Process Redesign is the second-lowest scoring pillar in the AI Maturity Index research baseline, at 17%. Only people and culture scores lower. And the reason is almost always the same: organisations are automating their existing processes instead of redesigning them.
Those sound similar. They are not the same thing, and the difference is the single biggest determinant of whether AI investment produces real commercial value or expensive busywork. RAND Corporation’s 2025 research found that 80.3% of AI projects fail to deliver intended value — and identified the root cause as process, not technology.
This pillar is where most AI pilots go to die. Not because the AI didn’t work. Because it was bolted onto a process nobody had questioned.
Automating a process means applying AI to speed up or remove manual steps from a workflow that otherwise stays the same. Redesigning a process means interrogating why the workflow exists in its current form at all, and rebuilding it around what AI actually makes possible — which often means removing steps entirely, not just accelerating them.
Automation asks: ‘how do we do this faster?’ Redesign asks: ‘should we still be doing this at all, in this order, with these handoffs?’ The first produces incremental efficiency. The second produces structural change — and it’s structural change that shows up in commercial outcomes.
The g2m AI Maturity Index measures process redesign across five questions: how AI has been applied to revenue-generating processes, how AI impact is measured, the fate of AI pilots, how much AI has changed team workflows, and how disciplined the organisation is about prioritising process improvement. The pattern across these questions reveals whether an organisation is genuinely redesigning or simply adding tools to what already exists.
• Automates existing broken processes rather than questioning them
• AI is bolted onto legacy workflows as an extra step, not a replacement for the process
• Success is measured by tool adoption — how many people are using it — not by business outcomes
• Pilots run, generate some interest, and then quietly die with no clear path to scale
• Approval bottlenecks and unclear ownership kill momentum before results can compound
• Interrogates the process itself before deciding whether or how to automate any part of it
• AI replaces the process — or significant parts of it — rather than being added on top of what already exists
• Business outcomes are measured from day one, not tool usage statistics
• Pilots are designed with a defined trigger for scale — when specific outcomes are hit, production deployment follows automatically
• Fast feedback loops are built into the design, so what isn’t working gets caught and corrected quickly
The highest-value AI implementations replace business processes entirely, not just streamline them.
— Iansiti & Lakhani, ‘Competing in the Age of AI’, MIT Sloan
AI pilots fail to reach production most often because they were never designed with a clear path to scale in the first place. A pilot run in isolation, evaluated on enthusiasm rather than measured business outcomes, with no predefined trigger for what happens if it works, tends to generate interest and then stall — not because it failed, but because nobody built the bridge from pilot to production.
Gartner’s research found more than half of generative AI proof-of-concepts get shelved before reaching production. The lowest maturity level on this question in the Index describes the most common starting point: we have run pilots but most have not progressed to production. The highest level describes a structurally different approach: pilots are designed with a defined scale trigger — when specific outcomes are hit, production deployment is automatic.
That distinction — a predefined trigger versus an open-ended ‘let’s see how it goes’ — is often the entire difference between a pilot that scales and one that becomes a slide in last year’s strategy deck.
AI impact in revenue processes should be measured against specific commercial outcomes — pipeline velocity, deal size, conversion rate, retention, cost to serve — defined before the initiative begins, not retrofitted afterwards to justify the spend.
Most organisations measure the wrong thing. They track tool adoption: how many licences are active, how many people logged in this week, how many queries were run. Those are activity metrics, not outcome metrics. An organisation can have high tool adoption and zero commercial impact, because adoption measures whether people are using something, not whether it’s working.
The maturity marker here is measuring AI impact against revenue, pipeline, and customer metrics from the outset of every initiative — not retrofitting an ROI story once leadership asks what the investment delivered.
Process redesign sits downstream of technology architecture (Pillar 3) and upstream of people enablement (Pillar 5) — and weaknesses in either direction show up here first. Buying well-integrated AI tools without redesigning the process they sit in produces automation of existing inefficiency — a faster version of a process that shouldn’t exist in its current form. And a redesigned process that the team hasn’t been enabled to use, or doesn’t trust, fails in implementation regardless of how well it was designed on paper.
This is also the pillar most directly shaped by executive mandate (Pillar 1). Genuine process redesign requires the authority to change how teams work, reassign accountability, and remove steps that people may be attached to. Without a leader willing to exercise that authority, process improvement stays at the level of individual initiative — the lowest maturity level on this pillar — rather than becoming a structured, organisation-wide capability.
The AI Maturity Index measures process redesign across how AI has been applied to revenue processes, how impact is measured, the fate of pilots, the depth of workflow change, and the discipline behind prioritisation. Most organisations score lower here than they expect — not because they lack ambition, but because automation is easier to start and redesign is the harder, more valuable work.
Free · 10 minutes · Instant results · Personalised report from g2m’s AI go-to-market specialists
Process automation applies AI to speed up existing workflow steps without changing the underlying process. Process redesign interrogates why the process exists in its current form and rebuilds it around what AI makes possible — often removing steps entirely rather than just accelerating them. Automation produces incremental efficiency; redesign produces structural change that shows up in commercial outcomes.
Most AI pilots fail to reach production because they lack a predefined path to scale. They are evaluated on enthusiasm or activity rather than measured business outcomes, with no clear trigger for what happens if the pilot succeeds. Gartner research found more than half of generative AI proof-of-concepts get shelved before reaching production. Pilots designed with a defined scale trigger — automatic production deployment once specific outcomes are hit — are far more likely to succeed.
AI impact on revenue processes should be measured against specific commercial outcomes defined before the initiative begins — pipeline velocity, deal size, conversion rate, retention, or cost to serve. Many organisations instead measure tool adoption metrics such as licence usage or login frequency, which indicate activity but not commercial value. Mature organisations tie AI impact to revenue, pipeline, and customer metrics from the outset of every initiative.
Process redesign scores low across AI maturity research because it requires more organisational effort and authority than simply buying and deploying AI tools. Automating an existing process is faster and feels less risky than questioning whether the process should exist in its current form at all. RAND Corporation found that 80.3% of AI projects fail to deliver intended value, identifying process — not technology — as the primary root cause.
• Part 1: Who Should Own AI in Your Organisation? The Case for an Executive Mandate
• Part 2: Why Your Data Is Probably Your Biggest AI Constraint
• Part 3: Why Buying AI Tools Isn’t the Same as Having a Strategy
• Part 4: The Difference Between Automating and Redesigning
• Part 5: Why Training Events Don’t Build AI Capability
Or start with the overview:
→ We’re Building Australia’s First AI Maturity Benchmark — Here’s Why We Need You In It