The Desire for AI Advice in Retirement Plans: A Latent Class Analysis

Traditional human financial planners remain a primary source of guidance for many households, even as digital alternatives emerge. Data from the Survey of Consumer Finances (2023) reveal roughly one-third of Americans worked with a professional financial advisor. Men have been somewhat more likely than women to use an advisor, and higher-educated individuals are also more inclined to seek professional advice (Ludwig et al. 2023). Trustworthiness is often cited as the most important factor when selecting an advisor (Ludwig et al. 2023);, this explains why human advisors continue to be valued for personalized service and accountability. The rise of “robo-advisors” in the late 2000s provided an automated, low-cost alternative for investment advice (Neal 2022), but these services have not rendered human advisors obsolete. Prior research suggests that human financial advisors offer value beyond algorithms, especially for complex planning and emotional support. In response, advisors are leveraging technology while highlighting the relational aspects of their role (Ludwig & Bennetts 2023). Notably, consumers strongly favor a knowledgeable human advisor over a robo-advisor; one experiment found that while experienced advisors were preferred, a robo-advisor was rated about on par with a novice advisor in terms of expected performance (Brenner & Meyll 2020). This implies that expertise and trust cultivated by traditional advisors remain factors that technology alone struggles to replicate.

The past decade has seen rapid growth in AI-driven personal finance tools, particularly robo-advisors for automated investing. Neal (2022) reported that by 2022, robo-advisors managed approximately $870 billion in assets, though this remained a small fraction of global investments. While consumer adoption of these services initially expanded, recent trends indicate signs of stagnation. Usage among U.S. investors peaked at nearly 28% in 2021 before dropping to about 21% in 2022, with the decline being especially pronounced among wealthier investors, many of whom returned to human advisors amid market volatility. These findings suggest that when market conditions become complex or uncertain, clients often seek the reassurance of human expertise.

Nevertheless, AI-based financial tools continue to attract users with their low fees and convenience. Robo-advisory platforms provide portfolio management at a fraction of traditional advisory costs and lower barriers to entry (e.g. low minimum balances) (Zhang et al. 2021).

Still, several barriers hinder widespread adoption. Many consumers remain hesitant to entrust their finances to an algorithm. Trust is a key issue: clients may be uncomfortable sharing sensitive data with a machine, and fintech startups lack the long-standing reputations of traditional financial firms (Neal 2022). Indeed, the use of AI chatbots and virtual assistants in finance has been slow, partly because customers prefer a human interface for discussing personal finances. Early adopters of robo-advice generally report satisfaction with its ease and efficiency, but a significant segment of the public is taking a “wait and see” approach.

Trust and comfort are crucial in determining whether individuals embrace AI-based financial advice or stick with human advisors. Human advisors benefit from interpersonal trust built through face-to-face relationships, whereas robo-advisors must engender trust through institutional credibility and technology reliability (Senteio & Hughes 2024). If a well-known financial institution offers a robo-advisor, clients may grant it more trust, whereas unfamiliar fintech brands face skepticism (Senteio & Hughes 2024). Overall, lack of trust in robo-advisors has been a major impediment to their adoption among some consumer segments (Zhang et al. 2021). Yet some experienced users come to trust robo-advice services nearly as much as human advisors (Figà-Talamanca et al. 2022).

On the other hand, comfort with technology varies widely. Those with positive attitudes toward AI tend to see robo-advisors as useful tools and are more inclined to try them (Zhang et al. 2021).

In contrast, those who are uncomfortable with AI or “algorithm-averse” are far less likely to follow machine-generated recommendations. In fact, simply knowing that advice has an AI component can trigger skepticism; one experiment showed that when a human advisor revealed they used an AI in formulating advice, it reduced some clients' trust in that advice (Figà-Talamanca et al. 2022). Thus, openness to non-human advice is strongly influenced by a person's trust in technology. When trust and tech-comfort are high, clients may view AI-based advice as a beneficial complement or alternative to human guidance. When trust is low, they will default to the perceived safety of human experts.

