Stop Automating Old Processes, Redesign Them Instead

Automating an old process makes it faster. It does not make it better, and Deloitte's 2026 survey found 84% of companies haven't redesigned work around AI at all (Deloitte, State of AI in the Enterprise, Jan 2026). This is for operations leaders and founders past the pilot stage, looking at an AI budget that produced activity but not much profit. It isn't a claim that automation is worthless. It's a claim that automation applied to the wrong layer of the problem is why so many AI programs plateau.

Why does automating the old process so often fail to pay off?

A process is a sequence of steps that turns an input into an outcome a customer or the business values. Automation accelerates the steps. It does nothing to question whether they're the right steps. That's exactly the gap McKinsey's 2026 global survey exposes. 80% of respondents report improved personal productivity from AI, yet only 37% attribute any profit impact to it, almost unchanged from the year before (McKinsey, Aug 2026).

That's not a small gap.

It's most of the AI spend in 2026 producing busier employees and no better P&L line.

Deloitte's survey splits companies into three groups by how deep their AI effort actually goes (Deloitte, Jan 2026):

• 34% are starting to deeply transform the business: new products, new processes, even new business models.
• 30% are redesigning key processes around AI while leaving the business model intact.
• 37% are using AI at a surface level, with little or no change to how work actually happens.

All three groups report some productivity gain. Only the first two are actually redesigning anything.

The report is direct about why that distinction matters: "AI doesn't just augment existing processes. It often requires fundamentally rethinking operating models and how work gets done." Its own example is a loan officer who's always approved loans on judgment and experience, now handed an AI system that produces a recommendation instead. Nobody automated that officer's job.

But the job itself needs new answers: when do you override the model, and how do you explain the decision to the customer? What happens to the officer's expertise if "reviewing the AI's output" quietly becomes the whole role? Bolt AI onto the old job description and none of those questions get asked. They just get deferred until something goes wrong.

What does a genuinely redesigned workflow look like instead?

Workflow redesign is the practice of rebuilding a process end to end around what's newly possible, instead of inserting a tool at the point that was easiest to reach. McKinsey found 74% of high-performing companies redesigned workflows this way from the ground up, versus roughly 25% of everyone else, and the difference compounds (McKinsey, Aug 2026).

BCG's fourth annual AI at Work survey, covering nearly 11,750 workers across 14 markets, puts a number on that compounding. Companies pursuing workflow redesign are 24 percentage points more likely to see measurable business improvement. They're also 22 points more likely to save a full day per week, and 20 points more likely to report higher job satisfaction (BCG, 2026). Strategic clarity on its own, the survey found, moves the improvement number by 25 points. Better tools alone move it by about 5.

Read that twice. Tool quality explains a fifth of what strategy explains. Most AI budgets are spent in exactly the wrong proportion to that ratio. It's the same imbalance behind BCG's 10-20-70 rule: roughly 10% of AI value comes from the algorithm, and 70% comes from changing how people and process operate.

A few markers separate a redesigned workflow from an automated one:

• The number of handoffs between people and systems goes down, not just faster, fewer.
• At least one step that used to exist gets removed entirely, not merely accelerated.
• The metric being optimized is a customer or business outcome, not an internal activity count.
• Someone can point to the moment the old process would have broken and show what the new one does instead.

What actual sequence turns automation into redesign?

Nick Colisto, CIO of the $9B manufacturer Avery Dennison, has a single test question for any workflow before a tool touches it (CIO.com, Aug 2026): "If we were designing this process today, would we build it the same way?" Most workflows fail that question, not because anyone designed them badly, but because they're the accumulated residue of old decisions. Old audit responses, old acquisitions, old regulations, each one reasonable at the time, never revisited since.

Colisto's answer is a four-step sequence, and the order is the entire point: automate is last, not first.

1. Simplify — remove the approvals, reports, and handoffs that don't need to exist. His own example: a purchasing process with six approval steps. AI can route and summarize all six faster. Only a leader can decide whether the business actually needs six, or whether three were always just inherited habit.
2. Standardize — build one consistent way of working across teams, so an improvement made once can be repeated everywhere instead of re-litigated team by team.
3. Redesign — rebuild the workflow around the outcome you actually want, not around the systems or org chart you happen to have. This is also where to decide on purpose what happens when something goes wrong. The workflow needs to notice it's uncertain, route the exception to a person, reverse an action already sent, and feed the correction back in so the mistake doesn't repeat. Skip this and the first edge case becomes a fire drill instead of a logged, learned-from exception.
4. Automate — only now, once the first three steps are done, apply AI to what's left.

