McKinsey’s Judgment Gap: What the New Research Found
New data published this week from McKinsey & Company, Deloitte, and IBM is sharpening a debate that every executive team is now having out loud: how much of corporate strategy should move to artificial intelligence, and how much has to stay with people. The reports, released as companies finalize 2027 planning cycles, draw a clearer line than most prior guidance between tasks AI can safely speed up and decisions that still require a human to own the outcome.
The timing matters. Enterprise software vendors spent the first three weeks of October 2026 shipping a wave of specialized “decision models,” AI systems built to score, classify, and recommend rather than write prose. That product push has collided with fresh governance data showing AI-related incidents are already reaching boardrooms at a rate that is forcing policy rewrites. Put together, the picture is less about AI replacing strategists and more about where the handoff between machine and executive actually belongs.
Don't miss new tech stories on Google
Add TrendinTech once in the Google app and our stories appear in your news suggestions.
McKinsey’s Judgment Gap: What the New Research Found
McKinsey & Company’s latest strategy research states that AI can accelerate the mechanical parts of strategy work: research, interpretation, synthesis, simulation, and communication. What it cannot do, according to the firm, is replace the judgment and alignment-building that turns an analysis into a decision an organization will actually execute. That framing is a departure from earlier, more sweeping claims about AI “doing strategy,” and it reflects a year of enterprises learning the difference between a fast answer and a good one.
McKinsey’s published analysis breaks the problem into two buckets. Routine decisions with clear parameters, the firm argues, are candidates for automation: dynamic pricing and promotions, demand forecasting and inventory routing, and real-time fraud or credit decisions all fit a pattern where the inputs are well understood and the cost of a wrong call is bounded and recoverable. These are the same categories that have quietly run on machine learning models inside retail and financial services firms for years, so the news here is less about a new capability and more about formal recognition that this is where AI belongs by default.
The second bucket is where the firm wants executives to slow down. Decisions involving greater uncertainty or structural change, including entering a new market, repricing during a disruption, or scenario planning when historical patterns may no longer hold, should stay human-led. The logic is straightforward: AI models are trained on historical data, and the decisions that matter most to a company’s trajectory are often exactly the ones where history is the least reliable guide.
Where AI Actually Speeds Up Strategy Work Today
Strip away the hype and the near-term case for AI in business strategy is fairly narrow but genuinely useful. Research synthesis, the unglamorous work of reading hundreds of pages of market data, competitor filings, and analyst notes, is the task every strategy team wishes it had more time for and never does. Large language models compress that work from days to hours. Scenario simulation, running a plan against dozens of variable combinations, benefits from the same speed-up, letting a strategy team see more branches of a decision tree before committing.
This is also the exact niche that a new category of enterprise AI products has targeted over the past three weeks. Rather than general-purpose chat assistants, vendors have begun shipping models built specifically to return a typed answer, a probability, or a score instead of a paragraph, aimed at the routing, classification, and approval workflows McKinsey flags as automatable. Microsoft’s Decision-1 model, built to evaluate predefined choices and assign probability scores rather than write narrative answers, and OpenAI’s Decisions API, which opened to all developers this month, both fit the pattern directly. Neither product claims to replace a strategist. Both are pitched as faster plumbing for the thousands of small, repeatable calls a large company makes every day.
The Decisions Executives Should Not Hand to AI
McKinsey’s list of decisions that should remain human-led reads like a checklist of the moments that define a company’s next five years rather than its next quarter. Entering a new market involves judgment calls about regulatory risk, local competitive dynamics, and brand fit that rarely show up cleanly in training data. Repricing during a disruption, the kind of call airlines and retailers had to make in rapid succession during recent supply shocks, requires reading signals that are changing faster than any model can be retrained. Scenario planning under genuine uncertainty, where the whole point is that historical patterns may not hold, is arguably the worst possible use case for a system that learns from the past.
