Where Should We Actually Use AI?

AI has put organizations in a strange spot: there are now far more plausible use cases than there are good reasons to actually pursue them. Ask almost any leadership team where AI could help, and the list grows fast. It can summarize meetings, search policies, build forecasts, draft communications, analyze customers, review contracts, support recruiting, write code, help service teams. Give that same group another hour, and they’ll find two dozen more ideas.

That abundance is a good problem to have. But it also removes a constraint that used to do a lot of the deciding for us. Traditional automation was expensive, integration was hard, and implementation took time, so only a relatively small number of opportunities ever justified the investment. AI erodes that filter. A useful experiment can come together in an afternoon, new capability increasingly shows up inside software the organization already owns, and employees can point an intelligent tool at work that never would have cleared the bar for a conventional technology project.

The more organizations experiment with AI, the less useful “can we build it” becomes as a test of what actually matters. A catalogue of a hundred AI use cases says a lot about a team’s energy and imagination. It doesn’t say the organization knows which problems matter, which work deserves to change, or where intelligence can produce an outcome worth sustaining.

Start with the outcome, not the AI

New technology has a kind of gravitational pull. A vendor shows off an impressive agent, and leaders start looking for somewhere to deploy one. A platform ships new AI features, and every function gets asked to produce use cases. An employee finds that a tool can do one thing well and immediately imagines ten more.

That kind of exploration is genuinely useful: plenty of good ideas surface precisely because people were given room to experiment. The trouble starts when the technology itself becomes the organizing idea.

“Use AI in recruiting.” “Build a sales agent.” “Automate Finance.” Each of those names a domain and a technology. None of them says what’s actually wrong with the work today, or what would genuinely get better if it changed. A better starting question is a harder one: what outcome are we actually trying to improve?

Maybe preparation takes too long. Maybe information is scattered across too many places, or service quality swings for no good reason, or a decision keeps landing too late. Maybe expertise never quite scales past the handful of people who have it, or the same avoidable mistake keeps happening. Or maybe the organization keeps spotting an important signal only after the window to act on it has already narrowed.

Once the outcome and the constraint are clear, AI becomes one option among several, alongside redesigning the process, fixing the underlying information, building an integration, switching on a capability that already exists in an enterprise platform, or simply deciding to stop doing the work at all. The goal was never to find somewhere to put AI. It’s to identify work that’s actually worth changing.

Faster work is not necessarily better work

Take what happens when a new client engagement kicks off. Once the agreement is signed, a project leader typically has to track down the executed statement of work, piece together whatever commitments were made during the sale, confirm staffing assumptions, figure out who the client stakeholders are, pull together the financials, catalog the risks, and turn a pile of scattered handoff notes into something the delivery team can actually use.

An intelligent system could plow through most of that prep in minutes. But a faster kickoff brief was never really the point. The point is that delivery starts with everyone actually sharing an accurate understanding of the commitments, the objectives, who’s responsible for what, the assumptions, the risks, and the decisions still waiting to be made before work begins.

Here’s why that’s not a small point: if the real problem is a weak handoff from sales, conflicting information or unclear ownership, an agent can simply reconstruct that same mess, only faster. It makes an activity more efficient while leaving in place the very thing that made it hard.

AI might still be part of the right answer: pulling together authoritative information, flagging contradictions, and getting the project leader ready to resolve them before kickoff. But the right answer might just as easily be a better handoff process, a properly maintained record of what was promised, or simply cutting out information nobody actually needs.

As automation gets easier, organizations should get pickier about what they automate, not more permissive. AI can absolutely make inefficient work faster. That’s not the same thing as making the organization better.

Look beyond the visible task

The most obvious AI opportunities usually look like tasks: draft the message, summarize the meeting, pull the fields, put together the report, answer the question. Those are genuinely useful, especially at the individual level, and no organization needs to run every small productivity gain through a strategic investment process. But the bigger opportunity is often sitting around the task, not inside it.

Think about how much time capable people spend just getting to the point where their expertise is useful. A salesperson digs through several systems before they even understand the customer. A finance professional reconciles numbers before real analysis can start. An HR leader has to assemble the workforce picture before advising the business. A manager chases updates from several people before deciding whether to step in.