As Colisto puts it: "Sometimes the greatest improvement comes from eliminating work rather than automating it."

Most companies skip straight to step four. They default to applying AI everywhere equally, because sorting workflows by stakes feels like more upfront work than just switching a tool on. That's not caution, it's a mismatch tax. Plenty of low-stakes, reversible tasks sit stuck behind unnecessary human review the old process never removed, which costs real throughput.

A few genuinely high-stakes steps get the same light-touch automation as everything else, because nobody separated them out in the first place. Simplify and Standardize are what catch this before Automate ever runs.

What should a leader actually do with this, starting this quarter?

Pick one workflow. Not five. One that already touches revenue, cost, or a customer complaint you've heard more than once.

This is the same diagnose-first move behind the AI Work Redesign Framework I use with clients. Don't open with "what can we automate." Open with "what does this workflow actually produce, and where does it break?" The five-step test below is the diagnose step made concrete enough to run without a consultant in the room:

1. Map the full workflow, every handoff, not just the step someone wants to bolt a tool onto.
2. Simplify first: cross out every approval, report, and handoff the team can't justify from scratch today.
3. Name the outcome a customer or the business actually values at the end of it, not the activity in the middle.
4. Write down, in one sentence per step, what happens when that step fails today.
5. Design the AI version to handle that failure sentence explicitly, with a named escalation path, before it touches the happy path.

If step four stumps the team, that's useful information. It usually means the "process" was never really designed, it just accumulated. But automating an accumulated process just locks the accumulation in place, permanently, and faster.

I'd push back on my own argument here for a second: this is more upfront work than most teams want to do before a pilot. The case for automating first and sorting it out later isn't crazy, speed has value, and a working pilot builds internal credibility that a redesign memo doesn't. Where I land is that the upfront mapping takes days, not months, for a single workflow. The 24-point gap BCG found didn't come from companies that waited a year to start. It came from companies that did the mapping before they scaled.

Not every process deserves this treatment. A low-stakes, already-simple task doesn't need a redesign ritual, just automate it and move on. The test above is for the workflows that actually touch revenue, cost, or a customer relationship, where getting it wrong compounds.

I'd like to hear where the five-step test breaks down in practice, especially in a smaller team without a dedicated ops function to do the mapping. If you've redesigned a workflow and skipped one of these steps anyway, tell me what happened. That's the gap I'm least sure about.

Frequently asked questions

What's the difference between automating a process and redesigning it?

Automating keeps the existing steps and speeds them up with AI. Redesigning asks whether the steps themselves should exist, then rebuilds the workflow around what AI can newly do, often removing steps entirely rather than accelerating them.

Why does automating an old process usually fail to deliver ROI?

Because the process wasn't the bottleneck to begin with, the handoffs and decision points were. Speeding up a flawed workflow just produces the same bad outcome faster, which is why McKinsey found most AI deployments report productivity gains with no attributable profit impact.

What's a concrete sequence for turning automation into redesign?

Simplify, standardize, redesign, automate, in that order. Cut unnecessary approvals and handoffs first, create one consistent way of working, rebuild the workflow around the outcome you actually want, and only then apply AI. Applying AI before the first three steps just locks today's inefficiency in at machine speed.

How do I know if my company is redesigning or just automating?

Check whether anyone mapped the end-to-end workflow and removed steps, or whether AI just got inserted at one point without anyone questioning the rest. If the process chart looks identical to a year ago except faster, that's automation, not redesign.

What's a practical first step for a mid-size company with no innovation budget?

Pick one workflow that already touches revenue or cost, write down what happens when it goes wrong today, and design the AI version to handle that failure case explicitly before automating the happy path.

Key takeaways

• Deloitte found 84% of companies haven't redesigned work around AI at all, while McKinsey found 80% report productivity gains but only 37% attribute any profit impact to AI.
• Workflow redesign compounds: BCG found it predicts 24 points more measurable business improvement, and strategic clarity alone explains 25 points of impact versus 5 points from better tools.
• Run the sequence in order: simplify unnecessary approvals and handoffs, standardize how the work gets done, redesign around the outcome, and only then automate.
• Design for the unhappy path first: how the system flags uncertainty, escalates, reverses, and learns, before automating the smooth case.
• Pick one revenue- or cost-touching workflow and run the five-step test this quarter rather than piloting automation broadly across many shallow use cases.