The time was saved.Where did it go?
MIT researchers studying 300 enterprise AI deployments found that approximately 95 percent produced no measurable profit and loss impact. McKinsey adds that only 6 percent of organizations can trace more than 5 percent of their operating profit directly to AI. Today's edition follows the distance between vendor promise and measured workplace outcome, and asks what actually changes human capability inside organizations.
95%of 300 enterprise AI deployments studied by MIT researchers generated no measurable profit and loss impact. The figure is approximate.
6%of organizations can trace more than 5 percent of their operating profit directly to AI, according to McKinsey.
The AI treadmill trap
Motion is not progress
At the level of the individual task, artificial intelligence works.
David Smailes, an executive coach and learning strategist, makes the case in a piece published by Training Magazine. He gives the pattern a name: the AI treadmill trap. He points to a study of more than 5,000 customer support agents. Working with an AI assistant, they resolved 15 percent more customer issues per hour.
Time was saved. The question for learning and business leaders is where it went. In most organizations it is absorbed by organizational slack. People complete more things, and complete them faster, but the extra activity does not turn into revenue or bottom-line value.
Organizations are confusing motion with progress. Exposing an employee to a faster tool does not develop workforce competence, and it does not automatically create commercial value. Handing people faster tools without strategic direction can simply help them execute flawed strategies faster.
Where the saved time goes
Sixteen minutes, and a decision about them
A multi-site study published in JAMA examined ambient AI medical scribes: listening tools that sit in the examination room, hear the conversation between physician and patient, and generate the clinical documentation automatically.
The technology worked as designed. It saved doctors about 16 minutes of documentation time every day. Set the plan for those minutes below and watch where they land.
The minutes are waiting for a destination.
Move the slider or choose a plan. Nothing is assumed: the study describes what happened when the minutes were left alone and what happened when they were redirected on purpose.
Reported after the workflow redesign
The sentence that makes it a plan
When this role gets back 16 minutes a day, those minutes will be used for scheduling additional patient interactions.
At first, that daily saving yielded no financial or operational gain for the hospitals. Documentation time dropped, but after-hours administrative work barely changed and financial returns were negligible.
Revenue appeared only when health systems including The Permanente Medical Group and the University of California, San Francisco explicitly redesigned their clinical workflows. They changed the operating system around the tool, and deliberately redirected the freed minutes toward scheduling additional patient interactions.
Once management redirected that capacity, participating physicians logged 1.81 more relative value units per week. A relative value unit, or RVU, is the standardized measure healthcare systems and insurers use to calculate reimbursement for physician services. The same physicians handled 2.8 percent more patient visits without any rise in billing claim denials. Across roughly 700 clinicians, the redesign added $2.1 million in new annual revenue.
Deploying AI without redesigning the workflow is like buying a faster car but staying stuck in the same traffic jam.
Same technology, different position
Who stands between the AI and the customer
Smailes compares how companies architect their deployments. Three of them sit at very different distances from the customer. Move the AI along the line and watch the people between it and the person on the other end.
Lumen kept the AI strictly upstream
Lumen Technologies gave its AI to the B2B sales team for pre-call customer research. Anyone who has worked in enterprise sales knows the burden: understanding a prospect's corporate footprint, reviewing its organizational structure and parsing its recent regulatory filings. Preparing for a single prospect call used to take a Lumen sales professional four hours. Generative AI cut that research to 15 minutes.
The human stayed in the final mile. The client conversation remained human to human, and the reps spent their effort on high-value conversations. Lumen's estimate is $50 million in annual revenue.
Estimated $50 million in annual revenue. Preparation time cut from 4 hours to 15 minutes.
JPMorgan Chase made the human the reviewer
JPMorgan Chase redesigned the workflow so that the human becomes a critical reviewer rather than an original creator. Its contract intelligence system turned highly compensated corporate attorneys into final-mile reviewers of AI-extracted contract attributes.
The result: 360,000 hours of manual legal review saved across 12,000 commercial credit agreements each year.