A key factor in AI adoption is self-assessed AI knowledge and confidence in AI-driven decision making. Individuals who perceive themselves as more knowledgeable about AI are more likely to experiment with AI-based financial tools and integrate them into their decision-making processes (Akhtar et al. 2024). This aligns with DOI theory, which suggests that early adopters tend to be more informed about new technologies than later adopters (Rogers 2003). In contrast, those with limited AI literacy may hesitate to trust or engage with AI-powered financial planning solutions, viewing them as too complex or opaque (Shin et al. 2022). Furthermore, confidence in AI-generated decisions plays a significant role in shaping attitudes toward robo-advisors and algorithm-driven financial insights. Individuals who feel confident in AI-generated financial recommendations are more likely to rely on them, whereas those with low confidence may view AI as an unreliable or risky alternative to human expertise. Prior research has shown that confidence in AI decision making increases after repeated exposure and positive experiences (Brenner & Meyll 2020), suggesting that personal experience with AI tools may act as a bridge between initial skepticism and eventual adoption. Understanding how AI knowledge and confidence interact with financial behaviour can help financial planners and AI developers tailor strategies that enhance client trust and facilitate AI integration in financial decision making.

Individual differences in demographics and financial background help explain heterogeneity in advice adoption. Income and wealth shape advice preferences: higher-net-worth investors have shown a clear preference for human advice, with many wealthy robo-adopter clients eventually reverting to traditional advisors during market turmoil (Neal 2022; Zhang et al. 2021). Gender differences have been observed as well. Men report using professional financial advice at somewhat higher rates than women, but other studies find no significant gender gap in robo-advisor adoption (Figà-Talamanca et al. 2022).

Age is a decisive factor: younger people are typically earlier adopters of robo-advisors and other fintech, whereas older individuals tend to stick longer with traditional human advice. One study found that ease of use is the primary driver for robo-advisor adoption among Gen Y/Z, whereas older generations (Gen X and beyond) require a clear perceived usefulness before adopting such technology (Figà-Talamanca et al. 2022). These generational patterns align with the notion that younger consumers are more comfortable with technology and willing to experiment, while older consumers are more cautious.

A survey was administered to employees of state and local government organizations in the United States (N = 2,000) during January 2025. The data collection was conducted by Morning Consult using their nationwide online panel with appropriate screening questions to verify employment status. The final analytic sample demonstrated balanced demographic characteristics, as shown in Table 1. Respondents had a mean age of 36.6 years (SD = 11.3), with 76.3% aged between 18 and 44. The sample was predominantly female (62.4%) and more highly educated, with 64.0% holding a bachelor's degree or higher. The racial composition was predominantly white and non-Hispanic (86.5%). Just over half of respondents (51.5%) reported having no dependent children. Household income levels were broadly distributed, with 57.6% of respondents earning $75,000 or more annually. The sample also demonstrated considerable engagement with financial professionals, with 42.3% currently working with one. Among those using financial professionals, tax preparers (33.2%) and financial planners (28.3%) were the most consulted. This sample provides a broad representation of public-sector employees, allowing for meaningful analysis of financial advice preferences and AI adoption patterns across different demographic groups.

This study focuses on public-sector employees, who represent approximately 13.4% of the U.S. workforce across local (6.4%), state (4.5%), and federal (2.5%) government positions (Gallagher 2023). While this represents a subset of the broader workforce, public-sector employees were selected because they typically participate in standardized employer-sponsored retirement plans with similar structures, reducing confounding variables related to plan design differences common in private-sector settings. The sample was stratified across various government sectors (education, technology, public safety, administration, etc.) to ensure representation across different types of public-sector work environments.

Race and ethnicity were measured following standard federal statistical guidelines, with race and ethnicity treated as separate constructs. Respondents selected from mutually exclusive racial categories (white, Black, Asian American, Native American, other), while Hispanic ethnicity was measured independently as a separate question. Respondents could identify as Hispanic in combination with any racial category, consistent with Census Bureau methodology (U.S. Census Bureau 2020). All percentages reported in demographic tables represent proportions within their respective category (race or ethnicity) and do not sum across categories.

To ensure consistent understanding across respondents, the survey provided a standardized definition of AI: ‘For the purposes of this survey, artificial intelligence (AI) refers to computer systems or software that perform tasks typically requiring human intelligence. Examples include tools for analyzing data, automating processes, generating content, or assisting with decision-making. AI can be embedded in applications such as chatbots, predictive analytics, workflow automation, and language translation.’ This definition was presented to all participants before AI-related questions to establish a common frame of reference.

Following Weller et. al.'s (2020) guidance on indicator variable selection for LCA, five key measures were used to define the latent classes. Current engagement with financial professionals was measured through active work with financial advisors (binary). AI-related indicators included: use of AI tools for understanding retirement savings options (binary), comfort level with AI for financial decision making (four-point scale from “very comfortable” to “not at all comfortable”), interest in employer-provided AI financial planning tools (interested/not interested/unsure), and current use of AI tools in the workplace (binary). Additionally, two measures capturing AI-related confidence and knowledge were included. Self-reported AI knowledge (four-point Likert scale from “much more knowledgeable” to “much less knowledgeable”) assessed how knowledgeable respondents felt about AI compared to their co-workers. Confidence in AI-based financial decision-making (four-point Likert scale from “very confident” to “not at all confident”) measured each respondent's trust in AI-generated financial recommendations.

