What Human Preference Means in Practice

Every AI transformation begins with a practical question: what can AI do?

That is the right place to start. But it is a risky place to stop.

Technical capability can tell us where AI might have an impact. It cannot tell whether people will trust the result, accept the experience, or still expect a human to be involved. It cannot tell whether automation will feel cold, careless, or simply wrong and inappropriate in the moment. And those moments shape how customers perceive your brand, how the best job candidates judge your company, and how your culture is built in practice.

In our previous article, we explained why we need to ask a second question: what should and what shouldn’t AI do? Building on research in trust, social cognition, and algorithm aversion, we developed a taxonomy of work that explains where human involvement is likely to remain preferred despite advances in AI capability. The framework suggests that human preferences are driven by the social and psychological meanings attached to work, particularly in contexts involving authorship, relationships, accountability, and moral judgment, while AI may be preferred in low-stakes, judgment-free interactions.

The taxonomy contextualizes the nuance introduced by human preference and acknowledges that Human+AI collaboration isn’t a simple yes-or-no question. What is possible does not determine what is preferred. Some work can move to AI with little friction, while other work loses a significant part of its value when the human disappears.

The challenge is that this distinction cannot remain a general concept. To use it in workforce planning, job redesign, and AI investment decisions, human preference has to become measurable.

That is what Korn Ferry is now building: a Human Preference Score that shows where human involvement still matters in the work itself.

Part of a Broader AI Impact Decision Framework

The Human Preference Score is part of Korn Ferry’s broader AI Impact Decision Framework, which starts with the work itself, not job titles. The framework first estimates technical AI capability (“What can AI do?”) at the level of responsibilities, then applies two types of constraints: human preference constraints and regulatory constraints.

We also look at and measure organizational readiness, because even strong AI potential may not turn into value if the company lacks the data, workflow clarity, governance, skills, or change capacity to deliver it.

Finally, it looks at what additional value AI can create beyond higher productivity. When AI is deployed, is the main effect released capacity, better quality, higher output, or a mix of these?

This approach keeps AI impact from becoming merely a headcount conversation, and instead focuses on Human+AI work transformation.

Human Preference Score

To make the AI Impact Scores useful for workforce decisions, we need to look beyond the job title. A job title tells us what employees are called. Responsibilities tell us what people do: prepare an analysis, advise a client, resolve an employee issue, manage a relationship, make a recommendation, or keep a process running. By scoring responsibilities across Korn Ferry success profiles, we can see where human involvement still matters in the work itself.

The Human Preference Score looks at how much the human touch matters for a specific type of work. It asks a simple question: even if AI is technically able and legally allowed to do the work, is AI the right way to deliver it? The score considers whether AI delivery could weaken human-related aspects such as relationship quality, trust, user experience, stakeholder acceptance, brand perception, or cultural expectations.

A low Human Preference Score does not mean the work is unimportant. It means people are less likely to care whether AI supports or performs more of it. A high Human Preference Score means human presence is part of the value. In those cases, AI may still perform a substantial part of the work. But replacing the human role too visibly may damage trust, experience, or acceptance.

This differentiation gives us a more practical way to think about AI adoption. Instead of asking only where AI has high technical potential, we can also ask where AI use is likely to be accepted, where it may create friction, and where human involvement should remain part of the work design.

What We Are Starting to See in the Data

We examined a sample of responsibilities across several functions to see where human preference constraints are weaker, where they become stronger, and what these patterns mean for workforce design. The Human Preference Score shows the share of technical AI potential that is limited by human preference constraints. A lower score means more of the technical potential can likely be realized with less human connection. A higher score means human involvement needs to remain more visible in how the work is designed and delivered.

Human Resources

• Lower Human Preference Constraint: HR Frameworks and Tools (~18%)
• Higher Human Preference Constraint: Diversity, Equity, and Inclusion (~38%)

Sales

• Lower Human Preference Constraint: Bid/RFP/Tender Analysis (~18%)
• Higher Human Preference Constraint: Customer Relationship Development and Prospecting (~29%)

Information Technology

• Lower Human Preference Constraint: Technical Database Support (~18%)
• Higher Human Preference Constraint: Enterprise Architecture (~26%)

Note: These figures are illustrative, unweighted averages aggregated across all responsibility levels.

