AI Consulting in Practice — Nathaniel Whittemore (NLW), Superintelligent

Where are organizations actually finding value in AI? That question is more useful for planning an implementation than the competing narratives about an AI bubble. Nathaniel Whittemore opens with that tension, criticizing studies he considers dubious, including an unnamed MIT report, before turning to what organizations report from their own work.

His perspective combines daily AI news analysis through The AI Daily Brief with executive interviews through Superintelligent, which he introduces as an AI planning platform. The first gives a broad view of adoption; the second exposes what happens inside organizations. Alongside enterprise surveys, he presents an initial analysis of approximately 2,500 self-reported AI use cases collected from his audience.

Enterprise AI usage is broadening across business functions. That establishes adoption, but not yet organization-wide scale: using a tool somewhere in a company is a different milestone from changing how the company operates.

Coding is a particularly visible inflection in 2025. Its reach extends beyond software engineering teams to other departments that can now communicate through code and build things themselves. The organizational change is therefore larger than faster software development: more people can express requirements in executable form.

Agents present a more mixed picture. The anticipated wholesale automation of enterprise work has not arrived, yet deployment has advanced. In KPMG’s Q3 AI Quarterly Pulse release, reported deployment of at least some agents rose from 11% in Q1 to 42% in Q3 2025. The Q3 survey covered 130 U.S. C-suite and business leaders at organizations with at least $1 billion in annual revenue. Whittemore describes these as production deployments, distinct from pilots; the measure does not establish automation across the whole enterprise.

As agents move into actual work, the implementation problem becomes partly human. Whittemore’s executive conversations suggest that some organizations moved through experimentation faster than expected. Their attention is shifting toward upskilling, enablement and how employees interact with agents. Sandboxes let people try those interactions, and he reports less resistance as employees gain experience.

Moving from an exciting experiment to a scaled organizational capability remains difficult. Whittemore cites McKinsey figures in which only 7% describe themselves as fully at scale with AI and agents, while approximately 62% remain in experimentation or piloting. These describe a broader adoption stage than the presence of a deployed agent.

Progress also varies by company size and function. Larger organizations are generally further along in scaling, despite the expectation that smaller companies should move faster. Within companies, the earlier pattern of similar experimentation rates across departments is giving way to greater divergence, with IT operations pulling ahead.

Leading organizations pursue a broader portfolio of changes. They do more than run isolated experiments: they consider AI adoption systematically, undertake multiple initiatives and connect them to organizational strategy. Their objectives include time savings and productivity, but also revenue growth, new capabilities and new product lines.

Spending intentions suggest continued commitment. The KPMG Q3 2025 AI Quarterly Pulse Survey reports average planned generative-AI investment over the following twelve months rising from $114 million in Q1 to $130 million in Q3. Whittemore also cites an approximately $88 million baseline in the preceding Q4. These are planned investments, not realized expenditure; the direction of change matters more to his argument than the absolute budget.

The Deloitte investment findings similarly report that 91% of surveyed organizations plan to increase AI investment over the twelve months through August 2026. Larger commitments bring greater pressure to show returns, while executives’ expected payback periods are shortening.

Whittemore compares the following distributions from KPMG’s annual CEO surveys. The 2024 global comparison asks about generative-AI payback; the 2025 chart refers to AI investment.

These selected categories show expectations moving toward earlier returns. They do not measure whether those returns have arrived.

Pressure to demonstrate value is increasing faster than confidence in measurement. Two distinct KPMG Pulse findings capture the gap: 78% say investor and board pressure for demonstrable returns will influence their generative-AI strategies over the next six months, and 78% say traditional metrics are insufficient to capture AI’s value. The first is more specific than the talk’s shorthand about ROI becoming a bigger consideration in the coming year. Whittemore hears the same measurement problem from CIOs: methods used for previous technologies and initiatives are falling short with AI.

That gap motivated a simpler research step: ask people what impact they are experiencing. The listener survey opened at the end of October; by the talk, roughly 1,000 organizations had submitted about 3,500 use cases. The presented analysis covers the first approximately 2,500.