Table 2 presents descriptive statistics for the key indicator variables used in the latent class model. Nearly half of respondents (48.3%) reported using AI tools at work, while 44.5% indicated they had used AI to improve their understanding of retirement savings options. Interest in employer-provided AI tools for financial literacy and retirement planning was mixed, with 57.4% expressing interest, 21.2% uninterested, and 21.4% unsure.

Confidence in making decisions based on AI-generated output varied widely, with 31.7% of respondents reporting they were somewhat confident and 22.1% very confident. Similarly, when assessing AI knowledge compared to that of co-workers, 36.6% of respondents felt their knowledge was about the same, while 31.1% considered themselves slightly more knowledgeable. Comfort with using AI tools for financial decision making was relatively low, with 24.7% of respondents not at all comfortable and 46.5% not too comfortable.

In addition to analysing the classes by demographic variables, class analysis was conducted on four types of AI-generated advice tools desired within an employer's retirement plan: estimating retirement income needs, tracking progress toward retirement goals, investment advice or rebalancing, and tax-efficient withdrawal strategies.

The analysis proceeded in two stages using Stata/SE 18.0. First, LCA was conducted to identify distinct segments based on AI adoption and financial advice preferences. LCA is particularly suited for this research because it identifies unobserved heterogeneity in the population that traditional regression approaches may miss. Unlike regression-based methods that assume homogeneous relationships between predictors and outcomes, LCA assumes the population consists of distinct subgroups with different behavioural patterns. This person-centered approach allows us to empirically discover naturally occurring adoption segments rather than imposing predetermined demographic categories and then examining how these behaviourally defined segments relate to specific retirement planning outcomes.

LCA was selected as the preferred analytical approach because it can identify unobserved subgroups within populations based on patterns of observed indicators (McCutcheon 1987; Weller et al. 2020). Unlike regression-based approaches that assume uniform adoption patterns, this approach uncovers the underlying factors that influence individuals' openness to AI in financial decision making. Specifically, LCA allows us to test H1 by determining whether a multiclass solution provides superior fit compared to a single homogeneous population, and H2 by examining whether the identified classes demonstrate systematic differences in financial professional usage and preferences for specific generative AI-powered retirement planning tools. This empirical strategy enables us to move beyond assuming uniform adoption patterns to empirically discovering how employees naturally segment based on their AI-related attitudes and behaviours.

Model selection criteria included the Bayesian information criterion (BIC; Schwarz 1978), the Akaike information criterion (AIC; Akaike 1987), and entropy indices. Lower information criterion values indicate better model fit, while higher entropy values (>.80) suggest better class discrimination (Weller et al. 2020).

Once the optimal number of classes was determined, the socio-economic characteristics of each class were examined to assess how socio-demographic factors, financial professional use, and demand for AI-generated advice within retirement plans were distributed across the identified latent groups. This second stage directly tests H2 by examining whether the empirically derived classes demonstrate systematic differences in financial professional usage and generative-AI retirement tool preferences, with patterns consistent with Rogers's DOI adopter categories.

The estimated probabilities of class membership are presented in Table 4. These probabilities indicate the proportion of respondents assigned to each latent class based on their financial advice preferences and AI adoption behaviours. The largest class, Class 5, comprises 30.8% of respondents, followed by Class 2 at 27.7%. Class 1 accounts for 18.9% of the sample, while Class 4 represents 13.5%. The smallest group, Class 3, includes 9.1% of respondents.

Figure 1 displays the distribution of class membership across the sample. The five identified classes range from 9.1% (Traditionalists) to 30.8% (AI-Comfortable Minimalists), with the largest segments being AI-Comfortable Minimalists (30.8%) and Employer-Driven AI Users (27.7%).

Figure 1.

Distribution of AI Adoption Segments in Financial Advice.

Table 5 presents the conditional probabilities for each indicator variable across the five latent classes, identifying distinct differences in AI adoption, financial decision-making confidence, and engagement with financial professionals. The classes vary significantly in their likelihood of using AI tools for work and retirement planning, their comfort with AI-based financial decisions, and their reliance on human financial advisors.