AI can increasingly take on some of that preparation. But the value isn’t really in removing the prep itself. It’s in what becomes possible once that prep stops eating up the same attention.

Does the salesperson actually spend more time with the customer? Does Finance get to investigate the variance instead of just compiling it? Does the manager coach the team instead of chasing updates? Does the specialist get to spend judgment on the cases that actually matter?

The decision may matter more than the activity

There’s another reason a task list undersells the opportunity: some of the most valuable work barely takes any time at all. Noticing that a major customer relationship is going sideways, that a project has picked up a dangerous dependency, that a workforce constraint is about to derail next quarter’s plan, or that a supplier has quietly become unreliable. Recognizing any of that can take a moment. But catching it early enough can be worth an enormous amount.

This is where the opportunity stops being about productivity in the usual sense. Enterprise systems have always been good at answering a question once someone knows to ask it. AI is increasingly good at something different: pulling together evidence around a decision and, in some cases, flagging that something deserves attention before anyone has even thought to ask.

A customer relationship doesn’t collapse all at once, it erodes. A project doesn’t blow up from one delay but from several small ones stacking up. A high performer doesn’t quit one day; they disengage gradually. None of that trips a workflow on its own. Someone has to notice the pattern.

Intelligent systems are increasingly capable of watching for those weak signals, pulling in the relevant organizational context, and surfacing a meaningful change to whoever is accountable for the outcome. That’s not a faster version of an existing task. It’s a genuinely different kind of organizational intelligence.

For a lot of organizations, the biggest AI opportunity may have less to do with doing existing work faster and more to do with what the enterprise notices in the first place, when it notices it, and which decisions get a scarce human’s attention.

Follow the outcome across the organization

Opportunity-spotting also gets skewed when it follows the org chart too closely. HR looks for HR use cases. Finance looks for Finance use cases. Sales looks for Sales use cases. IT looks for technology use cases. That makes sense: functions own the budgets, the systems, the people. But plenty of important outcomes don’t respect those lines.

Client mobilization runs through Sales, Delivery, Finance and Staffing. Employee onboarding touches Recruiting, HR, Payroll, Identity, Technology and the hiring manager. Customer onboarding can cross Commercial Operations, Security, Implementation, Service and Billing.

Improve one activity inside each function and you can make several steps faster while the overall experience stays just as fragmented. AI is especially useful right at these organizational seams, where it can pull context across systems, coordinate activity, and help people make sense of a situation that no single application or function fully owns.

But the design has to follow the outcome. It can’t just wrap each silo’s existing activity in something smarter. That means asking where information keeps getting rebuilt from scratch, where handoffs lose context, where decisions sit waiting for coordination, and where nobody actually owns the full picture of what’s being produced. Those seams often hold more value than the obvious task list ever will.

Invest according to the value of the work

This is where it helps to separate commodity work from contextual work from differentiating work. AI makes it technically possible to reproduce an organization’s existing complexity with remarkable efficiency. That doesn’t mean the complexity is worth keeping.

Commodity work usually rewards standardizing, simplifying, and leaning on dependable platform capability. Contextual work gets more valuable when intelligence can pull together the specifics of this particular organization and help someone reach a better conclusion. Differentiating work can justify deeper investment, where intelligence strengthens a capability central to how the enterprise creates value.

None of this is determined by what the function is called. Workforce scheduling might be routine administration at one company and a strategic capability at an airline, a hospital or a logistics organization. Customer intelligence might be useful support at one business and central to competitive advantage at another.

The question isn’t how unusual a process looks. It’s how much value the organization creates by doing it differently. That should shape how ambitious the AI investment is: standardize what doesn’t matter, apply organizational context where judgment matters, and invest most deliberately where intelligence can strengthen something genuinely valuable and hard to copy.

Sometimes the right answer is not AI

A mature AI strategy has to be willing to say that intelligence isn’t the answer here. Sometimes the process is poorly designed. Sometimes the data is wrong. Sometimes what’s needed is an integration. Sometimes the capability already exists in Workday or another enterprise platform and just hasn’t been adopted properly. And sometimes the work should simply stop.

AI happens to be unusually good at papering over structural weakness. It can make sense of inconsistent formats, work around bad search, reconstruct context from several systems, and help people route around awkward processes. Those are real capabilities. They can also make an unhealthy operating model a lot easier to tolerate.

