Regardless of if it is an enterprise solution or a SaaS product, the challenge is not finding AI ideas because there are plenty of them. The challenge is identifying which ideas are actually worth investing in.

The mistake that costs the most

When AI initiatives begin from the technology, “let’s use LLMs,” “we should build an agent,” “we need a copilot,” the project inherits a problem before it has even started. 

It’s a solution looking for a fit.

This pattern repeats across many enterprises. A confident kickoff. A polished demo. Budgets approved. Twelve months later, the model works, the integration works, and almost nobody uses it in their actual day-to-day work. Companies end up with an expensive technology showcase.

The fix isn’t to add more enthusiasm or have better change management. It’s structural.

You change where you start.

Start with the workflow, not the technology

AI is a technology. Like any technology, it only creates value when it enables something useful.

That means the starting point should not be the model, the tool, or the latest AI capability. It should be the workflow.

That workflow might be an internal operational process, a product journey, a customer journey, or any other process that directly affects how the company creates value and delivers it to users or customers.

To get to the workflow we start with structured and guided interviews with real users, customers, or employees. We map that workflow step by step. What we’re looking for is what people are trying to achieve, where they get stuck, what activities they perform manually, where they lose confidence and where they create some workarounds.

This is where real opportunities begin to appear.

Identify the friction points

Once the entire workflow is mapped out, the next step is to look for friction.

• Where do users wait?
• Where do they repeat themselves?
• Where do they search for information that should already be in front of them?
• Where do they make difficult decisions under uncertainty?
• Where does manual effort pile up because no one has had time to fix it?

This is the foundation of our ProductOpp™ methodology. Opportunity discovery starts with the workflow, the actual problem, and the user’s needs. Not with the technology.

Mapping friction points to AI capabilities

Next, we map friction points to AI capabilities.

The table below gives a good overview where AI is really good at when it comes to solving certain friction points.

Friction type	How can AI help	Potential value
Repetitive manual work	Automating routine tasks, data entry, document processing, or follow-up actions	Reduces manual effort, saves time, and lowers operational cost
Complex decisions	Supporting users with copilots, recommendations, or decision-support tools	Helps users make faster, more consistent, and better-informed decisions
Large information sets	Searching, summarizing, comparing, or extracting relevant information	Makes knowledge easier to access and reduces time spent looking for answers
Patterns hidden in data	Identifying trends, predicting outcomes, or recommending next best actions	Improves planning, prioritization, and personalization
User uncertainty	Guiding users through options, explaining trade-offs, or recommending next steps	Reduces hesitation and helps users move forward with more confidence

Not every use case is an opportunity

An AI assistant that summarizes customer support tickets may sound useful. But if agents already scan tickets quickly and the real bottleneck is internal approval, the assistant will not move the metric that matters.

A real AI opportunity needs several things going for it:

1. A clear user or operational need — someone is actually struggling, today, in a specific way. We should be able to clearly see that from the workflow.
2. Measurable business value — you can name the metric that should move (e.g. conversion increases as a result of more personalized offers)
3. Available, usable data — the model has something to work with (e.g. historical data of interactions with the customer)
4. A realistic integration path — it can plug into existing systems without “rebuilding the company”
5. A reason users would adopt it — it’s faster, easier, or more trustworthy than what they do today
6. A strategic fit — Often overlooked, the company must actually want this part of the experience or operation to be handled by AI. Some use cases may be technically possible, but still wrong strategically.

When you apply these filters, many AI ideas fall away quickly. They may be interesting, but they are not strong enough to justify investment.

The opportunities that remain are the ones worth assessing more seriously. That is where value and feasibility scoring helps.

Prioritize by value and feasibility

Once we have a shortlist of real opportunities, the next step is to prioritize them.

We usually assess each opportunity across several dimensions. Because it is impossible to assess everything with perfect accuracy at this stage, we use a simple scoring model from 1 to 5.

Typical dimensions include:

Dimension	Description	High score example
Business value	The expected impact on business KPIs and strategic goals.	Direct impact on revenue or major cost reduction
User value	How much the solution removes real pain or creates a significant lift for the user.	Solves a critical daily blocker for many users
Data readiness	Availability, cleanliness, and accessibility of required data sets.	Clean, usable data is already available
Integration complexity	The degree of technical change required in the current stack.	Simple integration into existing systems
Cost and operational effort	Total resources required to build, run, and support the solution.	Low effort, clear quick win
Trust, risk, and adoption	Likelihood of users trusting the output and adopting the new workflow.	Users already want help here, and risk is low.
Strategic fit	How well the opportunity supports the company’s strategic direction and desired role of AI.	Strongly aligned with strategic priorities

This exercise usually creates immediate clarity.

Some ideas that sounded exciting fall down the list because they do not create enough value for users or the business. Others move up because the pain is real, the business value is clear, and the data is already available. These are the ones we are looking for.

Prototype and validate before building

The most expensive way to validate an AI opportunity is to ship it. The smartest is to prototype it.

Before committing serious budget or engineering capacity, we create a mocked-up version of the AI-assisted flow and put it in front of real users.

At this stage, we are not only testing whether people like the idea. We are testing more practical questions.

• Do they understand what the AI is doing?
• Do they trust the output?
• Does it help them complete the task or make a better decision?
• Does it fit naturally into the way they work?
• Would they use it again tomorrow without being asked?

This is where most assumptions break. Users don’t trust the output. Or they trust it, but the workflow change is too disruptive. Sometimes they feel they do not have enough control. Sometimes the AI solves part of the problem, but not the part that matters most.

We have a dedicated article on how to design AI assisted experiences that users trust and use.

Build the business case

At the end of our process, leadership does not receive a long list of AI ideas. They receive a validated set of AI opportunities, shaped into clear business cases.

For each opportunity, the team understands:

• What problem it solves
• Who it is for
• Why users would adopt it
• What business value it can create
• What data and integrations are needed
• What needs to be designed and built
• How long it could take
• What budget it may require
• What ROI can reasonably be expected

This gives leadership more than an answer to “Where can we use AI?”

It helps them understand which AI opportunities are worth building, what it would take to build them, and whether the investment makes sense.

Summary

Doing AI well is not about ambition or speed. It is about the order of operations.

Start with the workflow. Find the friction. Match the friction to what AI is genuinely good at. Filter for real opportunities, not use cases. Prioritize honestly. Prototype before committing. Then turn the strongest opportunities into business cases, with the requirements, budget, timeline, and expected ROI needed to make a confident investment decision.

The pressure to “do something with AI” will not go away.

But it becomes much less dangerous once you have a clear way to decide what is worth doing, what it would take to build, and whether the investment makes sense.