Reported 360,000 hours of manual legal review saved each year, across 12,000 commercial credit agreements.
Klarna put the AI in front of the customer
Klarna decided to replace human customer service agents with customer-facing AI. Internally it projected a $40 million profit improvement. It looked brilliant on paper and compelling in executive presentations.
Then the company reversed course and deliberately restored its human customer service options. The AI system had eroded customer trust, and a cheaper interaction becomes extraordinarily expensive when it permanently damages the relationship with the client.
Projected $40 million profit improvement. Company later restored human customer service options.
How the technology is introduced
One door, or every door
Samsung: open access
Samsung gave engineers unstructured, open-ended access to ChatGPT and told them to explore and experiment on their own. Across three separate incidents, engineers accidentally pasted proprietary source code and sensitive internal meeting transcripts into the public model.
The lack of governance ended in a ban on generative AI tools across company devices.
Bayer: one chosen problem
Bayer did not hand out logins and tell people to figure it out. Leadership ran a structured effort that identified 700 specific operational pain points directly from frontline teams. Rather than attempt all of them, it targeted one: role-play coaching for pharmaceutical sales representatives.
It built a dedicated, secure AI system for 1,000 representatives and went from concept to launch in 72 days. That single intervention saved more than 6,000 training hours annually and improved the efficiency of one-on-one role-play coaching by more than 50 percent. Bayer did not train people on the tool as an abstract technology. It used it on a specific, high-friction human problem.
Choose one of the 700. Which would you have started with?
Bayer's chosen pain point is lit: role-play coaching for pharmaceutical sales representatives. The other 699 were set aside for now.
What learning leaders can do
If the tool works at the task level, how does L&D fix the workflow?
Smailes lays out four structural interventions. Each one builds on the last.
Start in the back end
Begin with back-end workflows where internal human experts can catch errors before a customer or client ever sees them. That avoids the trust erosion Klarna experienced.
Aim at friction, not at open exploration
Target high-friction, repetitive administrative tasks, as Lumen Technologies did with pre-call customer research. Do not grant open-ended experimentation with no clear business problem, which is the pattern behind the data leaks at Samsung.
Make the human the reviewer
Redesign the workflow so the human becomes a critical reviewer rather than an original creator, as JPMorgan Chase did with its contract intelligence system.
Write down where the hours go
The most important principle for the L&D profession: mandate a written time reinvestment plan before rolling out training. Before anyone is trained to use an AI tool, the organization answers one sentence in writing.
"When this role gets back X hours per week, those hours will be used for Y."
Without a plan and governance for the freed capacity, it evaporates. It is absorbed by organizational slack and turns into more internal emails, longer meetings and invisible administrative busywork. Try it with the sixteen minutes.
Unintended operational consequences
The cost nobody put in the training plan
The environmental footprint of AI is moving from an IT infrastructure concern to an everyday workforce capability issue.
Sustained Futures has released Future Skills A-Z, an operational guide for L&D and human resources teams. It maps 78 foundational terms across three domains: climate fluency, artificial intelligence skills and what it calls a future mindset. For workplace learning, the core term is inference energy: the computation and electricity used every time a model runs and generates an output.
The guide cites a 2026 analysis from the International Energy Agency. Software engineering has achieved efficiency gains for basic text generation. But when an employee asks for complex reasoning, multi-step agentic workflows, deep research or video generation, a single query can consume hundreds to thousands of times more energy than a simple text prompt.
Match the task to the tool
Prompt engineering has usually been taught as a speed or quality skill: how to format a prompt to get the best answer in the shortest time. This research reframes prompt hygiene as an operational sustainability matter. Taking 30 seconds to refine a prompt, instead of running brute-force, iterative queries until the system stumbles onto what is needed, becomes an environmental responsibility as well as a technical habit.