A clever agent shouldn’t end up as the permanent integration layer for a problem the organization already knows how to solve more directly. And the falling cost of automation shouldn’t become an excuse to keep reports, approvals, handoffs and exceptions around long after their original purpose disappeared.

A worthwhile opportunity survives without the words “artificial intelligence”

There’s a useful test for any serious AI proposal: describe the opportunity without mentioning the technology at all. The engagement-mobilization example passes that test easily.

Every new engagement should start with everyone sharing the same up-to-date understanding of the contractual commitments, the client’s objectives, the key assumptions, who’s responsible for what, the risks, and the decisions still needed before work starts. Right now, that understanding often gets pieced together by hand from several sources. The quality of the handoff is inconsistent, preparation eats up time that could go elsewhere, and important commitments sometimes only surface after delivery has already begun.

What we actually want is straightforward: relevant information assembled consistently, conflicts and gaps made visible, and a project leader who starts with a prepared understanding of the engagement instead of a search exercise, while still being the one accountable for validating the important conclusions and resolving ambiguity with the client.

Only then does the technology question become useful. Could AI assemble the information? What sources would it need? What could it reasonably infer, and what should it merely flag? Where does the project leader need to stay decisive? And which platform should actually run the work?

If an opportunity is still compelling before AI ever enters the picture, the technology has something real to improve. If the business case disappears the moment the word AI is removed, the organization may have enthusiasm for the technology without clarity about the work.

Measure the work that changes

That same discipline should shape how success gets measured. A functioning agent is not the outcome.

If the goal was faster preparation, preparation time should actually go down. If the goal was better decision-making, the organization needs to know what evidence of an improved decision would even look like. If the capability is meant to catch risk earlier, leaders should be able to tell whether meaningful signals are showing up sooner, and whether people are actually acting on them.

The measure has to follow the reason the work was chosen in the first place. That guards against a common trap in AI programs, where activity quietly becomes the evidence of success: counting agents, prompts, users, generated content or automated transactions, mostly because those things are easy to count.

Those numbers prove AI is being used. They don’t prove the organization changed something worth changing.

Build a portfolio that can learn

Not every AI opportunity deserves the same level of scrutiny. Personal productivity should stay easy to explore: low-consequence experimentation is exactly how organizations learn what this technology can do and how employees want to use it.

Anything going into production, though, needs a higher standard: a clear outcome, an observable improvement, and enough value to justify the operating responsibility that comes with it.

Strategic opportunities take more patience. Better decisions, earlier recognition, or a genuinely new organizational capability are all harder to measure than minutes saved. But they still need a credible line between the work being changed and the value the organization expects.

Some experiments will simply fail. Others will become features of the enterprise software the organization already owns, before internal development was ever justified. And some capabilities that work perfectly well technically still won’t produce enough value to earn a place in production. All of that is healthy.

The goal was never to predict with perfect accuracy which AI ideas will succeed. It’s to build enough discipline that the organization learns which forms of intelligence deserve a permanent place in the operating model, and has the confidence to stop the ones that don’t.

The decision leaders are really making

Before making a meaningful investment, leadership should be able to explain what outcome is worth improving, what’s actually standing in the way of it today, and what intelligence would materially change about that.

They should be able to tell the difference between preparation and coordination on one hand, and the judgment, relationships and accountability that still have to be human on the other. They should know why AI, specifically, beats redesign, integration or existing platform capability. And they should know exactly what evidence would tell them the underlying work has actually improved.

Notice that all of those questions are about the organization, not the model. That’s on purpose. The challenge ahead was never going to be a scarcity of AI capability. It’s a scarcity of attention, investment capacity and organizational willingness to change.

Organizations are about to have access to more intelligence than they can productively deploy. The strongest ones won’t be defined by the length of their use-case catalogues or the number of agents they’ve created. They’ll be defined by the quality of the choices they make about work.

They’ll simplify where being different creates no value. They’ll apply intelligence where organizational context sharpens judgment. They’ll invest where intelligence can strengthen a genuine advantage. They’ll catch consequential signals sooner. And they’ll stop doing work that no longer deserves to exist.

The question was never whether AI can pull off an impressive task. It’s whether the organization chose the right work to change.