. Did we solve the problem? Did we make things easier for the customer? Did employees become more effective? Did business results improve? And was the value worth the cost and risk?

The same principle applies to testing. More platforms are using simulated conversations and AI-based evaluators to test other AI agents. That can make testing faster and broader, but it introduces another question: who defines what a successful interaction looks like? An automated score is only as useful as the criteria behind it. The organization still needs to define resolution, customer effort, policy compliance, safety, and business outcomes, and have people validate that the evaluation reflects what matters. These are better ways to measure AI maturity than simply counting how many copilots, bots, or agents a company uses.

AI With a Purpose

This isn’t an argument against AI. AI is quickly becoming one of the most important technologies in customer experience, and agentic AI could reshape how journeys and business processes are built and delivered. That’s why it’s so important to choose where and how to use it.

An AI strategy shouldn’t start with a model, an agent, or a vendor demo. It should start with a problem. Define the outcome, understand the customer journey, assess the risks, and decide where human judgment matters. Then pick the technology that fits.

Sometimes the answer will be predictive, generative, or agentic AI. Other times, it will be a person supported by AI. In some cases, a regular workflow or automation is enough. And sometimes, the best choice is not to use AI at all.

The goal isn’t to use more AI. It’s to use AI where it truly makes a difference.

Sources and Reference Material

MIT Sloan, “What Leaders Still Get Wrong About AI,” Beth Stackpole, 18 May 2026.

Genesys, “2026 State of Customer Experience: Global Insights for CX in the Agentic Era,” 2026.

Genesys, Xperience 2026 product and customer announcements, September 2026.

Regulation (EU) 2024/1689 (the AI Act), including Article 50 transparency obligations.

Building Agentic AI, “Best Voice AI and IVA Platforms for Contact Centers,” Muhammad Arbab, 14 September 2026.