Human Preference Patterns

The first pattern is clear: lower Human Preference Scores tend to appear in work that is structured, procedural, internal, or easy to check. In the current sample, this includes areas such as administration, recordkeeping, policy localization, technical database support, tender analysis, data collection, and routine documentation. In these areas, people mainly need the work to be accurate, current, and fast. AI can often help without changing the meaning of the interaction.

Higher scores are where the work design question becomes more interesting. Customer service, business analysis, software implementation, data management, marketing, product development, finance planning, and some sales activities often reside here. AI can do useful work, but how it appears in the workflow matters. In customer service, for example, AI may prepare the case history, suggest the next best action, or draft a response, but a frustrated customer who needs reassurance may still expect a person to listen, take ownership, and rebuild trust.

The highest scores appear when work carries more social meaning. In the current sample, this is more common in employee relations, Diversity, Equity, and Inclusion, coaching, leadership, employer branding, employee engagement, key account relationships, sensitive health and well-being work, and high-visibility communications. These are moments where people notice who is speaking, who understands the context, and who will stand behind the answer. AI can still help, but the human role is an essential part of the value.

Function Labels Can Hide the Real Pattern

Human Resources is a good example. Some HR work may have a lower Human Preference Score, such as administration, payroll, onboarding, and candidate pipeline work. But employee relations, employee experience, coaching, assessment, succession, and organizational design score much higher. So, the answer is not “HR should stay human” or “HR can be automated.” The answer depends on which part of HR work we mean.

Sales show a similar split. Tender analysis, pipeline health, negotiation inputs, and standard sales support sit lower. Relationship development, account ownership, renewals, and customer focus sit higher. AI can help sales teams prepare better, respond faster, and manage more information, but the client relationship still needs a visible and accountable human owner.

Information Technology also has scores that differ by task. Technical support and database work tend to sit lower. Architecture, cybersecurity, infrastructure, data protection, and systems risk are higher. In these areas, the issue is not empathy. It is accountability, confidence in the system, and the cost of getting design decisions wrong.

This is why human preference constraints need to be measured at the responsibility level. A function-level view can help us see the broad pattern, but it can also mislead. Two responsibilities in the same function may need very different AI designs.

Why This Changes the AI Conversation

AI can do many things, and its capabilities are growing as we speak. But there is one thing AI can never replace: being a real human.

When work depends on trust, fairness, relationship, accountability, or humanity itself, removing the human can remove part of the value, or even all of it. When work is routine, private, or low stakes, the opposite may be true: too much human involvement can slow things down.

The Human Preference Score makes this measurable and visible. Once we are aware of the trade-off, we must answer one final question: if AI can do the work, but people no longer feel that someone cares, what exactly have we gained?

AI Impact Score Language

Methodology

Korn Ferry’s AI Impact Score is built on the Korn Ferry success profiles: a structured, calibrated library of profiles that describe what success in a particular role looks like when done well. This library spans industries and job functions. Rather than scoring a profile as a whole unit, the AI Impact Score separates each profile into its component responsibilities and assesses these individually for exposure to automation, augmentation, and transformation by AI. Large language models are used to assess impact, using a consistent responsibility taxonomy and calibrated scoring criteria applied across profiles. The process combines multiple model assessments, structured scoring rules, consensus methods, deterministic quality checks, and repeated review and validation by Korn Ferry experts. Because scoring happens at the most granular levels, the findings make it possible to see AI’s impact at whatever level of work a given decision requires.

What Comes Next

Korn Ferry is expanding and calibrating the AI Impact Framework across its library of more than 10,000 success profiles. As coverage grows, the framework will provide organizations with an increasingly comprehensive view of where AI can create value, where human involvement remains essential, and how work can be redesigned accordingly.

How Can I Get It?

Organizations can start with a light-touch AI Readiness Scan to understand their current readiness and biggest barriers. Deep Diagnostics then maps their roles and headcount to Korn Ferry’s success profiles, showing where realizable AI capacity sits and which areas deserve attention first. Work Redesign turns those priorities into specific changes applied to roles, responsibilities, skills, and human-AI workflows.