What ties these together, according to McKinsey, is not complexity for its own sake. A simple decision made under high uncertainty with irreversible consequences still belongs with a human. A complex decision made under stable, well-understood conditions can often be automated. That reframes the AI-versus-human debate away from “how hard is this decision” and toward “how much does this decision depend on information the model has never seen before.”
Deloitte: AI Can Clarify Uncertainty, Not Replace Purpose
Deloitte’s parallel analysis lands on a similar conclusion through a different lens. The firm states that AI can accelerate analysis and clarify uncertainty, helping leaders see a decision’s shape more clearly and faster, but it cannot replace human purpose, values, and judgment. That distinction between clarifying a decision and making it is becoming the standard dividing line across the consulting industry’s 2026 guidance, and it is notable that two of the largest strategy advisory firms in the world have converged on nearly identical language within the same reporting cycle.
The practical implication for software teams building internal strategy tools is that the interface matters as much as the model. A tool that quietly outputs a recommendation invites a different kind of trust than one that shows its reasoning, flags its uncertainty, and leaves the final call visibly with a named decision-maker. Enterprises that have gotten AI adoption right over the past year, per Deloitte and McKinsey’s parallel guidance, tend to be the ones that built that accountability structure into the workflow from day one rather than bolting it on after an incident.
IBM’s Boardroom Data: Governance Incidents Are Already Here
The governance side of this story has numbers behind it. IBM reported that 80% of surveyed executives in EMEA said an AI-related incident reached their board or C-suite during the prior year, and 40% said their company changed AI-governance policies or frameworks in direct response. That is a strikingly high figure for a technology still in the early stages of enterprise deployment, and it suggests the gap between how fast companies are rolling out AI and how fast their governance structures are catching up is wider than most public commentary has acknowledged.
IBM’s published announcements frame this less as a reason to slow AI adoption and more as evidence that governance has to be treated as a first-class part of the rollout, not an afterthought. The 40% figure on policy changes is arguably the more important number here: it shows companies are actually acting on incidents rather than absorbing them quietly, which is a reasonable sign that the feedback loop between AI failures and AI governance is functioning, even if it is running hotter than anyone would prefer.
A Different IBM Data Point: Hiring Got Faster, Not Just Cheaper
IBM also published a case study on the operational side of AI adoption that is easy to miss next to the governance headlines. Hiring managers cut administrative effort by 90% using AI and automation tools, and internal candidates started new roles an average of 12 days sooner as a result. That pairing, a huge efficiency gain alongside a modest but real speed improvement for the people actually moving into new jobs, is a useful reminder that the benefits of AI in business operations are not purely about headcount reduction. In this case, the technology removed friction from a process without removing the human decisions, who to hire, who to promote, embedded inside it.
Deutsche Telekom’s 2.5 Billion Euro Bet
On the financial side, Deutsche Telekom has set a target of 2.5 billion euros in savings from AI, one of the larger publicly stated AI savings goals from a European telecom operator. That figure places Deutsche Telekom alongside a small group of large incumbents treating AI less as an experimental budget line and more as a core line item in multi-year cost planning. The target also puts pressure on the company’s internal strategy and finance teams to show, decision by decision, which of those savings come from automatable processes McKinsey would classify as low-risk and which depend on judgment calls that still need a human signature.
The New Wave of Decision Models Reshaping Enterprise Software
The business-strategy debate did not happen in a vacuum this month. October 2026 has been one of the busiest stretches on record for enterprise AI releases, and a striking number of them are purpose-built for exactly the kind of bounded, parameter-driven decisions McKinsey says are safe to automate. Beyond Microsoft’s Decision-1 and OpenAI’s Decisions API, Cloudflare shipped its open-weight Clef and Clef-flash models, and Amazon released a compact Strands Decider model, all variations on the same theme: smaller, cheaper, purpose-built systems that return a score or a classification instead of an essay.
General-purpose model releases have kept pace. Gemini 4 Argon reached general availability this month, Claude Haiku 5.5 shipped as Anthropic’s fastest small model, and Liquid AI’s LFM2.5-24B-A2B pushed on-device inference speeds further than most open-weight models have managed before. None of these are being marketed primarily as strategy tools, but all of them are the substrate that strategy-adjacent products, from research synthesis assistants to the decision-scoring models above, are being built on top of.
