From Efficiency Theater to P&L Impact

Australian software giant WiseTech Global announced a sweeping restructuring this week, cutting roughly 2,000 jobs; nearly a third of its global workforce; in a direct and explicit pivot toward artificial intelligence. CEO Zubin Appoo made a stark declaration on a conference call, stating that "the era of manually writing code as the core act of engineering is over." The company plans to deeply integrate AI into its software development and customer service operations, claiming the technology can shrink the time needed for complex coding projects from months down to a single day. This move is being watched closely worldwide as one of the most drastic examples yet of a major tech firm actively swapping human headcount for AI efficiencies.

In a significant step toward national workforce readiness, the U.S. Department of Labor (DOL) officially released a voluntary "AI Literacy Framework" on February 20th. Designed for employers, educators, and workforce development boards, the framework aims to establish a standard baseline understanding of artificial intelligence across the labor market. Rather than focusing solely on highly technical engineering capabilities, the DOL's guidance emphasizes practical, everyday AI competencies, safe usage guidelines, and ethical considerations. This signals a concerted federal push to democratize AI skills and ensure the broader American workforce is prepared for the rapid acceleration of digital transformation.

New reporting this week from the Society for Human Resource Management (SHRM) highlights how artificial intelligence is actively accelerating the "Great Flattening" of corporate structures. Across various industries, organizations are continuing to strip away layers of middle management. However, rather than simply overloading the remaining executives, companies are increasingly relying on AI tools to bridge the administrative gap. Managers are now handling a much wider span of direct reports because AI is absorbing the routine performance tracking, scheduling, and project management tasks that traditionally required dedicated middle-management personnel.

There's a tone shift happening in enterprise AI conversations, and if you haven't noticed it yet, you will soon.

For the past eighteen months, the dominant question was "how do we deploy agents?" Organizations scrambled to stand up pilots, select platforms, hire talent, and get autonomous systems into production. Budget approvals came with relative ease because nobody wanted to be left behind. The conversation was about capability and ambition.

That era is ending. The conversation has moved to accountability.

PwC's Global CEO Survey published in January found that 56% of CEOs report zero measurable ROI from AI in the past twelve months. McKinsey's latest analysis puts it even more starkly: while nearly eight in ten companies report using generative AI, just as many report realizing no significant bottom-line impact. Only one percent of companies view their AI strategies as mature. MIT research found that just 5% of integrated AI trials generate sustained value at scale.

These numbers are creating real consequences. The four major hyperscalers lost a combined $1 trillion in market capitalization following their latest quarterly earnings reports, as investors rotated out of tech amid fears of infrastructure overbuild. CFOs are tightening scrutiny. A quarter of planned AI investments are being deferred to 2027 until value can be demonstrated. The patience that funded the experimentation phase is running out.

For organizations building a digital workforce, this pressure isn't a crisis. It's a maturity milestone. Every significant enterprise technology wave hits this inflection point: the moment when "promising" stops being good enough and "proven" becomes the requirement. The organizations that can demonstrate real returns will accelerate their investments. The ones that can't will retreat to pilots that never scale. The difference between those two outcomes comes down to measurement.

The first problem most organizations encounter when trying to measure agentic AI value is that they reach for the wrong metrics. They apply frameworks designed for traditional automation to a technology category that behaves entirely differently.

Traditional automation ROI was relatively simple to calculate. You had a process that required X human hours. You automated it. Now it requires Y human hours (where Y is less than X). The savings equal the difference multiplied by the loaded cost of those hours. Clean, simple, defensible in a board presentation.

Agentic AI doesn't work this way, and organizations that try to measure it this way end up either understating its value or, more commonly, claiming value they can't actually find in the financial statements.

The data illustrates the disconnect. Deloitte found that 66% of organizations report AI-driven productivity gains. But only 20-30% of those organizations convert those gains into measurable financial impact. Task-level speed improvements of 14-55% are common, yet 37-40% of the time saved is lost to fixing low-quality AI output, rework, or non-productive uses of the freed-up time. The hours "saved" evaporate before they ever reach the bottom line.

This is the measurement trap. Organizations report impressive efficiency numbers in pilot reviews and quarterly updates, but when the CFO looks at the P&L, the improvements aren't there. Headcount hasn't changed. Operating costs haven't declined. Revenue hasn't grown proportionally. The productivity gains are real at the task level but invisible at the financial level, and that gap is where executive confidence goes to die.

The trap catches organizations because hours saved is the easiest thing to measure but the least reliable predictor of business value. An agent that saves a knowledge worker two hours per day creates value only if those two hours get redirected to work that generates revenue, reduces costs, or improves outcomes. If the worker simply fills the time with lower-priority tasks, meetings, or administrative overhead, the "savings" exist only on a dashboard.

This disconnect between task-level efficiency and financial impact isn't new. Economists have been studying the productivity paradox since Robert Solow observed in 1987 that computers appeared everywhere except in the productivity statistics. Every major technology wave produces the same pattern: measurable improvements at the individual task level that take years to show up in aggregate economic data.

