Finding the Right Problems to Solve with AI

Introduction

AI is everywhere right now. New models, agents, and tools appear almost daily, each promising to transform how companies work. But when teams come to us asking for help with AI, the challenge is rarely how to build something; instead, the question is what to build, and whether AI is even the right tool in the first place.

I was listening to an episode of the Super Data Podcast where Jon Krohn interviewed John Roese, Global CTO and Chief AI Officer at Dell, and one moment stuck with me. They were discussing why so many AI projects fail to move the needle inside organizations, and Roese put it succinctly: in many cases, AI is being used to automate work that was essentially BS to begin with.

That observation mirrors what we see in practice. You can apply AI to almost anything, but if the underlying work or data doesn’t materially matter to the business, making it faster or cheaper won’t suddenly create value.

This article walks through a practical way to identify high-impact opportunities for AI, and how to avoid wasted effort, hidden risks, and unnecessary complexity.

Don’t Automate BS Work

The idea is simple but powerful: before you automate a process, it’s worth asking whether that process should exist at all.

AI can make bad processes run faster. But that often just means you’re producing low-value outcomes more efficiently.

Before reaching for AI, ask:

• If this task disappeared tomorrow, would anyone outside the company notice?
• Does improving this actually change customer outcomes, revenue, or risk?
• Are we optimizing for internal optics rather than real impact?

This kind of upfront discipline is unglamorous, but it’s one of the strongest predictors of whether an AI initiative will matter. The best AI projects we’ve seen don’t start by picking a model or framework. They start by ruthlessly questioning whether the work itself is worth doing.

Start with Friction, Not Technology

A common mistake is starting with a solution:

“We should use AI to do something with customer support.”

Instead, start with friction:

• Where do teams spend time on repetitive or manual work?
• Where do errors regularly creep in?
• Which processes depend heavily on human judgment, but follow recognizable patterns?

Good AI candidates often live in spreadsheets, inboxes, internal tools, and Slack threads.

Examples:

• Support agents rewriting similar responses all day
• Operations teams manually reviewing documents or images
• Engineers babysitting flaky alerts or noisy dashboards
• Sales teams repeatedly updating CRM records, summarizing calls, or chasing down context spread across emails and notes
• Customer success teams manually triaging tickets, identifying churn risk from scattered signals, or preparing account summaries before check-ins
• Marketing teams repurposing the same content across channels with slight variations
• Finance or ops teams reconciling data between systems that were never designed to talk to each other

If a task is already painful, slow, or expensive, that’s a signal worth investigating.

Look for Decisions, Not Just Automation

AI isn’t only about replacing tasks. Some of the best use cases support better decisions.

Ask questions like:

• Where are we making decisions with incomplete or delayed information?
• Where do teams rely on gut feeling because data is hard to interpret?
• Where does context live across too many systems?

AI systems can summarize, classify, surface patterns, and flag outliers. This can often be done without being fully autonomous. These “decision support” use cases tend to be lower risk and easier to validate.

Evaluate Cost, Risk, and Reliability Early

Not every problem should be solved with AI. That’s okay.

Before committing, it’s worth testing potential use cases:

• Cost: Does the value outweigh model and infrastructure costs at scale?
• Latency: Does this need to be real-time, or can it be asynchronous?
• Risk: What happens when the system is wrong?
• Compliance: Are there privacy, security, or copyright concerns?
• Maintenance: Do you have enough team members or an agency to keep the system running and up to date?

Auditing prompts, pipelines, and model behavior early can surface these issues before they become production incidents.

Prefer Small, Measurable Wins

The most sustainable AI efforts tend to start small:

• A faster internal workflow
• A clearer signal in noisy data
• A cheaper or more reliable existing system

These projects are easier to test, easier to roll back, and easier to learn from. They also create momentum and trust, both critical for more ambitious AI work later.

Final Thoughts

AI doesn’t need to be complicated to be valuable. The hardest part is often choosing the right problem to solve.

Start with real friction, the kind people complain about in meetings or work around with makeshift solutions. Question whether the underlying work matters.

Focus on measurable improvements. And don’t be afraid to walk away from an AI project if a simpler solution makes more sense.

By starting with real friction, auditing what you already have, and focusing on measurable improvements, teams can use AI in ways that are practical, responsible, and effective.

If you’re unsure where AI fits into your systems, our team at OmbuLabs.ai opens a new window can help. Let’s talk opens a new window . 🤖