Class 1 exhibits the highest AI adoption rates, with 75.4% using AI at work and 95.5% leveraging AI for retirement planning. This group also demonstrates the strongest confidence in AI-generated financial decisions, with over 48% extremely confident. Conversely, Class 3 represents the most AI-sceptical segment, with only 9.2% using AI at work and 3.8% adopting AI for retirement planning, while 89.5% express no confidence in AI-based decisions.

Interest in employer-provided AI financial tools is highest among Classes 1 and 2, with 87.0% and 85.4% expressing interest, respectively, while Class 3 again remains the most resistant, with 52.8% explicitly rejecting AI-based financial tools. Engagement with financial professionals also differs across classes, with Class 1 displaying the strongest reliance on traditional financial advice (72.0%), whereas Class 3 exhibits the lowest engagement with financial professionals (15.4%).

Table 6 presents the proportion of each latent class that has engaged with different financial services and expressed preferences for AI-driven retirement planning tools. The results reveal clear distinctions in financial advice-seeking behaviour and AI integration across classes. Financial service use varies significantly by class. Class 1 and Class 2 show the highest engagement with financial professionals, with over half of Class 1 respondents consulting a financial planner and 45% working with an investment advisor. In contrast, Class 3 exhibits the lowest levels of professional engagement, with only 5.7% consulting a financial planner and 2.3% using an investment advisor.

Figure 2 illustrates the systematic differences in AI retirement tool preferences across the five latent classes. The visualization reveals several important patterns for retirement plan implementation. First, Innovators and Early Adopters demonstrate consistently high interest across all four AI tools, with investment advice generating the strongest preferences (61.4% and 56.1%, respectively). Second, Laggards show uniformly low interest across all tools, with investment advice and tax strategies being particularly unappealing (10.8% and 10.2%). Third, the Late Majority segment demonstrates moderate but meaningful interest in simpler tools like income estimation (46.5%) and goal tracking (44.4%), suggesting potential for targeted implementation strategies. Finally, the Early Majority shows strong interest in income estimation and goal tracking (above 50%) but more moderate interest in complex functions, indicating a preference for straightforward AI applications over sophisticated advisory features.

Figure 2.

AI Retirement Tool Preferences by Latent Class.

The relationship between latent class membership and financial professional usage reveals important patterns that challenge assumptions about AI replacing human advisors. As shown in Figure 3, Innovators and Early Adopters demonstrate the highest engagement with traditional financial services, with over 70% working with financial professionals and approximately 45%–50% using specialized advisors like financial planners (CFPs) and investment advisors. This pattern directly contradicts the expectation that high AI adopters would rely less on human expertise. Conversely, Laggards show the lowest engagement across all professional services, with only 22% using tax preparers and minimal usage of specialized financial advisors. The Late Majority and Early Majority segments show moderate but meaningful engagement with basic services like tax preparation (35% and 27%, respectively) but limited use of comprehensive planning services.

Figure 3.

Financial Professional Usage by Latent Class.

These patterns support H2's prediction of systematic differences in AI tool preferences across latent classes. Class 1 and Class 2 again indicate the strongest demand for AI-powered investment advice and rebalancing (61.4% and 56.1%, respectively). Meanwhile, Class 3 remains the least receptive, with just 10.8% interested in AI-driven investment guidance and 10.2% in tax-efficient withdrawal strategies.

The classes demonstrate systematic differences in both financial professional usage and AI tool preferences, supporting H2. Class 1, AI-Integrated Consumers, represent the earliest adopters of AI in financial planning, exhibiting the highest engagement with AI-driven financial tools and traditional financial professionals, along with strong confidence in AI-based decision making. Class 2, Employer-Driven AI Users, also show a high level of AI engagement but primarily through employer-provided tools rather than independent exploration. Class 5, AI-Comfortable Minimalists, demonstrate moderate familiarity with AI but rely less on financial professionals, suggesting a preference for technology-assisted financial decision making without full reliance on AI tools. Class 4, Sceptical Adopters, use AI infrequently but remain somewhat open to employer-sponsored AI tools, reflecting a more cautious stance toward automation. Class 3, Traditionalists, have the lowest engagement with both AI and financial professionals, favoring conventional, self-managed financial approaches. These five groups map onto the DOI framework, with AI-Integrated Consumers resembling innovators, Employer-Driven AI Users aligning with early adopters, AI-Comfortable Minimalists corresponding to the early majority, Sceptical Adopters fitting the late majority, and Traditionalists representing laggards. These findings confirm that AI adoption in financial planning follows a diffusion pattern, with distinct segments emerging based on trust in AI, financial advisor reliance, and openness to employer-driven AI tools.