That volume of releases is itself part of the story McKinsey and Deloitte are responding to. When a strategy team can choose between a dozen credible model options in a single month, the hard part of AI adoption stops being “which model” and starts being “which decisions do we actually let any of them touch.”
Satya Nadella on Why Humans Still Run the Show
Microsoft chairman and chief executive Satya Nadella has been one of the more consistent voices arguing that AI’s role in business should be framed as assistance rather than replacement. “I think that one big design decision we made was to think about this as a copilot. Not as an autopilot, but as a copilot. Designing in such a way that the human is in control. Human agency, human judgment is what is still the core and then building the product around it,” Nadella said, describing the design philosophy behind Microsoft’s AI product line.
Nadella has also pushed a specific mental model for how output from these systems should be treated inside a company. “You are the editor. That metaphor of thinking of anything that gets generated as a first draft, whether it’s a piece of code or a piece of text or an image, that’s probably a good way for us to think about how things get accelerated. But at the same time, humans have to do their job, too,” he said in the same conversation. The framing lines up closely with what McKinsey and Deloitte are now telling corporate clients: treat AI output as a draft, not a decision.
On the broader question of corporate AI strategy, Nadella has argued that companies need to build two kinds of capital in parallel. “Every company is going to have to build what I think of as human capital and token capital. Human capital comprises the knowledge, judgment, relationships, ingenuity, and pattern recognition of its people, while token capital is the firm’s AI capability it builds and owns,” he wrote. He went further on what actually drives value from that AI capability: “I believe human agency will be the driver of token capital growth. Humans will set ambitious goals, connect dots across domains, build relationships, and recognize patterns that matter most. Without human direction, you have compute running in circles.”
Where AI Helps Versus Where Humans Must Lead
The clearest way to summarize the current consensus from McKinsey, Deloitte, and IBM is to sort specific business decisions by who should own them. The table below maps the categories each firm has publicly described.
| Decision Type | Example | Who Should Lead | Why |
|---|---|---|---|
| Dynamic pricing and promotions | Retail markdowns, surge pricing | AI, human oversight | Clear parameters, historical data is reliable, errors are correctable |
| Demand forecasting and inventory routing | Warehouse allocation, restocking | AI, human oversight | Repeatable pattern, high data volume, low per-decision stakes |
| Real-time fraud and credit decisions | Transaction approval, credit scoring | AI, human oversight | Bounded risk per decision, established statistical baselines |
| New market entry | Launching in an unfamiliar region | Human-led | Regulatory, cultural, and competitive factors with little training data |
| Repricing during a disruption | Supply shock, sudden demand shift | Human-led | Conditions change faster than models can retrain |
| Scenario planning under uncertainty | Long-range strategic bets | Human-led | Historical patterns may not hold by design |
Competitive Field: Consultancies, IBM, and the Hyperscalers
Four distinct types of organizations are now shaping how companies think about AI in strategy, and each is pulling in a slightly different direction. Strategy consultancies like McKinsey and Deloitte are positioning themselves as the arbiters of where the human-AI line should sit, publishing frameworks rather than products. IBM is playing a dual role, publishing governance research while also selling the AI infrastructure and services that create the governance problem in the first place. Hyperscalers and model vendors, Microsoft, OpenAI, Google, Anthropic, Cloudflare, and Amazon among them, are shipping the actual decision-making tools at a pace that is arguably outrunning the governance guidance meant to contain them. Telecom and industrial operators like Deutsche Telekom sit downstream of all three, trying to hit cost targets using tools whose risk profile is still being written in real time.
