Best for: Gaining AI experience with speed and minimal investment
Risk level: Minimal
Interdependencies: Virtually none

Begin by working with your data team to deepen your understanding of your data assets. Seek opportunities for your data to provide patterns and insights not previously possible through machine learning.

Choose problem spaces where your organization already has strong domain expertise, so you can tell if the experiments are working.

After each experiment, include an organized reflection mechanism to capture learnings and develop new questions for the next experiments (After Action Reviews, postmortems, retrospectives, etc.).

It can be tempting to treat experiments as a secondary or side effort, causing you to underutilize your leadership skills. Don't do this. Especially because they don't have a typical business case and typically have negative ROI from an immediate revenue perspective, it's critical to inspire and motivate the organization around the learning goals and positive steps the organization is taking into this new AI era.

An asset management firm recognized they possessed data on thousands of fund managers and millions of investment data points across clients over time. Their unique approach to structuring client portfolios and organizing data provided potential differentiation.

Senior portfolio managers, directors, and heads of research worked with data teams and their head of innovation to collectively come up with questions their data may help them answer in new or faster ways. They used machine learning to run experiments seeking patterns and insights about portfolio performance across various factors. Their seasoned Portfolio Managers, not beholden to experimental outcomes, proved instrumental in evaluating effort quality.

Best for: Achieving customer benefit with speed and manageable budget
Risk level: Low
Interdependencies: Minimal impact on surrounding features

Start by analyzing existing user flows for individual features that have potential to be enhanced with better data insights (ML, AI, etc.).

Focus on features you can change without breaking other parts. Look for challenges that address genuine user pain points, could be implemented without interrupting existing flows, and leverage existing data.

What to avoid at this level: demanding comprehensive organizational AI strategy encompassing every product and feature, creating detailed ROI calculations requiring board-level approval, or forcing integration with other features or multi-feature releases.

Shopify's product team—led by Miqdad Jaffer, Head of Product at the time—identified a common problem: many items for sale on the platform lacked descriptions, creating pain points for both sellers and buyers. It promised to be a great problem for AI enhancement. It was a pain point for both sellers and buyers. It could be addressed without interrupting existing flows. And it could leverage data they have, gathered from millions of items on their platform over the years.

They worked with data engineers to see how well the AI system could generate descriptions and then manually assessed hundreds of samples. Through design, prototyping, and testing with actual sellers, they built confidence to pilot with select sellers before rolling out platform-wide. Today when the platform uses AI to come up with an item description, it is a suggested description. The seller decides to approve, modify, or reject the description.

Instead, they maintained sharp focus, allowing them to realize success quickly with minimal risk and cost while generating valuable AI experience. Allow your goal to be as simple as neutral or better for the customer. Remember you're still learning and experimenting.

Best for: Transforming entire user journeys without impacting other product areas
Risk level: Moderate
Interdependencies: Contained within individual flows

Level 3 sits between enhancing individual features and bigger transformations. This level makes sense when you can identify complete user journeys that can be redesigned without disrupting other journeys in your product.

To identify good Level 3 candidates, look for flows with clear start and end points, minimal connections to other product areas, and measurable user outcomes. Look for situations where data can be brought to the solution in new ways that weren't previously possible.

Resource-wise, expect Level 3 projects to require cross-functional collaboration between product, design, engineering, and data science teams, with timelines typically spanning quarters rather than weeks. Stakeholders will need to understand that you're reimagining workflows, not just adding features, which requires patience during the design and testing phases.

What to avoid at Level 3: choosing flows with extensive dependencies on other systems, or attempting to redesign too many flows simultaneously without learning from the first.

Legal professionals preparing cases inevitably work with massive amounts of documents—millions of electronic records, communications, briefs from prior cases, precedents, discovery materials, and more. Previously, legal software empowered efficient document uploads so that teams of legal professionals could access and manually analyze them.

A leading legal technology company recognized AI's capability to extract concepts and summarize information from documents, allowing them to redesign the entire flow. The new experience transformed document upload from a simple efficient upload and organization function into an intelligent analysis system that extracts concepts, generates summaries, and identifies interconnections and insights.

Best for: Addressing customer problems that span multiple existing product flows
Risk level: High
Interdependencies: Significant impact across multiple product areas

When you examine your existing products, it doesn't take long to find a user journey that works, but perhaps isn't set up as well as it could be. The user might have to move between products to complete their task. Or there may be duplicative ways to accomplish a task, when one way would be better.

There are many understandable reasons for this: legacy technical architecture, company growth by acquisition, and more.

Where Levels 1-3 focus on isolated and discrete areas of your product to allow for delivery without excessive dependencies, Level 4 solutions aren't as tidy. They'll impact more than one existing user journey. This doesn't mean you have to redesign the entire product suite, but you will have dependencies and interconnections to design o.

What this typically means in practice is your team gets to create more boldly and divergently, grounding their efforts with individual user goals, or Jobs To Be Done. It's less constrained than Levels 1-3, and more constrained than Level 5, where you envision entirely new solutions.

An e-commerce platform has separate tools for inventory management, sales analytics, marketing campaigns, and supplier coordination. The team identified one Job to Be Done that could be significantly enhanced with AI: helping merchants know what to stock and when.

The Job to Be Done: "Help me stock the right products at the right time so I don't run out of popular items or get stuck with inventory I can't sell."

Previously, merchants had to manually check sales data in one system, review marketing campaign performance in another, look at seasonal trends in a third tool, and coordinate with suppliers through email or a separate system. A merchant might have seen that a product was selling well in their sales analytics, but missed that their marketing team had just launched a campaign that would drive 3x more demand, or failed to notice that their supplier had a lead time issue that would create stockouts.

The AI solution develops an inventory advisor that connects data from sales, marketing, supplier systems, and external factors like weather, trends, and seasonality to predict optimal inventory levels. The AI identifies patterns like "winter coat sales spike 2 weeks after the first cold snap, but only if we have active social media campaigns running, and Supplier A has a 3-week lead time while Supplier B has 1 week." It provides specific stocking recommendations and automatically flags when marketing campaigns might outpace inventory availability.

This crosses multiple existing product flows—inventory management, marketing planning, supplier coordination, sales forecasting—but focuses on solving one specific job rather than reimagining the entire merchant experience.

AI integration works when you balance big ideas with practical action. Each level provides learning opportunities that inform and improve subsequent efforts, but the key lies not in perfect planning but in purposeful action that builds organizational capability while delivering genuine value to your customers.