More
Call to Insights
To Know
-
Start again
Rethinking how we work before adding AI
Call to Insights
—
Nov 2026
Most companies put AI on top of the processes they already have. Our Leverages started by asking which steps should exist at all.
- Miguel Graça
At their recent All Hands, our People team opened with a slide that read "We thought we knew the process… then we looked at how it really works." They shared how they’re redesigning some of their own processes with AI, focusing on what they found when they stopped describing their work from memory and watched it happen.
People is one of the five Leverages doing this work, alongside Finance, Marketing, Legal and Contract Management. Last edition's piece on the AI for Productivity Academy explored how these teams learned to use AI in their day-to-day work. This piece covers the next step: using AI to change the process itself.
The gap between using AI and changing how work gets done is showing up well beyond our own experience.
McKinsey's 2025 global survey found that 88% of organisations use AI in at least one function, but only about 6% get 5% or more of their EBIT from it. Workflow redesign was one of the factors most strongly linked to that result, with high performers about three times more likely to redesign workflows when deploying AI.
MIT's NANDA initiative published a harsher number the same year. About 95% of the enterprise generative AI pilots it studied showed no measurable effect on profit and loss, pointing to how companies fitted the tools into existing workflows.
Theodoros Evgeniou of INSEAD places AI's impact at five levels: task,job, process, organisation and value proposition. While task automation is already mature, the gains still to come sit at the process level and above.
A model placed on top of an unchanged process just runs the same steps faster, including the ones nobody needed.
Since last spring, a small team has worked with colleagues across these functions to redesign their processes, an initiative called internally Spark for Leverages. Each process starts with half a day spent with the people who run it.
Before anyone draws a map, the group agrees on what the process actually is: Who provides what, what goes in, what comes out and who receives the result. Then they watch the work happen, under one rule: Show me, don't tell me.
What looked like a few simple steps turned out to involve emails, spreadsheets, manual checks, handovers, and waiting. The team mapped this real version and tagged each step with one of lean manufacturing's seven wastes, such as waiting, motion or over-processing. A dot vote then selected the first three areas to redesign.
The questions at each step were simple. Why does this step exist? Why do we type the same information twice? Why does the process stop until someone remembers to send an email? Some steps disappeared. Others were merged. Only then did technology enter the conversation.
In one redesigned process, a single entry now starts the whole workflow and pre-fills information colleagues used to enter by hand. Other processes are being tested or mapped. Each one builds on the methods and components developed before, making the next redesign easier.
Every redesign depends on the people who run the process, because they know how it really works. But involving them is about more than understanding the process. It is also about ownership. Research from the National University of Singapore and Emory University found that, when asked to share their expertise with AI, workers became more concerned about losing control, being replaced or having less say in how their knowledge was used when AI was involved.
That is why the small team is not meant to stay. The goal is for each function to rethink its own processes. Process owners learn the method through workshops and training, keeping control of how their processes change.
At the All Hands, our People colleagues were already doing that: presenting the work as their own, explaining the method and taking its lessons beyond the initiative.
When we talk with clients about AI-driven business efficiency, we make the same argument: change how the work is done, not just the technology. The method is the one described here: observe the real process, identify the waste, redesign, and only then choose the technology.
Watching the process also captures what an agent needs to know: who makes each decision, which rules apply and where the exceptions are. When we help clients build ontologies, we turn that knowledge into a form agents can act on. We now apply the same approach to our own processes.
When a client asks whether it works, we can show them our own processes and the people who redesigned them can share what they learned.
McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation, November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, August 2025, as reported by Fortune. https://finance.yahoo.com/news/mit-report-95-generative-ai-105412686.html
Theodoros Evgeniou (INSEAD) and NTT DATA Global AI Office, Impact of AI on the IT industry: Talent transformation, March 2026. https://www.insead.edu/system/files/2026-03/INSEADNTT_AIREPORT.pdf
Tianqi Song, Zicheng Zhu, Yi-Chieh Lee (National University of Singapore) and Hancheng Cao (Emory University), When My Skill Becomes Agent Skill: How Knowledge Workers Share Their Expertise with AI Systems, October 2026, preprint. https://arxiv.org/abs/2610.09608
Most companies put AI on top of the processes they already have. Our Leverages started by asking which steps should exist at all.
- Miguel Graça
At their recent All Hands, our People team opened with a slide that read "We thought we knew the process… then we looked at how it really works." They shared how they’re redesigning some of their own processes with AI, focusing on what they found when they stopped describing their work from memory and watched it happen.
People is one of the five Leverages doing this work, alongside Finance, Marketing, Legal and Contract Management. Last edition's piece on the AI for Productivity Academy explored how these teams learned to use AI in their day-to-day work. This piece covers the next step: using AI to change the process itself.
When we talk with clients about AI-driven business efficiency, we make the same argument: change how the work is done, not just the technology. The method is the one described here: observe the real process, identify the waste, redesign, and only then choose the technology.
