- Process transformation was already hard before GenAI. Few do it consequentially.
- Tutorials look great on YouTube, but they fail in a business environment. Governance, security, and knowledge management are needed, but hardly covered.
- Expert skills remain scarce, because few have done the AI work in a business. Don’t confuse influencers with actual experts.
- Instead of productivity, measuring KPIs and PPIs is the better approach that is actually tied to a goal that business leaders measure, too, and care about.
- Business users are not developers. The bar is pretty high as the agentic platforms are not exactly self-service or provide the dead simple, reliable results that would be needed. Discoverability and interoperability remain behind expectations.
These are just a few of my thoughts on this topic…
"Perhaps it’s just too early for us to see AI-driven productivity gains, but how long do we need to wait?" Maybe longer than we expect given our history of mis-understanding technology trends. I met Hal Varian for the first time in late 1999 when he came to Boston to participate in the meeting of a committee charged with solving the "productivity paradox." I told him they were wrong to start measuring the gap between spending and the (expected) increase in productivity from the advent of PCs in the early 1980s. I argued that the real productivity impact occurred much later with the implementation and widespread use of local area networks, getting so much more from PCs by connecting them. That took time--when the first LAN was installed at NORC (in 1985) and I asked why do we need it, the answer was: So we can print on the laser printer in Dick's office (rather than walk a few yards and stick a floppy disk in his PC). It took another ten years for organizations to find and adopt productivity-enhancing uses of LANs, resulting in the increase in productivity in the second half of the 1990s.
Agreed 100%. I'm a retired tech-bro turned government CIO, so my perspective is on an area AI is going to have a huge impact, if we architect it right.
I think SME already see the positive impact from AI implemented especially in demand generation, content creation worklows. SME are typically tight on human resources, when one person does several roles or even everything in case of a solo-entrepreneur. In such cases AI could significantly increase the business outcome. Big enterprises are completely different story, where the resistance of the employees might one of a serious blockers.
Tom, It has been a while since we connected and good to see your provocative reasons highlighted here (I just recently joined Substack). It absolutely tracks well with what I've been calling the "missing layer" problem (see my post https://augmentedhealthintelligence.substack.com/p/the-missing-layer).
Most Gen AI productivity efforts are focused on automation of tasks and not really the end to end decision process. Workslop isn't a quality failure, it's an unassigned decision — nobody owns the "is this good enough to pass on" call, so it either doesn't get made or gets made redundantly downstream. It is difficult to measure productivity gains from a decision you never explicitly designed rights around. I think the redesign challenge you highlight is mostly around decision governance and architecture, not process flow — and that's a smaller, faster problem than rebuilding entire workflows, if organizations would actually name it as the target to focus on.
I am also thinking back to what we can learn from prior tech waves and apply to the current era. ERP, Analytics and AI have defined the past 30 years and as you and I have discussed on a number of occasions it is not just about the tech but rather the changes in process and org that are harder. I would also argue that now, more than ever, is the era where we need to examine decision architecture much more deliberately in order to get the value that should be there.
Hey Professor Davenport! Been a bit since I sat in your classroom at BU. This is a strong critique of the "individual productivity" trap!
The point about personal infrastructure is particularly critical. True process-level productivity requires that organizations have a systematic way to identify exactly where process friction, cognitive inefficiencies, fatigue, and administrative waste live across workflows.
And...this is exactly the gap we are bridging at Fullstory with our Workforce solution. By capturing aggregate, objective employee behavioral data across internal applications, we reveal the ground-truth digital bottlenecks holding teams back.
When you pair this deep behavioral context with our MCP integration, the game changes. MCP acts as the standardized bridge to connect that rich, human-centric workflow data directly to LLMs. This allows organizations to dynamically map automation opportunities and "agentify" with confidence. This supports shifting from fragmented individual prompts to targeted, high-value enterprise agents that automate or augment entire end-to-end processes.
Real productivity gains don't come from telling people to work faster with a chatbot, it comes from using behavioral data to systematically redesign and automate the workflows themselves, or as someone else said below, "how work actually moves".
The list resonates, but I'd add a reason that cuts across several of yours: governance lag.
Productivity improvements require organizations to change how decisions are made, how work is assigned, and who is accountable for outcomes. AI tools can accelerate tasks, but they can't resolve the institutional ambiguities they create. When AI generates a recommendation, who owns the decision? When it speeds up a process, does headcount stay the same, redeploy, or get cut? Those aren't technology questions.
Every one of those unresolved questions creates friction that slows adoption or dilutes benefit. The organizations that see productivity gains are the ones that have answered them — not because they got better AI, but because they got clearer governance.
Asking if AI has improved firm-level productivity is akin to asking whether the introduction of Microsoft Excel did the same. Yet, the lack of a definitive answer on that front has not kept enterprises from adopting Excel and related “productivity” software.
That is correct, Doug. But I don't think that our economic growth and stock market valuations were closely tied to Excel. I am certainly not arguing that genAI isn't useful, just that it is hard to measure value from it.
This is the bit many people skip. Giving everyone AI is easy. Changing how work actually moves is the hard part. The real opportunity begins when the excitement around the tool turns into a better way of working.
It’ll be hard for existing companies to reengineer processes, but new ones have an opportunity to build from the ground up with these capabilities in mind.
In the same way small companies disrupted huge ones over the past several decades, brought on by advantages of building new era software, we will probably see a similar pattern play out by AI-natives. It’s just going to take time and new, good ideas to gain traction.
This resonates in so many levels:
- Process transformation was already hard before GenAI. Few do it consequentially.
- Tutorials look great on YouTube, but they fail in a business environment. Governance, security, and knowledge management are needed, but hardly covered.
