The AI skill is starting to appear in pay and promotion decisions
On October 14, 2026, Reuters reported that U.S. financial-services firms are beginning to reward AI proficiency in employee pay and advancement. In Selby Jennings' AI Talent Pulse Report 2026, 25% of surveyed finance HR professionals said their organization had linked raises, bonuses or promotions to AI proficiency. The figure represents responses about current workplace practices, not an independently audited count of every Wall Street firm.
The underlying research was conducted by third-party research provider Censuswide on August 19–26 among 500 U.S. financial-services HR professionals aged 22–64. These are respondent-level findings, not a register of individual employees' AI bonus payments or the total cost of such programs. The survey's findings do not identify one common compensation formula.
The shift nevertheless matters. Employers are moving beyond asking whether applicants can operate AI tools toward attaching economic value to employees' abilities after they are hired. But deciding a skill deserves a premium and proving how much business value it produced are fundamentally different decisions.
Forty-two percent have assessment guidelines, not a universal pricing formula
Another 42% of respondents said their organizations had created guidelines for assessing employees' AI skills. Thirty percent reported dedicated teams or roles focused on proving the return on investment from AI, while 30% said workers must disclose AI-tool use in formal performance reviews. These are separate questions, not stages of a single funnel for the same firms.
Selby Jennings executive Dylan Gomez told Reuters that there was no established formula for translating AI usage and performance into compensation and anticipated that policies could become more defined by next year. A skills checklist may help with appraisal, but it does not automatically settle how to allocate a raise or incentive between workers.
A person who produces the same report more quickly, one who improves analytical accuracy and one who prevents a damaging financial error may all use AI well. The actual value depends on what the role is meant to accomplish, how much risk it carries, and whether any claimed improvements can be independently verified.
Yet 31% cannot confidently evaluate the candidates they want to hire
The survey's sharpest contradiction appears in recruiting. Thirty-one percent chose the absence of anyone internally who could confidently evaluate candidates as the biggest obstacle to hiring AI-enabled talent. Another 16% said they did not know precisely what 'AI-enabled' meant for the positions being recruited, while 23% could not find anyone internally who had performed the work themselves.
These responses must not be treated as proof that the exact 25% of organizations linking AI to compensation are the same ones lacking evaluators. The two numbers answer different survey questions. They do indicate that buying into AI skills as valuable and developing the capability to define and assess them are progressing at different speeds across the surveyed finance market.
An engineer building models, an analyst using one for credit review and a risk officer challenging an AI-produced recommendation are not interchangeable profiles. If a company cannot define what the job requires, its hiring tests and subsequent performance-review incentives may reward a label instead of the actual capability.
More AI usage is not necessarily better employee performance
Thirty percent of respondents reported requiring AI-use disclosure as part of formal reviews. Such reporting may help organizations understand adoption, approved tool use, security practices and training gaps. But counting usage is not the same as measuring the value of what the employee produced.
Imagine two fictional employees. One makes dozens of AI calls to draft internal reports. The other uses AI less often but identifies a flawed calculation before it influences a consequential decision. A usage-based measure would favor the first worker; a business outcome or risk-control measure might favor the second. This comparison is illustrative, not a real case from the survey.
Rewarding simple tool activity risks incentivizing performative usage that does not improve decisions. A more defensible review asks what business problem the tools addressed, what changed as a result, and how the employee tested the output before another person relied on it.
In finance, catching a wrong answer can be worth more than speed
Banks use AI in contexts that can include credit assessment, fraud detection, compliance checks and investment research. Cutting processing time and preventing a risky false conclusion are both potentially valuable outcomes, but their economics and accountability are different. A fast workflow with undetected errors can cost more than it saves.
In a September 2026 speech, a Bank for International Settlements Financial Stability Institute leader discussed AI's challenges for model explainability, independent validation and supervisory judgment. U.S. banking regulators separately issued revised model-risk guidance in April, but that 2026 guidance explicitly excludes generative and agentic AI from its defined scope and does not impose new prescriptive rules on all AI models.
A compensation framework should therefore not collapse throughput, analytical reliability and risk prevention into a single AI usage score. Each role needs a standard that captures what matters in that specific financial workflow. That is a Banseog assessment proposal, not a bonus schedule adopted by the surveyed companies.
Who deserves the reward when AI outcomes are created by a whole team?
Commercial AI systems commonly involve several contributors. Developers build the tools, data teams maintain inputs, employees use the outputs, and control functions test and challenge what the system recommends. Measuring one team's productivity gain does not establish how much each contributor personally added.
Suppose a fictional investment analysis team improves turnaround time by 30%. The figure is an illustration rather than a reported banking result. Product design, data readiness, redesigned workflow and an analyst's decisions could each account for part of the gain. An incentive paid to one visible user could overlook those who made the performance improvement possible.
Banseog separates three judgments: capability—what the person can do; verifiable outcome—what materially improved; attributable contribution—how much of that result the individual was responsible for. Each link requires evidence before it supports an individual compensation decision.
