AI is boosting individual productivity across your team. But until you redesign how work actually flows, those gains will never show up on your P&L.

Everyone on your team is more productive. Your analysts build models faster. Your marketers generate content in minutes. Your developers ship code they didn't write.

Walk through any mid-market company right now, and you'll find dozens of people using AI every day.

And yet.

Revenue hasn't budged. Margins haven't improved. The P&L looks basically the same as it did eighteen months ago.

So where did all that productivity go?

We think about this question constantly. We regularly work with organizations trying to answer it. And the answer from what we can tell is generally the same each time. The AI isn’t the problem. It works fine.

But the organization has not adapted itself to take advantage.

The productivity illusion

But how could that be? Individual team members are reporting material improvements in productivity. And they aren’t wrong. We see it ourselves - our engineers are able to ship far more than they ever were before.

The issue: faster at the individual level does not result in better at the organizational level.

An analyst today might be able to build a model in an hour. But the decision that model is meant to inform still takes three weeks. Still has to route through the same approval chain. Still has to go through the laborious process of building internal consensus.

And so most of productivity gains are local optimizations. Individual tasks are made more efficient. But organizations don’t create value at the task level. Organizations unlock value through the coordination of many tasks. The flow of decisions. The handoffs between teams. And, at least so far, AI isn’t making that part faster.

Simply put, internal politics and old workflows are not designed to deal with the pace that AI can theoretically unlock.

Now we have to do the hard part

Closing this gap requires the hardest work in any enterprise - changing how the organization actually operates.

Many aspects of this are familiar to anyone who’s done change management before. Everyone has to be clear on the objective. You need to make sure the initiative has support, at the highest levels you can. It needs clear milestones, quick wins, momentum. It needs thorough documentation and training.

This is unglamorous. Politically complicated. It requires understanding the business at a level that most vendors don't have (and aren't interested in developing). It requires sitting with the people who do the work and watching how they actually do it, which is rarely what the process documentation says.

We've been in this work for fifteen years. First as a development shop, then as a consultancy, now as an implementation partner focused on data and AI.

The difference this time is the scope and the framing. Previously, digital transformation was largely about augmenting how people do the work. But people were still doing the work.

What’s new is that the AI or agent or swarm (or whatever term you want to use) is able to do many of the tasks we’re used to doing manually, at similar (or sometimes better) quality, 24/7, in a fraction of the time and cost.

But just because you can write 100 reports in the time it used to take to write one, doesn’t mean your organization has the ability to edit or review or discuss or implement the recommendations in those reports. At least not how it’s currently constructed.

It’s not that different than suddenly having the ability to get 100 times the raw materials for your widget factory. That doesn’t mean much if the rest of the org isn’t able to process it.

Hopefully the implication is clear: we have to rethink the entire process for how we do things. To fully benefit from these tools, we have to figure out a way to move at their pace.

But how do you actually do that?

How to Start Redesigning Workflows With AI In Mind

It starts with first principles. Instead of looking at your current workflow and asking, "where can an agent help?" you have to think bigger. That question implicitly preserves every assumption baked into the current process (the approvals, the handoffs, the reviews) and just makes individual steps faster.

The right question is: if we were building this function from scratch today, knowing what agents can do, what would it look like? The answer is almost never "the same process, but faster."

An example might be helpful. Take financial reporting. Today, most mid-market companies close their books on a monthly cycle. An analyst pulls data from a couple different systems, reconciles discrepancies, builds a report, sends it to the controller, who reviews it, flags questions, sends it back, gets answers, approves it, and sends it to the CFO.

That cycle takes ten to fifteen business days. It's been ten to fifteen business days for twenty years.

An agent doesn't make that cycle take five days. An agent makes the cycle disappear.

An agent is capable of reconciling transactions continuously. It can flag discrepancies the day they occur instead of when someone notices them during close.

The controller doesn't review a monthly report. They review a dashboard of exceptions the agent couldn't resolve on its own. And the CFO doesn't wait for month-end to know where the business stands. They know today, because the data is always current.

That's not a faster version of the old process. It's a different process. The monthly close as a discrete event stops existing.

