What the system is
An artificial intelligence internal mobility platform is described as a talent activation tool. It matches employee skills to internal openings, surfaces personalized growth paths from performance data, supports succession planning, and reduces turnover by helping people find their next role without leaving. The framing is developmental and the intent is usually genuine.
The system is also a promotion screening system, and it should be governed as one. It determines which employees are shown which opportunities, which are identified as ready, and which are proposed to hiring managers considering internal candidates. Failure to promote and failure to transfer are established theories of liability under Title VII, the Age Discrimination in Employment Act (ADEA), and the Americans with Disabilities Act (ADA). The Uniform Guidelines on Employee Selection Procedures apply by their terms to selection procedures affecting hiring, promotion, and other employment opportunities, not to external hiring alone. New York City’s Local Law 144 reaches automated employment decision tools used to evaluate employees for promotion as well as to screen candidates, a scope point that organizations deploying internal mobility platforms routinely miss because they think of the law as a hiring statute.
Novara Consulting Group’s position is that internal mobility systems carry higher governance risk than external screening systems, for reasons specific to the population they act on, and that the developmental framing is what allows the risk to go unexamined.
The adverse action nobody records
External hiring generates a record. The candidate applies, the employer receives the application, a decision is made, and a rejection is sent. Both parties know something happened.
Internal mobility works differently. The system determines which openings appear in an employee’s feed. An employee who is not matched to a role does not receive a rejection, because she never applied. She never saw the posting. There is no application, no decision record, no notice, and nothing for her to contest. From her perspective nothing occurred at all, which is precisely why nothing is examined.
The absence of a rejection record does not convert the decision into a non-decision. It converts it into a decision the employer cannot reconstruct.
From the employer’s perspective a selection decision was made. An algorithm determined that this employee was not a candidate for that opportunity, on criteria the employer did not write, using data the employer did not validate for this purpose. The absence of a rejection record does not convert the decision into a non-decision. It converts it into a decision the employer cannot reconstruct.
This matters most where the employee has a protected characteristic or protected activity in her file. Consider an employee who requested an accommodation eighteen months ago, or took protected leave, or raised a complaint. If she is subsequently not surfaced for opportunities, she has a retaliation claim, and retaliation claims are the most common category the EEOC receives. The standard is whether the action would dissuade a reasonable worker from engaging in protected activity, and the denial of promotional opportunity comfortably meets it. The employer’s defense requires showing why she was not matched. If the answer is that a model produced a ranking and the scores were not retained, the employer has no defense to offer, only an absence.
Performance data is the wrong input, and it is the employer’s own product
The platforms surface growth paths from performance data. That input deserves more scrutiny than it receives, because performance data is not an observation of the world. It is a record of what managers wrote down.
Manager ratings carry documented, well-studied variation by rater, by relationship, and by group. Feeding them into a mobility model does not remove that variation. It converts subjective judgments into a numeric input, propagates them across every future opportunity the employee is or is not shown, and attaches to the result the appearance of analytical objectivity. The organization has not replaced managerial bias with data. It has given managerial bias a score and a longer reach.
There is a further problem specific to the employees this article is concerned with. Performance records reflect periods of disability-related leave, reduced schedule, accommodation, phased return, medical absence, or caregiving. A rating covering a period of reduced capacity is lower, and legitimately so as a description of that period. A model trained to identify readiness from performance history will systematically rank down employees whose records contain such periods, and it will do so without any variable labeled disability appearing anywhere in the system. The employer does not need to have inferred disability for the outcome to track it. The performance record already encodes it.
Reassignment is an accommodation, and the system now sits in front of it
This is the point that distinguishes internal mobility from every other application discussed in this series, and it is the one most likely to produce a claim the employer does not see coming.
Under the ADA, reasonable accommodation may include reassignment to a vacant position. Where an employee can no longer perform the essential functions of her current role even with accommodation, reassignment to a vacant position for which she is qualified is an accommodation the employer is required to consider, and the employer’s obligation extends to identifying vacancies rather than merely inviting the employee to compete for them in the ordinary way.
An internal mobility platform controls the visibility of vacancies. Once deployed, it becomes the mechanism through which employees learn what positions exist. An employee in the reassignment process who is not surfaced a suitable vacancy, because the model scored her poorly on the basis of a performance record depressed by the very condition generating the accommodation request, has been failed twice by the same data.
The organization is unlikely to recognize this as an accommodation failure, because the accommodation process and the mobility platform sit in different functions, are owned by different people, and were procured for different reasons. Nobody asked whether the talent marketplace would become the reassignment mechanism. It became one anyway, on the day the vacancy list moved into it.