Technical screens can expand without solving the hiring judgment problem
Fifty percent of surveyed HR professionals said the share of underqualified AI candidates in their interviews had increased over the past year. This does not mean half of all applicants are unqualified. It reflects respondents' perceived change in the candidates reaching their own interviews.
Separately, 42% said their organizations had added technical assessments to help screen out underqualified candidates. That is a different question from the 42% who reported employee AI-assessment guidelines. Even a technically valid demonstration may fail to show how an applicant would make an accountable decision in the finance role itself.
The candidate evaluator needs both technical understanding and context about the financial decisions to be made. A skills test without a clear work definition may reward the easiest behavior to observe. The 31% evaluator-gap finding is therefore also a question about the employer's internal knowledge, not merely about recruiting process speed.
BANSEOG VIEW | Pricing AI talent requires evidence of actual value
The same finance HR survey reports that 25% link AI proficiency to incentives and advancement, while 31% select a shortage of internal AI candidate evaluators as their leading recruitment obstacle. We do not know whether these findings overlap within individual companies. Together, however, they identify distinct problems: making AI valuable in pay decisions and developing the expertise to know what its value actually is.
Employers can start with four separate questions. What does proficiency mean in this exact role? Which verified workflow outcome improved? Who identified or prevented consequential errors? And what contribution can reasonably be attributed to the employee rather than the wider system or team? These are proposed management questions, not established industry compensation rules.
The next advantage may not belong to organizations with the highest volume of AI usage. It may belong to those that can define valuable skills, check business results, identify accountable contributions and compensate talent accordingly. AI rewards can strengthen strategy only when companies know what they are rewarding.
BANSEOG VIEW
Banseog View — AI activity, verifiable outcome and individual value
The 25% rewards, 42% new employee assessment guidelines, 30% AI-use disclosure and 31% internal candidate evaluator shortage are distinct questions answered by 500 U.S. finance HR professionals. They do not establish that all four conditions co-occur at the same firms.
AI usage can be measured, but verifiable operational improvement, avoidance of errors and attribution to an individual require different evidence. Banseog proposes capability → verifiable outcome → attributable contribution as an analytical distinction, not an existing standardized bonus rule.
The BIS September 2026 speech highlights governance and human judgment; the OCC's April 2026 revised guidance explicitly excludes generative and agentic AI from its scope. Neither source establishes a newly binding blanket regulatory bonus or AI-performance evaluation system.
SOURCES
Primary sources and references
- Selby Jennings — AI Talent Pulse Report 2026 (U.S. finance HR survey)
Primary Censuswide survey 500 U.S. financial HR respondents Aug. 19–26, 2026. Separate metrics for pay-linked AI skill (25%), employee guidelines (42%), AI disclosure (30%), ROI roles (30%), evaluators lacking (31%), and recruiting assessment issues.
- Reuters — Wall Street firms tie bonuses, promotions to AI skills (Oct. 14, 2026)
Contemporaneous financial press report and Selby Jennings executive interview saying no established common AI-based pay formula yet.
- Semafor — A quarter of financial firms tie pay and promotions to AI proficiency (Oct. 14, 2026)
Interprets the survey with source interview: the practice appears more frequently at fintech startups and smaller organizations than at large name-brand banks.
- Bank for International Settlements — Supervising banks in an AI-shaped economy (Sept. 18, 2026)
Financial Stability Institute leader's speech addressing explainability, model validation, oversight and strategic judgment; it does not itself establish binding rules.
- Office of the Comptroller of the Currency — Revised Model Risk Management Guidance (Apr. 17, 2026)
Risk-based validation and governance principles for covered traditional models; generative/agentic AI are explicitly outside this particular guidance's defined scope and the document is not prescriptive regulation.
Prepared October 15, 2026. Primary Selby Jennings AI Talent Pulse Report 2026: Censuswide survey of 500 U.S. finance HR professionals aged 22–64 conducted Aug 19–26, 2026. 25% said employers linked raises/promotions/bonuses to AI proficiency; 42% created guidelines for employee AI skill assessment; 30% required AI use disclosure in formal reviews; 30% created AI ROI dedicated departments/roles; 31% named lack of confident internal candidate evaluators the #1 obstacle to recruiting AI-enabled talent. These are separate items and were not cross-tabulated to prove co-occurrence within employers. 50% observed an increase in the share of underqualified interview candidates; not 50% of all candidates. 42% extra technical candidate assessments is a different metric from 42% employee skill guidelines. Reuters October 14 reported no established compensation formula. BIS September 18 speech is supervisory perspective, not enforceable regulation. OCC's April 17, 2026 updated model risk guidance excludes generative/agentic AI and does not establish binding prescriptive rules. Banseog's analysis of activity, business outcomes, risk and personal contribution is editorial and not asserted as an employer-adopted formula or audited compensation finding.