Most executives we talk to are grappling with the same challenge: How to use AI to actually deliver value at their organization.  

Here’s what’s holding them up: Most companies can tell you where revenue comes from in their business, but very few can quickly name the core value-generating processes – aka, the step-by-step work that actually creates margin.

When you can’t identify these processes, AI deployments become a scattershot of a thousand micro-use cases, and few are worth the effort.

What I’m seeing on the ground

Today, the most fruitful AI deployments are enhancements to existing workflows, not wholesale replacement of workflows. This usually means giving the people responsible for a workflow the tools to help them do it faster, better, and cheaper. But this is harder than it sounds, for a few reasons:

1. You need total process clarity (and most don’t have it). If you can’t articulate how value is generated in a process, you can’t target AI to the right tasks. This is why so many teams pursue easy builds like chatbots that don’t fix real business problems – they’re easy to ship, but there’s nothing to be gained. You need to understand each part of the workflow, and which parts of that workflow impact your margins.
   
   
2. Workflow redesign is an iterative process (and most don’t want that). You can’t know the end-state of an AI-powered workflow ahead of time (or get to a final product in your V1). You need to measure the impact of an initial change, iterate, and repeat. After a few versions, the workflow will usually have changed in ways you didn’t expect.

The world has spent the last 25 years pursuing digital transformation – moving from paper processes to digital ones. AI’s new capabilities mean we need to embark on a revamp of that effort: AI transformation. I know what you’re thinking, and I cringe to myself just saying those words, but it’s real.  

Digital transformation earned a bad reputation (with its armies of consultants and months spent on meticulously defining business ontology) – but many of the insights learned during that era are applicable for this new AI revolution, just delivered far faster.

The bottom line: Business leaders need to work through their analysis paralysis, identify the top 3-4 opportunities, and get started – even if they’re not 100% convinced of the direction. AI is moving too quickly to delay action, so speed and iteration beats months of abstraction.

The mindset shift needed

Most organizations – and frankly, most tech teams – default to thinking about outputs when they think about AI augmentation: “We should build an AI that can search our knowledge base”, “let’s automate expense tracking,” etc.

But those outputs are the result of workflows. If you focus on fixing the workflow itself – especially the parts that are labor-intensive, error-prone, or data-starved – you often find bigger, faster gains.

The mindset shift you need to make is concentrating on outcomes – e.g., lower time to resolution for service requests, reduced number of HR inquiries, more on-time expense reporting, etc. Then you examine the workflows involved in achieving that outcome – and more specifically, the steps getting in the way.

If you need a metric to help you prioritize, margins give you an objective, financial measure to do so. You can think about this in a couple of ways:

1. Positive margin on inputs vs. outputs. Improving efficiency or throughput of a workflow to increase earnings without proportional cost increases.
   
   
2. High cost of failure. Focusing less on directly creating revenue, but avoiding expensive inefficiencies – financially, operationally, or reputationally.

Take the example of a hospital:

1. AI in a treatment workflow impacts margin because that’s where reimbursement happens.
2. In a triage workflow, margin isn’t directly created – but the cost of mistakes (misdiagnosis, delayed care) is high enough to warrant extra resources and attention.

The key is to shift from gut feelings about what AI should be solving, to a structured, repeatable, measurable process of uncovering where AI can have the highest impact.

5 steps to picking high value use cases

Let’s put all the pieces together. Here’s my framework for ROI-generating use case discovery:

Step 1: The template (aka your industry blueprint)

Start with the high-level overview of your industry-typical processes. What do you do?

Sticking with the hospital theme, this might look like: you admit → triage → stabilize → assess → transfer/treat → discharge.

Every industry has a relatively stable set of big activities that most players perform, even if they do them differently. This helps you start with something other than a blank page. It also helps you avoid overlooking obvious levers from jumping straight into your company’s quirks instead of considering the baseline steps that are universal in your space.

Most companies also have a set of non-core-value functions like HR, accounting, marketing, etc. These are essential, but often aren’t core differentiators. And because their processes are pretty similar across all organizations, AI-powered enterprise software will emerge soon enough to drive efficiency in these generalized functions.

I’m not saying that you can’t use AI to make these departments more efficient – it’s more that investing in software development here is very unlikely to deliver meaningful ROI. So note them but set them aside.

Step 2: The actions (aka your organization’s core activities)

In this step, you customize that general template to your specific business operations. Actions are the concrete, repeatable, high-level workflows that happen in your business every single day.