Watching the process also captures what an agent needs to know: who makes each decision, which rules apply and where the exceptions are. When we help clients build ontologies, we turn that knowledge into a form agents can act on. We now apply the same approach to our own processes.
When a client asks whether it works, we can show them our own processes and the people who redesigned them can share what they learned.
McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation, November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, August 2025, as reported by Fortune. https://finance.yahoo.com/news/mit-report-95-generative-ai-105412686.html
Theodoros Evgeniou (INSEAD) and NTT DATA Global AI Office, Impact of AI on the IT industry: Talent transformation, March 2026. https://www.insead.edu/system/files/2026-03/INSEADNTT_AIREPORT.pdf
Tianqi Song, Zicheng Zhu, Yi-Chieh Lee (National University of Singapore) and Hancheng Cao (Emory University), When My Skill Becomes Agent Skill: How Knowledge Workers Share Their Expertise with AI Systems, October 2026, preprint. https://arxiv.org/abs/2610.09608
Most companies put AI on top of the processes they already have. Our Leverages started by asking which steps should exist at all.
- Miguel Graça
At their recent All Hands, our People team opened with a slide that read "We thought we knew the process… then we looked at how it really works." They shared how they’re redesigning some of their own processes with AI, focusing on what they found when they stopped describing their work from memory and watched it happen.
People is one of the five Leverages doing this work, alongside Finance, Marketing, Legal and Contract Management. Last edition's piece on the AI for Productivity Academy explored how these teams learned to use AI in their day-to-day work. This piece covers the next step: using AI to change the process itself.
The gap between using AI and changing how work gets done is showing up well beyond our own experience.
McKinsey's 2025 global survey found that 88% of organisations use AI in at least one function, but only about 6% get 5% or more of their EBIT from it. Workflow redesign was one of the factors most strongly linked to that result, with high performers about three times more likely to redesign workflows when deploying AI.
MIT's NANDA initiative published a harsher number the same year. About 95% of the enterprise generative AI pilots it studied showed no measurable effect on profit and loss, pointing to how companies fitted the tools into existing workflows.
Theodoros Evgeniou of INSEAD places AI's impact at five levels: task,job, process, organisation and value proposition. While task automation is already mature, the gains still to come sit at the process level and above.
A model placed on top of an unchanged process just runs the same steps faster, including the ones nobody needed.
Since last spring, a small team has worked with colleagues across these functions to redesign their processes, an initiative called internally Spark for Leverages. Each process starts with half a day spent with the people who run it.
Before anyone draws a map, the group agrees on what the process actually is: Who provides what, what goes in, what comes out and who receives the result. Then they watch the work happen, under one rule: Show me, don't tell me.
What looked like a few simple steps turned out to involve emails, spreadsheets, manual checks, handovers, and waiting. The team mapped this real version and tagged each step with one of lean manufacturing's seven wastes, such as waiting, motion or over-processing. A dot vote then selected the first three areas to redesign.
The questions at each step were simple. Why does this step exist? Why do we type the same information twice? Why does the process stop until someone remembers to send an email? Some steps disappeared. Others were merged. Only then did technology enter the conversation.
In one redesigned process, a single entry now starts the whole workflow and pre-fills information colleagues used to enter by hand. Other processes are being tested or mapped. Each one builds on the methods and components developed before, making the next redesign easier.
Every redesign depends on the people who run the process, because they know how it really works. But involving them is about more than understanding the process. It is also about ownership. Research from the National University of Singapore and Emory University found that, when asked to share their expertise with AI, workers became more concerned about losing control, being replaced or having less say in how their knowledge was used when AI was involved.
That is why the small team is not meant to stay. The goal is for each function to rethink its own processes. Process owners learn the method through workshops and training, keeping control of how their processes change.
At the All Hands, our People colleagues were already doing that: presenting the work as their own, explaining the method and taking its lessons beyond the initiative.
When we talk with clients about AI-driven business efficiency, we make the same argument: change how the work is done, not just the technology. The method is the one described here: observe the real process, identify the waste, redesign, and only then choose the technology.
Watching the process also captures what an agent needs to know: who makes each decision, which rules apply and where the exceptions are. When we help clients build ontologies, we turn that knowledge into a form agents can act on. We now apply the same approach to our own processes.
When a client asks whether it works, we can show them our own processes and the people who redesigned them can share what they learned.
McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation, November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, August 2025, as reported by Fortune. https://finance.yahoo.com/news/mit-report-95-generative-ai-105412686.html
Theodoros Evgeniou (INSEAD) and NTT DATA Global AI Office, Impact of AI on the IT industry: Talent transformation, March 2026. https://www.insead.edu/system/files/2026-03/INSEADNTT_AIREPORT.pdf
Tianqi Song, Zicheng Zhu, Yi-Chieh Lee (National University of Singapore) and Hancheng Cao (Emory University), When My Skill Becomes Agent Skill: How Knowledge Workers Share Their Expertise with AI Systems, October 2026, preprint. https://arxiv.org/abs/2610.09608
.png)