| Organization | Role in the AI Strategy Debate | Key Reported Figure |
|---|---|---|
| McKinsey & Company | Published the judgment-gap framework separating automatable from human-led decisions | Three decision categories flagged as human-led: market entry, disruption repricing, scenario planning |
| Deloitte | Published parallel guidance on AI’s role in analysis versus judgment | Positions AI as accelerant for analysis and uncertainty, not a replacement for purpose and values |
| IBM | Published governance incident data and an HR efficiency case study | 80% of EMEA executives report an AI incident reached the board; 40% changed governance policy |
| Deutsche Telekom | Set a public AI cost-savings target | 2.5 billion euro savings target from AI |
| Microsoft, OpenAI, Cloudflare, Amazon | Shipped specialized decision-scoring models in October 2026 | At least four purpose-built decision models launched within a three-week window |
Historical Context: From Decision Support Systems to Agentic AI
The tension between automated analysis and human judgment in corporate strategy is not new. Decision support systems built on statistical models were already common in large retailers and banks by the 1990s, handling exactly the kind of bounded, repeatable calculations McKinsey now flags as automatable. What changed over the past three years is scope. Earlier systems were narrow by design, built for a single function like credit scoring or inventory replenishment. The generation of large language models and, more recently, the decision-scoring models released this October are general enough that a company can plausibly point them at almost any decision in the business, which is precisely why the question of where to draw the line has become urgent enough for McKinsey, Deloitte, and IBM to all publish guidance on it within the same month.
The 2023 to 2025 period saw a wave of enthusiasm about AI “co-pilots” for executives, largely focused on summarization and drafting. The current wave, built around typed outputs and probability scores rather than free text, represents a narrower but more operationally serious bet: instead of asking AI to help write the strategy memo, companies are now asking it to directly approve a transaction, route a support ticket, or flag a pricing change. That is a meaningfully higher-stakes use case, and it explains why governance data like IBM’s 80% figure is suddenly a headline number rather than a footnote.
Market Impact: Software Vendors and the Governance Industry
The practical effect of this guidance is already visible in how enterprise software is being sold. Vendors pitching AI tools to corporate strategy and operations teams are increasingly leading with the narrow, bounded use cases McKinsey has blessed, pricing and inventory optimization, fraud screening, routing, rather than broader strategic planning tools. That is a shift from the 2024-era pitch of AI as a general strategic advisor toward a more modest, defensible claim: AI as a fast, auditable layer underneath decisions a human still signs off on.
A parallel market is forming around governance itself. IBM’s data point, that 40% of EMEA companies changed AI governance policy after a board-level incident, is effectively evidence of a new compliance and tooling market taking shape in real time. Expect to see more enterprise software vendors bundling audit trails, human-in-the-loop checkpoints, and decision logs directly into their AI products over the next two quarters, not as a nice-to-have but as a response to boards that have already been burned once.
What Could Go Wrong: Accountability Gaps
The risk that McKinsey, Deloitte, and IBM are all circling, from different angles, is what happens when a company blurs the line between the two decision categories. A pricing algorithm that quietly starts making calls closer to a market-entry decision, for instance adjusting strategy in a new geography based on patterns learned elsewhere, can drift into exactly the high-uncertainty territory McKinsey says should stay human-led, without anyone explicitly deciding to let it. IBM’s governance numbers suggest this kind of drift is already common enough to be reaching boards at a meaningful rate rather than remaining a rare, headline-grabbing failure.
The fix both consultancies are recommending is less technical than organizational: name a human owner for every AI-assisted decision before it is deployed, not after an incident forces the question. That sounds obvious, but IBM’s 40% figure on post-incident policy changes implies a lot of companies are currently doing it in reverse.
5 Predictions for AI in Business Strategy Through 2027
- Purpose-built decision models, in the mold of Microsoft Decision-1 and OpenAI’s Decisions API, will multiply faster than general-purpose strategy assistants, since they map directly onto the bounded use cases McKinsey has already approved.
- More large enterprises will publish explicit AI decision-rights frameworks, naming which categories of calls can be automated and which require a named human sign-off, following the IBM governance pattern.
- Governance incidents reaching the board will keep rising in absolute numbers through 2027 simply because AI deployment is scaling faster than oversight structures, even as the rate per deployment likely falls.