Succession planning and age
Succession planning is named in the proposition and carries its own exposure. A model that identifies high-potential employees, readiness for advancement, or successors for critical roles is producing a categorization that determines who receives development investment, visibility, and opportunity.
The inputs available to such a model are heavily correlated with age. Tenure, career stage, time in role, recency of credential, pace of prior advancement, and the shape of a career trajectory all track years in the workforce. A system that identifies emerging talent from these signals will produce an age-patterned output whether or not anyone intended it, and the ADEA does not require intent. If the resulting categorization drives differential investment, the employer has made a series of employment decisions on a basis it cannot explain, recorded as a talent strategy.
The governance question is the same one that applies to any individual-level score: what action follows from the categorization, and would that action survive review as an employment decision rather than as an analytics output. If nothing follows, the model produces no value. If something follows, it is reviewable.
Two claims to strike
The proposition attaches two assertions that should be removed from external communication.
The first is that the initiative aligns with diversity goals. Title VII, as amended in 1991, makes it unlawful in connection with the selection or referral of candidates for employment or promotion to adjust scores, use different cutoff scores, or otherwise alter the results of employment-related tests on the basis of race, color, religion, sex, or national origin. The statutory language names promotion expressly. Nothing suggests that mobility platforms do this, and the point is not that the organization has done something wrong. The point is that a public statement connecting an automated promotion tool to demographic objectives is a document the organization wrote, cannot retract, and may be asked to explain. The accurate and safer formulation is that the organization is broadening the range of employees considered for opportunities, has documented the job-related basis for the criteria, and monitors outcomes.
The second is that the system minimizes bias. There is no public evidence base supporting that claim for internal mobility tools, and the input analysis above suggests the mechanism runs the other way. An organization that says this has adopted a vendor’s assertion as its own representation about its promotion practices, in writing, without substantiation.
What to establish before deployment
Six items, answerable from evidence rather than vendor description, before an internal mobility system governs who sees which opportunity.
- Determine whether the platform will become the vacancy identification mechanism for ADA reassignment. If it will, establish a parallel route that does not depend on the model’s ranking, and name the person responsible for operating it.
- Retain, for each employee and each opportunity, the score, the inputs, and the model version. A decision not to surface an opening is a selection decision, and it is the one the organization will be least able to reconstruct because no rejection was generated.
- Identify which performance data feeds the model and over what period, and determine how the model treats records containing protected leave, reduced schedule, or accommodation periods. If nobody has asked, the answer is that it treats them as low performance.
- State, for each scored construct, which documented requirement of the target role it corresponds to. The Uniform Guidelines apply to promotion, and the component analysis they contemplate requires this.
- Establish whether employees are told they are being scored, what they can see about their own profile, and whether they can correct it. An employee who cannot see why she is not being matched cannot raise it, which suppresses the internal complaints that would otherwise surface the problem before a charge does.
- Confirm scope under local law. Where the organization has employees in New York City, an automated tool evaluating employees for promotion falls within Local Law 144 regardless of whether the organization thinks of it as a hiring system.
Conclusion
The developmental case for internal mobility is sound. Employees do want growth and transparency, internal movement does reduce turnover, and organizations that surface opportunities systematically do better than organizations relying on whoever a manager happens to think of.
What makes these systems risky is that they operate on employees rather than applicants, and employees arrive with performance histories, protected activity, accommodation records, and tenure. Every one of those is an input the model will use or a claim the model can generate. The system also produces its adverse actions silently, which means the organization will not learn about a pattern from complaints, because the affected employee has nothing to complain about. She simply keeps not being shown things.
The organizations that will govern this well are the ones that treat the mobility platform as a promotion process, document it as one, retain the record as one, and check whether it has quietly become the route through which a legally required accommodation is supposed to travel.
References
- Americans with Disabilities Act of 1990, 42 U.S.C. § 12111(9)(B), reassignment to a vacant position as reasonable accommodation; 42 U.S.C. § 12203, prohibition on retaliation.
- Age Discrimination in Employment Act, 29 U.S.C. § 621 et seq.
- Burlington Northern & Santa Fe Railway Co. v. White, 548 U.S. 53 (2006), standard for materially adverse action in retaliation claims.
- New York City Local Law 144 of 2021, Automated Employment Decision Tools, covering tools used to evaluate employees for promotion.
- Title VII of the Civil Rights Act of 1964, as amended, 42 U.S.C. § 2000e-2, including § 2000e-2(l).
- US Airways, Inc. v. Barnett, 535 U.S. 391 (2002), reassignment and seniority systems.
- Uniform Guidelines on Employee Selection Procedures, 29 C.F.R. Part 1607, §§ 1607.4 and 1607.15; EEOC recordkeeping requirements, 29 C.F.R. Part 1602.