The reason is consistent. Technology creates efficiency at the point of use, but capturing that efficiency as financial value requires organizational adaptation. The work needs to be restructured. Roles need to change. Capacity freed by automation needs to be redirected deliberately, not left to fill itself. As we explored in "The Automation Trap" (Feb 12), organizations that layer agents onto existing workflows without redesigning the work will see the same pattern: impressive demos, underwhelming results.

For agentic AI, the productivity paradox has an additional dimension. Agents don't just do existing work faster. They make new categories of work possible. A sales team with AI-generated competitive analyses for every deal isn't just doing the old job more efficiently. They're doing a qualitatively different job, one that involves strategic decision-making informed by analysis that was previously impossible at that scale. How do you measure the ROI of capabilities that didn't exist before?

The answer requires a different measurement philosophy. Instead of asking "how much time did we save?" the question needs to become "what business outcomes improved, and can we trace those improvements to our agent deployments?"

The organizations successfully demonstrating agentic AI value have shifted their measurement approach across three dimensions: from activity metrics to outcome metrics, from point-in-time snapshots to longitudinal tracking, and from isolated agent performance to system-level impact.

The shift from activity to outcomes is the most important. Activity metrics tell you what agents are doing: tickets resolved, documents processed, emails drafted, transactions completed. These are useful for operational monitoring but they don't answer the value question. Outcome metrics tell you what changed because agents did those things: customer retention rates, deal close rates, error rates in financial reporting, time-to-revenue for new products, compliance incident frequency.

The connection between activity and outcome is where most measurement programs break down. It requires establishing baselines before agent deployment, defining which business outcomes each agent deployment is expected to influence, and building the instrumentation to track those outcomes over time. This isn't just a data problem. It's an organizational discipline that needs to be established before agents are deployed, not after executives start asking for proof.

Deloitte's research shows that this shift is already underway. Direct financial impact, combining top-line revenue growth and bottom-line profitability, nearly doubled to 21.7% as the primary ROI metric organizations track. Productivity gains, the default justification throughout 2024 and 2025, fell from 23.8% to 18.0% as the leading metric. The market is telling us what it values, and what it values is P&L impact, not efficiency theater.

The longitudinal dimension matters because agentic AI value often materializes on different timescales than traditional automation. A rules-based bot delivers its full value on day one: it executes the programmed task at programmed speed. An agentic system improves over time as it encounters more scenarios, as the organization learns to use it more effectively, and as supporting infrastructure matures. Measuring agentic ROI at a single point in time, particularly early in deployment, almost always understates the trajectory.

The system-level dimension recognizes that agent value often shows up in places that aren't directly connected to where the agent operates. A customer service agent that resolves issues faster may not reduce support costs much if volumes remain constant, but it may significantly improve customer retention, which shows up in the revenue line months later. Measuring only the direct operational impact of the agent misses the cascading benefits that often deliver the majority of the value.

You can't calculate returns if you don't understand your investment, and most organizations significantly underestimate the true cost of their agentic AI deployments.

The obvious costs are model inference, cloud infrastructure, and platform licensing. These show up on invoices and are easy to track. But they're often the minority of total cost. IDC warns that by 2027, organizations will face up to a 30% rise in underestimated AI infrastructure costs, driven not by overspending but by under-forecasting expenses unique to AI-specific projects.

The less obvious costs include data preparation and quality maintenance, which we covered in "The Pillars of Data Quality" (Jan 7). They include integration development and ongoing maintenance, which we explored in "The Quiet Crisis" (Feb 19). They include the organizational change management costs of redesigning work around agents. They include the human oversight costs of monitoring and correcting agent behavior. And they include the opportunity costs of the technical and business talent dedicated to making agents work instead of doing other things.

A discipline called AI FinOps is emerging to address this challenge. Where traditional FinOps manages cloud infrastructure costs, AI FinOps tracks the economics of intelligence: the cost of every inference, every tool call, every retry, every agent action, traced to a business owner and a cost center. Two metrics in particular are gaining traction.

Cost-per-inference measures how much each model invocation costs under real usage conditions, not the theoretical rate-card price. This metric reveals inefficiencies that averages obscure: overly verbose prompts, redundant calls, unnecessarily powerful models being used for simple tasks.

Cost-per-action goes further by measuring the full lifecycle cost of achieving a business outcome through an agent. A single agent action might involve multiple inferences, tool calls, retries, and data retrievals. Cost-per-action captures all of that, giving you a true picture of what the agent's work actually costs and enabling meaningful comparison against what the same outcome would cost through human execution.

Without this cost visibility, ROI calculations are built on incomplete data. You're dividing uncertain benefits by unknown costs and calling the result a return on investment. That's not measurement. That's guesswork.

The practical challenge of connecting agent activity to business outcomes requires building what I call a value chain: a documented, measurable path from what the agent does to why the business cares.

Here's how it works. Start with the business outcome you're trying to influence. Not "deploy a customer service agent" but "reduce customer churn by 15%." Work backward from that outcome to identify the operational levers that drive it. Customer churn might be influenced by first-contact resolution rates, average response times, issue escalation frequency, and customer satisfaction scores.