The mechanism reaches the corporate level. Picture an enterprise of 10,000 employees haphazardly starting energy-intensive, multi-agent reasoning chains for simple daily tasks that a lightweight older model or a basic database lookup could resolve in a fraction of a second. Data center power demand and corporate carbon emissions climb quickly. Training employees to judge whether they need a heavy analytical model or a basic retrieval tool ties environmental governance to everyday digital literacy, and links individual behavior to net-zero targets and ESG commitments. Prompt engineering becomes a matter of corporate resource management.
In force today, 30 October 2026
From reasonable steps to all reasonable steps
The United Kingdom's updated statutory prevention duty comes into legal force today under the Employment Rights Act 2025.
Employers were previously required to take reasonable steps to prevent workplace sexual harassment. The updated legislation raises that burden to all reasonable steps. The word carries weight. Legal scrutiny shifts from how an HR department reacts after an incident to whether the organization built proactive, systemic prevention beforehand.
The most important change for anyone managing client-facing or customer-service teams is third-party liability. If a client, customer, supplier or independent contractor harasses an employee, the employer can be held legally liable unless it can prove it took all reasonable steps to prevent it. And if an employment tribunal upholds a sexual harassment claim and finds the duty was breached, it can increase the compensation awarded to the victim by up to 25 percent.
Who does the training cover?
In response, HR and learning software provider Ciphr updated its 20-minute e-learning course. It now includes scenarios on third-party harassment by clients and contractors, and the completion data and assessment scores connect directly and permanently to the employee's central HR record.
Everyone knows the compliance routine: click play on a mandatory video, mute the tab, answer email in another window until the multiple-choice quiz appears. Course completion is not capability. Yet before an employment tribunal, an audited, timestamped completion record can serve as the organization's legal shield against a 25 percent uplift. That is the paradox of the profession. Completion does not equal behavior change, but an audited record that an employee understood the policy is the baseline of institutional defense.
Because of third-party liability, compliance training can no longer look only inward at how colleagues behave in the office. It has to teach customer-facing employees how to hold professional boundaries with external clients, how to handle power dynamics when a client oversteps, and how to reach safe escalation channels without fearing the loss of a commission or damage to a major business contract.
After the training ends
Trained, then back to the spreadsheet
Think of the last time an organization rolled out a new platform, CRM or protocol. Employees check the box, pass the post-test and return to their desks. Many go right back to what they were doing before.
A TalentLMS study called Habit Hugging surveyed 1,500 full-time US employees who had completed training on a new tool or workflow within the past 12 months.
42%
are still using legacy tools, manual processes and shadow spreadsheets that their managers believe have been permanently retired.
What employees do after training
55% say the old way feels more familiar
48% say the old way is faster
60% say immediate results matter more at their company than following the new process
45% avoided questioning a flawed new tool so they would not look resistant to change
When adoption fails, leadership usually blames clunky software or stubborn employees. The survey points somewhere else. Corporate incentives routinely reward immediate daily output over the natural, temporary slowdown of a learning curve. With a production quota due Friday afternoon, the hidden spreadsheet beats the unfamiliar enterprise interface. This is not a failure of the software, and it is not a failure of the learner. It is employees protecting themselves, rationally.
The manager multiplier
49of every 100 employees consistently use the new workflow
When the direct manager actively uses the new tool, consistent adoption sits at 49 percent.
What this shows: a survey comparison of 1,500 employees. What it does not show: that manager behavior alone causes the difference.
Exposing employees to digital content does not develop competence if manager behavior does not reinforce it and workloads leave no room to practice. Only 36 percent of the companies in the survey check whether skills are being applied on the job. L&D cannot be only a content delivery mechanism. It has to intervene in workflow integration, psychological safety and manager accountability.
The missing variable
The answer arrives in two seconds. Who judges it?
When adoption fails because of habits, and AI deployments stall because workflows stay the same, the missing variable is foundational human capability.
That premise anchors an executive roundtable Sponge is hosting on 12 November with senior leaders from The LEGO Group, Deloitte, Fosway Group and 8Levers. Its core data point comes from the World Economic Forum's Future of Jobs Report: 63 percent of employers cite skills gaps as the number-one barrier to organizational transformation.
