Tom Davenport’s Post

Tom Davenport the parallel between AI wash and the 1990s reengineering-as-layoffs pattern is the right comparison, and most boards are still rewarding the demonstrable activity over the harder work. The piece I would add from the enterprise delivery side, which sits downstream of the redesign question you are engaging, is that even successful process redesign has an operating discipline problem after the redesigned workflow goes live. AI-driven processes drift. The redesign at month one is not the redesign at month twelve. The discipline that catches that drift is the layer most organisations have not yet built.

Tom Davenport So true Tom! I remember reading Re-engineering the Corporation (Hammer & Champy, 1993) and using that framework in our airline IT/Analytics projects. Redesigning workflows, business processes, decision-making processes is difficult for many reasons. Change being one big reason. People still don't like change and giving up control of "the way they have always done things" to a computer, a model, or an AI Agent. A key success factor is reminding them that the HUMAN is still in charge of the process and able to do exception handling on nuanced edge cases. But the program or the agent is now alerting them to issues and handling most of the routine tasks and decision-making. The business value and economic impact should be the primary motivator for change. Increasing efficiency or throughput, reducing cost, improving service level, etc. (This is isn't new--- we did this in the '90s with Airline Analytics projects.) I'm in the middle of a large AI-driven project right now involving transforming a predominantly manual, labor intensive, time consuming, complex set of processes using Agentic AI coupled with mathematical decision-making models. Our mantra is streamline, automate, and optimize. (Don't pave the cowpath!)

Unfortunately the skill set that's needed for successful process redesign has atrophied since the turn of the century as leaders have opted for short term incremental improvements.

Process redesign fails for the same reason AI projects fail — the assumption that the underlying data is accurate enough to build on. Most of the time, it isn't.

I find the change in any organization is the emotional change; entire careers are suddenly reconfigured because the old ways of doing things no longer make sense. Once the emotional buy-in is accomplished, finding ways to re-orient enterprise work with AI tools becomes refreshingly easy, clear and ongoing. It is the emotional change of leaving the old way of doing things that is the core blocker at this point. 5 stages of grief for any organization to go through.

Good piece, especially the AI wash point. One thing I'd add: even genuine process redesign often disappoints because the org around it didn't change with it. The departments, reporting lines, and buying units were built for the old way of working, and they pull the new process back toward it. The deeper redesign is at the org architecture level, which is harder than process work and almost no one attempts it.

Tom, to me if a redesign is the answer and I believe that is correct; then let’s look fully at the current architecture. If I do that, I see it as wrong for the AI era. The change is bottom up where the foundation holds all organizational meaning and data and the LLM and workflows operate over the top. This kind of architectural build is now possible. It is hard for all the reasons in the article; but hard isn’t impossible and if folks want to get this to work properly everyone is going to need to come together. For me it starts with a fresh look. Take a look https://hutch.news/AgenticCost

The "AI wash" framing is accurate, and you're right about the 90s parallel. But there's a layer most companies skip between task-level AI use and end-to-end redesign: the stabilization step. You can't automate what hasn't been standardized first. What stalls companies in practice isn't reluctance to redesign. It's that they start redesigning before they know which parts of the process are genuinely variable versus just undocumented. Process mining helps, but usually what it reveals is that the data to answer that question doesn't exist yet.

I find that the category management precedent is worth sitting with here. Supermarkets didn’t get value from better purchase-order workflows. The value came from moving the assortment decision closer to the demand signal. The process changed because the decision location changed, not the other way around. Most AI deployments are automating decisions rather than optimising them structurally, which is why it stays hard and gains are modest.

Tom, great insight as usual. Handoffs is the key word that comes to mind when thinking of the 30+ years of process redesign you pioneered. Many of your cases and some I've lived personally tell us most opportunity lives between the tasks, not inside them. I believe AI breaks that frame in a specific way. For thirty years, handoff redesign assumed humans on at least one end of every handoff. AI agents can now sit on both ends. The handoff map redraws itself. Which means the binary "judgment required, yes or no" question that ran the playbook also collapses, because the executor isn't binary anymore. Curious if you see it the same way.