Why Layering AI Onto Old Workflows Gets You 30% ROI. Reimagining Them Gets You 30x.

I have spent the last few months talking to enterprise leaders about AI transformation. Almost all of them are doing it. Most of them are doing the wrong version of it.  

A recent MIT study found that despite billions in enterprise investment into generative AI, 95% of organizations are getting zero return. That is not a technology problem. It is a strategy problem. 

There are two fundamentally different approaches playing out right now. The first is what I think of as the efficiency layer. You map your existing workflows, find the friction points, and insert AI to speed things up. Automate the report. Summarize the meeting. Generate the first draft. This produces real gains, typically 20 to 30 percent automation across targeted workflows. It feels like progress.  

The problem is the math. When you account for the AI investment required to achieve those gains, the ROI is almost negligible. You have not transformed anything. You have made the existing system marginally faster.  

The second approach is harder. Instead of asking where AI fits into your current processes, you ask a different question: if we were designing this operation from scratch today, knowing what AI can do, what would it look like? You stop retrofitting and start reimagining. The teams doing this are not seeing 20 to 30 percent gains. They are building toward 20 to 30x impact. That is not a rounding error. It is a different game entirely.  

At APPLY, every new outcome-based engagement starts with about 50 percent agents working alongside human practitioners in matrix teams. Humans set direction, make judgment calls, own accountability. Agents handle execution at a speed and scale no human team can match. We are not at 30x across every deployment yet, but the trajectory is clear, and the difference between organizations building toward that and those chasing incremental gains is visible from the first conversation.  

What actually separates the two groups? In my experience it comes down to three things, and none of them are the technology itself.  

Process design 

Most AI implementations struggle not because the model underperforms but because the process it was dropped into was never designed for the speed and variability AI introduces. When you genuinely reimagine work with AI, you go back to first principles. What decisions need to be made? What does an agent need to make them well?  

Where does human judgment stay in the loop and why? These are not technology questions. They are operational design questions, and most organizations have not asked them seriously yet. The ones that do find that their workflows look fundamentally different on the other side, not just faster versions of what they had before.  

Enablement 

Your team isn’t just learning a new tool. They are navigating a shift in what their expertise is worth and how it gets applied. I have seen technically successful AI deployments fail because the people using them felt threatened rather than empowered. There is real anxiety in organizations right now about what AI means for people’s roles.  

Ignoring that anxiety does not make it go away. It shows up as resistance, workarounds and adoption that looks good on a dashboard but does not actually change how work gets done. 

The practitioners who adapt fastest are not the ones who were told to use AI. They are the ones who were shown what becomes possible when they do. That distinction sounds small but it changes everything about how a rollout lands.  

Real enablement means explaining not just how the tools work but why the work is changing, what new capabilities become possible when agents handle execution, and what the human role looks like in that model. It means giving people a reason to lean in rather than a mandate to comply. The organizations getting this right are investing as seriously in the human transition as the technical one. That investment is what turns a deployment into a transformation.  

Governance 

This is where failures compound quietly, and it is the area I watch most closely. Small governance compromises get accepted one at a time until the risk becomes invisible through familiarity. A named human needs to own every AI output before it reaches a client or customer. Not a policy document. Not a committee. A person who is accountable for the result regardless of what tools produced it.  

The risk pattern I see most often is what happens when organizations move fast on deployment without building that accountability structure first. They do not fail dramatically. They fail slowly, through outputs that erode trust over time, compliance gaps that surface months later, and personalization decisions that violate the implicit contract customers thought they had with the brand. By the time it is visible it is already expensive. The organizations that treat governance as a foundation rather than a constraint are the ones building something worth trusting.   

The race is not over. You haven’t missed it. But the organizations treating AI as a layer on top of existing operating models are not in the same race as the ones redesigning around it. The gap between 30 percent and 30x lives in process, enablement and governance. The technology is the easy part.