The survey organizes impact into eight categories: time savings, increased output, quality improvement, new capabilities, improved decision-making, cost savings, increased revenue and risk reduction. This makes room for benefits that a narrow cost calculation might miss, while giving respondents a simple vocabulary for describing value.

These results come from voluntary submissions by an unusually engaged daily AI podcast audience, not a representative sample of enterprises. Returns are respondents’ perceptions, and high ROI combines the significant and transformational response categories.

Negative ROI means respondents believe they have spent more than they have gained so far. It does not necessarily mean they consider the program a failure.

Expectations are even more optimistic: 67% expect high growth in ROI over the next year. Among teams currently reporting negative ROI, 53% still expect high growth. These forecasts describe confidence about future returns, separately from the current results.

Time savings account for approximately 35% of submitted use cases. Increased output and quality improvement are also prominent, making everyday productivity the main entry point. Reported time savings cluster between one and ten hours, especially around five hours.

Recovering a few hours can matter substantially even when the eventual ambition is to create entirely new capabilities. Whittemore illustrates this with five to ten hours saved per week, describing the annual benefit as seven to ten workweeks. That is an illustration rather than a fixed conversion: the equivalent depends on the working-year and workweek assumptions. The practical point is that recurring productivity gains can be valuable before an organization achieves a more ambitious transformation.

Time savings dominate the aggregate, but priorities differ within it. Organizations with 200–1,000 employees show a greater concentration of increased-output use cases. Whittemore tentatively connects this to companies that have reached some scale but are still pursuing expansion; the survey has not established that explanation. Leadership respondents likewise emphasize increased output and new capabilities more than time savings.

Leaders report more optimism and transformational impact than junior respondents. One possible explanation is the work they choose: a senior leader may be evaluating a project whose scope is inherently more transformational if it succeeds. Among use cases submitted by leadership respondents, 17% are reported to already deliver transformational impact and ROI. Role-related project selection may therefore help explain the difference, rather than seniority itself producing better returns.

The smallest organizations also show a concentration of early transformational benefits. That observation concerns reported outcomes, whereas the earlier enterprise comparison concerned progress toward scale. Even the survey’s 1–50-person category may obscure important differences: a three-person company and a forty-person company can operate very differently. Whittemore proposes examining that range more closely. Coding and software use cases, meanwhile, show above-average ROI and below-average negative ROI in this sample, without a numerical uplift specified.

Risk reduction is the least common primary benefit in the preliminary analysis. The AI ROI Benchmarking Study asks respondents to choose just one primary benefit, so category shares do not capture every benefit a use case may produce. Risk reduction is the primary benefit for 3.4% of use cases, but 25% of that subset reports transformational ROI—the highest transformational share among the categories described.

A possible explanation comes from conversations with practitioners in back-office, compliance and risk functions. Much of their difficulty involves the sheer volume of work, a burden AI can help address. That offers a plausible mechanism for large gains, although the survey does not demonstrate that volume reduction caused the reported returns.

Industry composition matters too. Technology and professional services are heavily represented in the listener sample. Whittemore describes some other industry samples as reasonably sized, without giving counts, and reports higher average impact for healthcare and manufacturing use cases than for the sample overall. Those differences are leads for further study, not a settled ranking of industries.

More content, better content and a few recovered hours characterize the first layer of enterprise adoption. Automation and agents point toward a further layer. In the preliminary analysis, use cases mentioning automation or agents substantially outperform on self-reported ROI. No numerical uplift or causal comparison is supplied, so the finding identifies an association with more advanced use cases rather than proving that adding an agent improves returns.

Breadth of adoption shows a related pattern: people or organizations submitting more use cases tend to report better ROI. Whittemore acknowledges that several explanations are possible, but connects the pattern to the earlier distinction between isolated experiments and systematic adoption. Exploring AI across disciplines and organizational functions creates more opportunities to find valuable applications than stopping after a single trial.

This remains a first look at roughly the first two-thirds of the use cases collected at that point. At the time of the talk, the survey was scheduled to remain open another week, with a full study planned for early December. The next step is to move beyond generic discussions of impact and intuitive impressions: collect more concrete reports, investigate the differences and use that evidence to decide where to explore next.