Select a block, then select a gap in the span. Dragging also works.
The skills employers lack are not only technical. Analytical thinking is now employers' leading core capability, accelerating alongside creative thinking, resilience and lifelong learning.
Much training goes into the mechanics of tools: which prompt formula, which interface settings, which plugins. The World Economic Forum data points to a different bottleneck: the human critical thinking needed to evaluate what the tool produces. When an AI system returns a detailed, confident answer in two seconds, the employee needs the analytical skill to know whether it is accurate, biased or completely hallucinated.
As AI automates routine cognitive tasks, the value of an employee shifts toward metacognitive skills: critically evaluating model errors, managing the ethical boundaries of client data, and synthesizing cross-functional insights that isolated software platforms cannot see. The argument for L&D is a shift from tool-centric training to human-centric enablement. Budgets that concentrate on these high-judgment capabilities build more durable value than superficial familiarity with software interfaces that will change within six months.
Policy this week
Skills funding in one direction, payroll costs in the other
The UK Autumn Budget, delivered on 28 October, committed £100 million to double the size of mayoral apprenticeship brokerage pilots. They expand from 7 to 14 English regions across 2027 to 2029, to help small and medium enterprises hire young apprentices. The government also restored the Union Learning Fund with £15 million annually and allocated £83 million to AI economic transition research and Pathfinder projects that help manage workforce shifts.
Neil Carberry, chief executive of the CIPD, issued a warning about the wider policy environment. Skills funding and AI transition planning, he said, are taking place in an isolated policy silo, while the Budget raises employment taxes, including employer National Insurance contributions, alongside rising regulatory burdens.
Those forces pull in opposite directions. An administration can subsidize apprenticeships, but if the wider tax environment discourages a company from adding headcount in the first place, the training subsidy cannot overcome payroll realities. Learning strategy cannot be separated from the economics of employment.
Mayoral apprenticeship brokerage pilots, English regions
7 regions
Learning technology
Two architectures, two pressures
At the Moodle Global Partner Summit in Barcelona, more than 100 learning technology leaders from 70 countries gathered. CEO Scott Anderberg's keynote centered on the intersection of artificial intelligence, data sovereignty and open-source architecture. Moodle is expanding its certified integration ecosystem and marketplace while keeping the core platform open source.
Separately, Bridge LMS was named a 2026 Top 20 LMS company by Training Industry for a fully connected approach: unifying learning delivery, skills graphs and manager-led performance enablement in one closed-loop system.
The two platforms answer different pressures. Moodle's emphasis on data sovereignty addresses the risk of being locked into closed, proprietary AI clouds where enterprise intellectual property is harvested, and keeps custody with the organization as it uses AI. Bridge's link between learning and manager-led performance reviews speaks to the manager multiplier problem behind the shadow spreadsheets: when an employee learns a skill, the manager is accountable for measuring its use.
Both reflect an industry that is maturing. The market is moving away from treating employees as productivity metrics, counting who clicked through a video fastest, and toward verifiable, secure capability building.
What it all comes back to
Technology is a magnifier of human judgment, not a replacement for it.
Whether the subject is the illusion of AI productivity, the legal reality of compliance training or the stubbornness of shadow spreadsheets, technology amplifies the human systems underneath it. Organizations cannot automate their way out of broken workflows, misaligned management incentives or a basic lack of critical thinking.
If AI amplifies whatever workflow, management habits and culture you have today, what are you about to put a magnifying glass over?
The workflow
Buying a faster car does not clear the traffic jam. A tool dropped into a workflow that was never redesigned speeds up what the workflow already does.
Management habits and incentives
When managers reward immediate output over the learning curve, and do not use the new tool themselves, the old way wins on Friday afternoon.
Critical thinking
A confident answer arrives in two seconds. Someone still has to judge whether it is accurate, biased or invented.