Keeping to the hospital theme, you would define exactly how your hospital works through all 6 of the above steps. Do you run separate triage for ER vs. urgent care? Does stabilization involve a rapid response team or an ICU step? These differences matter, because AI opportunities hide in the specifics.

To nail this step:

• Get the right people in the room. Bring in the managers and ICs from each function who actually know what happens day-to-day.
• Describe, don’t label. Avoid abstract terms like “customer success” or “operations.” Break out the specifics, like “respond to client inquiries” or “schedule field service visits”.
• Stick to the essentials. You’re looking for the 5-10 activities that, if they stopped tomorrow, your business would grind to a halt – or that, if you could double the efficiency, you’d make a lot more money.

Step 3: The stages (aka sub-steps inside each action)

Stages are the smaller, sequential steps that have to happen for the action to be completed.

This is where you stop talking about big buckets of work and start describing the flow of that work. This step is critical for a couple of reasons:

1. It exposes variability. At the action level, two departments might look identical. But at the stage level, you can see where methods diverge – and whether those differences are helping or hurting.
   
   
2. It reveals hidden bottlenecks. Bottlenecks rarely appear at the “action” layer (“triage patients”). They appear in the tactical steps inside it (“wait for equipment to free up” or “call in a specialist”).

You’re not writing an SOP yet – in fact, it’s really important NOT to get caught up in the “micro” details. The point of this exercise is to create a map of the business so you can identify where you’re getting stuck in the mud.

Step 4: The resources (aka what it takes to complete each stage)

List out all the ingredients needed for a stage to be accomplished successfully, broken down into: The necessary people, processes, data, and technology.

Not only does this step reveal the levers that you can pull with AI, it can highlight dependencies where your process relies too heavily on one type (e.g., manual labor) or lacks another (e.g., structured data).

If a stage is bottlenecked by people doing repetitive tasks, for example, that’s an opportunity for AI to automate or assist. If technology is outdated or data is siloed, AI can act as a bridge without heavy integration work. And so on, and so on.

Step 5: Assessing AI fit

Once you know the resources behind a stage, you can ask: where could AI add speed, accuracy, or intelligence?

This should correlate to the two-pronged approach we talked about earlier – where can AI:

1. Directly create margin, or
2. Avoid an expensive failure

For the resources of each stage, ask yourself:

People: Can AI take over repetitive, time-consuming tasks to free human capacity? Can it augment human judgment with faster, better data?

Process: Can AI shorten or simplify the workflow by automating steps or removing handoffs?

Data: Can AI generate, clean, interpret, or enrich the data needed for this stage? Or deliver it more efficiently when and where it’s needed?

Technology: Can AI act as an interface between disconnected systems without requiring heavy integration work? Or replace older “best-effort” attempts from the past?

And for each of these, you need to understand:

• Does AI have the technical capability to improve this resource today?
• Would that improvement create or protect margin in a meaningful way?
• Can the solution be implemented without prohibitively high cost, complexity, or risk?

How to do this well

Now you have a clearly broken down process with obvious areas of opportunity – but you still need the second part of the equation: You need to start quickly and you need to iterate.

A leadership workshop can usually name the actions and most stages in a sitting, then screen for value-creating moments and likely AI fits. The point isn’t to be exhaustive, it’s to identify a few high-leverage candidates fast. Then move into the iteration loop:

1. Set goals and metrics: Measure the time spent and value generated in the existing workflow to benchmark AI against later
2. Enhance first: Add AI to the existing workflow to help humans
3. Pilot & learn: Watch what changes; capture human heuristics
4. Reconfigure: Combine/eliminate stages now made redundant; graduate humans from doers to reviewers/managers of AI
5. Repeat: Expect multiple versions before the workflow looks “new” and beware anyone promising instant “AI-ified” systems via agents
6. Measure: Measure the time spent and value generated again and observe the changes

Final pro tip: AI is worse than you’d think at BIG things, and much better at small things. AI has thinking capabilities, so it can handle messy tasks that normally would have needed a human – or would have been impossible or ridiculously difficult with traditional programming – like making small decisions based on unstructured data, tracking and extracting meaning from a conversation, transforming messy information into usable data, generating readable reports, creating on-brand assets, etc.

So choose an AI-enabled, human-like workaround first, before committing to a programming-heavy integration. This lets you move on high-value opportunities faster and start capturing ROI without waiting months for complex build-outs.

P.S. If you want a no-BS, solutions-oriented partner to help you capture the opportunity of AI, find Ed at Machine & Partners.