That means roles change. Your accounting team shifts from data assembly to exception judgment.

The skills change. Pattern recognition and business context matter more than being great at Excel or Quickbooks.

The management cadence changes. You don't need a close meeting if there's nothing to close.

Or take customer support - what would it look like to reimagine that? An agent handles the initial interaction, pulls the customer's full history, checks known issues, and resolves the 60-70% of tickets that follow a known pattern in minutes, not hours. But here's the part people miss: the agent also categorizes what it couldn't resolve and why.

This means the humans on the team aren't working a ticket queue. They're working a curated list of genuinely novel problems, pre-loaded with every piece of context the agent already gathered.

Their job isn’t to "answer tickets." Their job is now "solve problems that don't have a known solution yet, and create the known solution so the agent handles it next time."

That’s likely a common pattern. The human's output isn't just the completion of a widget. It's equipping a new capability for the agent. Every problem a person solves teaches the system to solve it autonomously next time. The problems they work on get harder and more interesting. And the system gets better every week, because every human intervention is also a training event.

In other words, agents handle the volume. Humans handle the judgment. And there's a feedback loop connecting the two so the system improves continuously.

But you can't get there by layering agents onto your current org chart. You have to be willing to look at the whole workflow and ask what it would look like if humans only touched the parts that require actual human judgment.

Three questions to start

If this sounds abstract, here's how to make it concrete. Pick a process that shows up on your P&L. One you spend real money on. And ask three questions:

1. What would this look like if it ran continuously instead of in batches? Most business processes are batched because humans need to batch. We do payroll biweekly. We close the books monthly. We review pipelines on Tuesdays. But agents don't need batches. When you remove the batch constraint, the entire process architecture changes (and usually gets simpler.)
2. Where are humans currently moving information between systems instead of making decisions? Half of most knowledge work is copying data from one place, reformatting it, and putting it somewhere else. That's not a step to accelerate. It's a step to eliminate. When you map your actual workflows honestly, the amount of human effort spent on information logistics versus actual judgment is staggering. Agents don't make the logistics faster, they make them unnecessary.
3. What decisions are currently batched because the analysis was expensive? A lot of approval gates exist because it used to be costly to run the analysis. You needed a VP to approve the research plan because a bad plan wasted two weeks of analyst time. When an agent can run every scenario in an hour, that gate isn't protecting anything. It's just adding latency. The VP's judgment still matters, but it should be applied to which scenario to act on, not whether to run the analysis.

These aren't theoretical exercises. When we work through them with clients, the answer is usually that 40-60% of the steps in a given workflow exist because of constraints that agents have already removed. The steps just haven't been removed yet, because nobody's asked the question.

How to get this funded

The standard ROI math doesn't work here. You can model a 10% improvement to step 3 of an existing process. When you're asking whether step 3 should exist at all, the spreadsheet breaks.

But the work still has to get funded.

What we've found actually works is simpler than a business case. Run those three questions on a real process. Map what's actually happening. When a CFO sees that their monthly close takes 15 days because it was designed for humans to batch work (and what you’re suggested is that the constraint that required batching is just gone) you don't need an ROI model. The reaction is almost always "why are we still doing this?" The absurdity sells itself once you make it visible.

The other thing that moves executives is competition. HBR published a piece recently about why AI transformations stall. They identified something they call "the efficiency trap". companies measuring AI in hours saved, wondering why nothing shows up on the P&L. We see it constantly. The way out is to stop talking about efficiency and start talking about capability. Not "we'll do the same thing 20% faster" but "we'll be able to do something we literally cannot do today."

That reframe changes the whole conversation. If a competing healthcare system moves to continuous claims reconciliation and starts catching revenue leakage in real time, that lands differently in a board room than "we could save some FTEs."

We’re in a brave new world. And while we love the traditional digital transformation playbook, the incremental, horizon 1 improvements, we suspect that to fully unlock AI in your organization you have to think bigger and bolder. You have to be willing to challenge “the way we do things here”, be willing to ask what it might look like to reinvent workflows from first principles.

‍