- Expert skills remain scarce, because few have done the AI work in a business. Don’t confuse influencers with actual experts.
- Instead of productivity, measuring KPIs and PPIs is the better approach that is actually tied to a goal that business leaders measure, too, and care about.
- Business users are not developers. The bar is pretty high as the agentic platforms are not exactly self-service or provide the dead simple, reliable results that would be needed. Discoverability and interoperability remain behind expectations.
These are just a few of my thoughts on this topic…
"Perhaps it’s just too early for us to see AI-driven productivity gains, but how long do we need to wait?" Maybe longer than we expect given our history of mis-understanding technology trends. I met Hal Varian for the first time in late 1999 when he came to Boston to participate in the meeting of a committee charged with solving the "productivity paradox." I told him they were wrong to start measuring the gap between spending and the (expected) increase in productivity from the advent of PCs in the early 1980s. I argued that the real productivity impact occurred much later with the implementation and widespread use of local area networks, getting so much more from PCs by connecting them. That took time--when the first LAN was installed at NORC (in 1985) and I asked why do we need it, the answer was: So we can print on the laser printer in Dick's office (rather than walk a few yards and stick a floppy disk in his PC). It took another ten years for organizations to find and adopt productivity-enhancing uses of LANs, resulting in the increase in productivity in the second half of the 1990s.
Agreed 100%. I'm a retired tech-bro turned government CIO, so my perspective is on an area AI is going to have a huge impact, if we architect it right.
https://thegoodideafairy.substack.com/p/how-to-hack-a-government-agency-for
I think SME already see the positive impact from AI implemented especially in demand generation, content creation worklows. SME are typically tight on human resources, when one person does several roles or even everything in case of a solo-entrepreneur. In such cases AI could significantly increase the business outcome. Big enterprises are completely different story, where the resistance of the employees might one of a serious blockers.
Tom, It has been a while since we connected and good to see your provocative reasons highlighted here (I just recently joined Substack). It absolutely tracks well with what I've been calling the "missing layer" problem (see my post https://augmentedhealthintelligence.substack.com/p/the-missing-layer).
Most Gen AI productivity efforts are focused on automation of tasks and not really the end to end decision process. Workslop isn't a quality failure, it's an unassigned decision — nobody owns the "is this good enough to pass on" call, so it either doesn't get made or gets made redundantly downstream. It is difficult to measure productivity gains from a decision you never explicitly designed rights around. I think the redesign challenge you highlight is mostly around decision governance and architecture, not process flow — and that's a smaller, faster problem than rebuilding entire workflows, if organizations would actually name it as the target to focus on.
I am also thinking back to what we can learn from prior tech waves and apply to the current era. ERP, Analytics and AI have defined the past 30 years and as you and I have discussed on a number of occasions it is not just about the tech but rather the changes in process and org that are harder. I would also argue that now, more than ever, is the era where we need to examine decision architecture much more deliberately in order to get the value that should be there.
Hey Professor Davenport! Been a bit since I sat in your classroom at BU. This is a strong critique of the "individual productivity" trap!
The point about personal infrastructure is particularly critical. True process-level productivity requires that organizations have a systematic way to identify exactly where process friction, cognitive inefficiencies, fatigue, and administrative waste live across workflows.
And...this is exactly the gap we are bridging at Fullstory with our Workforce solution. By capturing aggregate, objective employee behavioral data across internal applications, we reveal the ground-truth digital bottlenecks holding teams back.
When you pair this deep behavioral context with our MCP integration, the game changes. MCP acts as the standardized bridge to connect that rich, human-centric workflow data directly to LLMs. This allows organizations to dynamically map automation opportunities and "agentify" with confidence. This supports shifting from fragmented individual prompts to targeted, high-value enterprise agents that automate or augment entire end-to-end processes.
Real productivity gains don't come from telling people to work faster with a chatbot, it comes from using behavioral data to systematically redesign and automate the workflows themselves, or as someone else said below, "how work actually moves".
The list resonates, but I'd add a reason that cuts across several of yours: governance lag.
Productivity improvements require organizations to change how decisions are made, how work is assigned, and who is accountable for outcomes. AI tools can accelerate tasks, but they can't resolve the institutional ambiguities they create. When AI generates a recommendation, who owns the decision? When it speeds up a process, does headcount stay the same, redeploy, or get cut? Those aren't technology questions.
Every one of those unresolved questions creates friction that slows adoption or dilutes benefit. The organizations that see productivity gains are the ones that have answered them — not because they got better AI, but because they got clearer governance.
Asking if AI has improved firm-level productivity is akin to asking whether the introduction of Microsoft Excel did the same. Yet, the lack of a definitive answer on that front has not kept enterprises from adopting Excel and related “productivity” software.
That is correct, Doug. But I don't think that our economic growth and stock market valuations were closely tied to Excel. I am certainly not arguing that genAI isn't useful, just that it is hard to measure value from it.
This is the bit many people skip. Giving everyone AI is easy. Changing how work actually moves is the hard part.
The real opportunity begins when the excitement around the tool turns into a better way of working.
I agree, Vinod. I wonder how long that will take?
For sure longer than it should. Give us a new tool and our first instinct is usually to recreate the old problem more efficiently.
This is the bit many people skip. Giving everyone AI is easy. Changing how work actually moves is the hard part. The real opportunity begins when the excitement around the tool turns into a better way of working.
It’ll be hard for existing companies to reengineer processes, but new ones have an opportunity to build from the ground up with these capabilities in mind.
In the same way small companies disrupted huge ones over the past several decades, brought on by advantages of building new era software, we will probably see a similar pattern play out by AI-natives. It’s just going to take time and new, good ideas to gain traction.