Then connect your agent deployment to those specific operational levers. Your customer service agent improves first-contact resolution by handling routine issues autonomously and routing complex issues with full context to human agents. It reduces response times by operating 24/7. It decreases escalation frequency by accessing the right data from the right systems in real time (assuming you've addressed the integration challenges we discussed last week).

Now you have a traceable chain: agent activity drives operational metrics, which drive business outcomes, which appear on the financial statements. Each link in the chain is measurable. Each link can be validated. And when the CFO asks "what are we getting for this investment?" you can walk through a story that starts with agent actions and ends with revenue retention, not a slide that says "we saved 10,000 hours."

The companies seeing the highest returns, those achieving 3x to 6x their investment within the first year, share this characteristic. They don't measure agent performance in isolation. They connect it to the financial metrics that matter to the business. McKinsey found that the roughly 6% of companies capturing outsized AI value do so primarily because they redesign workflows end-to-end and build measurement systems that track outcomes, not just activities.

One of the hardest aspects of ROI discipline is accepting that not every agent deployment will generate positive returns. Some won't, and honest measurement means being willing to act on that finding.

Gartner predicts that 40% of agentic AI projects will fail by 2027. That doesn't mean 40% of organizations will fail. It means that organizations deploying agents at scale should expect a meaningful percentage of their deployments to underperform. The question is whether you detect underperformance quickly and respond decisively, or whether you let failing deployments consume resources while you wait for results that aren't coming.

This requires establishing clear decision criteria before deployment, not after. What does success look like at 30, 60, and 90 days? What operational metrics need to move, and by how much, to justify continued investment? What are the kill criteria that trigger a pivot or shutdown? These aren't pleasant conversations, but they're the conversations that separate organizations running disciplined AI programs from organizations running expensive science experiments.

The decision framework should distinguish between three scenarios. The first is clear success: the agent is meeting or exceeding outcome targets, the value chain is traceable, and scaling will multiply the returns. Invest more. The second is promising but stalled: the agent shows task-level improvements but the outcome metrics haven't moved yet. Investigate whether the problem is measurement lag, integration gaps, or process design issues. Set a time-bound window for the outcome metrics to materialize. The third is clear underperformance: the agent isn't producing the expected operational improvements, or the operational improvements aren't translating to business outcomes despite adequate time and infrastructure. Redirect the investment.

The organizations that build this rigor into their agentic AI programs will compound their advantages over time, because every deployment either delivers value or generates learning that makes the next deployment more likely to succeed.

If you're building the measurement program for your digital workforce, here's where to focus.

Start with the financial metrics that your leadership already cares about. Revenue growth, gross margin, operating costs, customer lifetime value, cash conversion cycle. These are the outcomes that matter. Your measurement program needs to demonstrate movement in these metrics that can be attributed, at least in part, to agent deployments.

Layer in the operational metrics that connect agent activity to those financial outcomes. These will vary by function: cycle time, error rates, throughput, first-contact resolution, compliance incident frequency, time-to-close, processing costs per unit. The key is choosing operational metrics that have a demonstrated relationship to the financial outcomes you're targeting.

Track the cost metrics with AI FinOps discipline. Total cost of ownership per agent, cost-per-action, infrastructure utilization, and the ratio of agent cost to human cost for equivalent work. These metrics ensure the return calculation has a reliable denominator.

And monitor the health metrics that predict future performance. Agent accuracy rates, escalation frequency, user adoption and satisfaction, data quality scores across the systems your agents depend on. Declining health metrics are leading indicators of declining ROI, and catching them early is far cheaper than discovering the problem in quarterly results.

The measurement isn't the hard part. The hard part is the organizational commitment to use the measurements honestly, to celebrate what's working, to fix what's underperforming, and to stop what's failing.

The agentic AI industry is entering its accountability era, and that's a good thing. The pressure to prove ROI will do more to accelerate meaningful enterprise adoption than any new model release or platform announcement.

The data tells two stories simultaneously. On one hand, 80% of companies report no measurable earnings impact from AI. Only 22% say agents have proven tangible value. The productivity gains that justified early investments are evaporating before they reach the financial statements. On the other hand, the organizations that get measurement right are seeing 3x to 6x returns. Companies that redesign workflows and build proper measurement infrastructure are pulling away from the pack.

The gap between those two groups isn't luck or technology selection. It's measurement discipline. The winners connect agent activity to business outcomes through traceable value chains. They track true costs with AI FinOps rigor. They make honest decisions about what's working and what isn't. And they commit to measuring outcomes, not activities.

2026 is the year that separates organizations with AI strategies from organizations with AI results. The difference will come down to one question: can you prove what your digital workforce is worth?

---

Building a measurement framework that connects your digital workforce to business outcomes starts with understanding where you stand today. The Complete Agentic AI Readiness Assessment includes detailed evaluation frameworks for assessing your measurement maturity, identifying value chain gaps, and establishing the metrics infrastructure that turns agentic AI investments into demonstrable returns. Get your copy on Amazon or learn more at yourdigitalworkforce.com. For organizations ready to build rigorous measurement programs, our AI Blueprint consulting helps design value frameworks, implement AI FinOps practices, and create executive dashboards that connect agent performance to P&L impact.