- Consultancies will increasingly sell governance and decision-rights frameworks as a standalone service line, separate from traditional strategy engagements, as demand from boards burned by incidents grows.
- Cost-saving targets like Deutsche Telekom’s will become more common among large incumbents, but companies will face growing pressure to disclose which savings came from McKinsey’s automatable category versus judgment calls that still involved a human.
What This Means for Technology and Strategy Teams
For engineering and product teams building internal tools, the practical takeaway from this month’s reports is narrower than the headlines suggest. The safest near-term AI investments are the ones McKinsey already named: pricing, forecasting, routing, and fraud detection, where data is abundant and mistakes are recoverable. Strategy teams evaluating new AI tools should ask vendors directly which of McKinsey’s two buckets their product is designed for, since a tool marketed vaguely as a “strategy assistant” without that distinction is more likely to drift into the high-uncertainty territory both McKinsey and Deloitte say should stay with people.
The governance side deserves equal attention. IBM’s data makes clear that the question is no longer whether an AI-related incident will reach a company’s board, for 80% of surveyed EMEA executives it already has, but whether a policy framework exists before that happens rather than after. Teams that build decision logs, human checkpoints, and clear ownership into AI-assisted workflows now are building the exact infrastructure that 40% of companies only added after a visible failure.
Frequently Asked Questions
What is the main finding from McKinsey’s new AI strategy research?
McKinsey & Company states that AI can accelerate strategy research, interpretation, synthesis, simulation, and communication, but human judgment and alignment-building remain necessary for sound decisions, particularly on high-uncertainty calls.
Which business decisions does McKinsey say are safe to automate?
McKinsey identifies routine decisions with clear parameters as candidates for automation, including dynamic pricing and promotions, demand forecasting and inventory routing, and real-time fraud and credit decisions.
Which decisions should stay human-led according to McKinsey?
Decisions involving greater uncertainty or structural change should remain human-led, including entering a new market, repricing during a disruption, and scenario planning when historical patterns may be unreliable.
What did IBM find about AI governance incidents?
IBM reported that 80% of surveyed executives in EMEA said an AI-related incident reached their board or C-suite during the prior year, and 40% said their company changed AI-governance policies or frameworks as a result.
How much is Deutsche Telekom aiming to save from AI?
Deutsche Telekom has a stated target of 2.5 billion euros in savings from AI, one of the larger publicly disclosed AI cost targets among European telecom operators.
What does Deloitte say about AI replacing human judgment?
Deloitte states that AI can accelerate analysis and clarify uncertainty but cannot replace human purpose, values, and judgment in making final decisions.
What are AI decision models, and how are they different from chatbots?
Decision models, such as Microsoft’s Decision-1 and OpenAI’s Decisions API, return typed answers like probabilities, classifications, or scores instead of open-ended text, targeting routing, approval, and verification workflows rather than conversational tasks.
Does IBM’s hiring data mean AI is replacing recruiters?
No. IBM reported that AI and automation reduced hiring managers’ administrative effort by 90% and helped internal candidates start new roles an average of 12 days sooner, but the hiring decisions themselves remained with people.
Related Coverage
Daniel Okafor
Daniel Okafor is the Senior AI Reporter at TrendinTech, where he covers large language models, machine learning research and the practical use of artificial intelligence across business and government. He has a particular interest in evaluation: how models are benchmarked, where those benchmarks fail and what that means for the companies betting on them.
All stories by Daniel Okafor (381)
![OpenCode Console Breach Exposes OpenAI, Anthropic Keys [2026]](https://trendintech.com/wp-content/uploads/2026/10/opencode-console-breach-openai-anthropic-google-keys-2026-photo-3-640x360.webp)
![Meta Blames Claude AI Tool for No Kings IG Ban [2026]](https://trendintech.com/wp-content/uploads/2026/10/meta-claude-ai-no-kings-instagram-suspension-2026-photo-3-640x360.webp)