The five latent classes identified in the analysis exhibit distinct demographic profiles in terms of age, gender, and education (Table 7). Classes 1 and 2 skew younger, with roughly half of their members under age 35 (50.92% and 47.35%, respectively), whereas Classes 3 and 4 are older, with more than 70% of their members being over 35 (70.45% and 72.84%, respectively). Correspondingly, Classes 3 and 4 have a greater share of middle-aged respondents. For example, about 30% of Class 3 and Class 4 is age 45–64 (32.95% and 30.45%, respectively), compared to ~20% in Classes 1 and 2. Class 5 tends to be in the middle, with 42.11% being 18–34 and 21.13% being 45–64.

Figure 4 reveals key demographic patterns that both confirm and challenge conventional assumptions about technology adoption. As expected from DOI theory, Innovators and Early Adopters skew younger and higher income, with over 70% of each segment holding bachelor's degrees or higher. However, two patterns challenge traditional tech adoption stereotypes: first, gender distribution shows Innovators and Early Adopters have nearly equal gender splits (48% male), while Late Majority and Laggards are approximately 80% female; second, the Late Majority segment has the highest proportion of graduate degree holders among older segments yet demonstrates low AI adoption, suggesting education alone does not predict technology acceptance in retirement planning contexts.

Figure 4.

Socio-demographic Relationships by Latent Class.

Gender composition also varies markedly but is skewed due to the makeup of the sample. While the sample is 62.5% female, Classes 3 and 4 are about 80% female, while Classes 1 and 2 have a roughly even gender split (around 52% female versus 48% male). Class 5 is also female-majority, at approximately two-thirds female. Educational attainment follows a similar pattern of divergence. Classes 1 and 2 have the highest education levels, with over 70% of respondents holding a bachelor's degree or higher, in contrast to only 48.86% in Class 3. Class 4 is notable for the highest share of graduate/professional degree holders (34.57%), and Class 5's education profile falls between Class 3 and the more highly educated Classes 1 and 2.

Marital status and racial/ethnic makeup also differ across the classes, alongside key financial characteristics. Classes 1 and 2 report the highest marriage rates (60.37% and 57.59%, respectively), whereas Class 3 has the fewest married individuals (41.48%) and a correspondingly higher proportion of divorced or widowed respondents (19.89% combined, reflecting its older age profile). The proportion of “single, never married” hovers around one-third in all groups (28.81% in Class 4 to 32.54% in Class 2).

Racial and ethnic composition shows Class 2 and Class 4 are predominantly white (about 80% of each), while Class 1 is more diverse (73.75% white), with 20.47% Black, and the highest Hispanic representation at 18.64%. Class 5 falls in between, with 25.73% of members identifying as non-white and the highest Asian American representation (3.83%), whereas Class 3 is 76.14% white with relatively small Black (13.64%) and Hispanic (11.36%) segments, and the largest share reporting “other” race at 7.39%.

Finally, household financial characteristics distinguish the classes even further. Class 1 has the highest income with over one-third (34.91%) of its members reporting household earnings of $150,000 (USD) or more, and only 7.61% report income under $35,000. Class 2 also skews toward higher income, with 28.88% with household earnings of $150,000 or more. In contrast, Class 3 is the most financially constrained, with only 6.82% above $150,000 and 32.39% under $35,000.

Classes 4 and 5 cluster more in the middle-income ranges each, with roughly 16% in low-income brackets and—11.93% (Class 4) and 16.08% (Class 5) in the highest brackets. Consistent with these patterns, Classes 1 and 2 are far more likely to use financial professionals for advice (71.39% and 69.84%, respectively), each working with one or more advisors, whereas only around 15.34%, 24.69%, and 16.08% of Classes 3, 4, and 5 do so.

Financial advisors can leverage these findings by tailoring their service models to each segment's comfort level with AI. Figure 3 demonstrates that AI adoption complements rather than replaces advisory relationships—early adopters actually show the highest engagement with traditional advisors (72% for Innovators versus 15% for Laggards). This suggests advisors should embrace AI tools as client attractors rather than threats. For tech-savvy clients, professionals can integrate AI platforms to provide rapid data analysis while maintaining their role in complex decision making and emotional support. The unexpected finding that gender does not predict AI adoption as traditionally assumed means advisors should focus on behavioural indicators (comfort with technology, confidence in AI decisions) rather than demographic stereotypes when introducing AI-enhanced services.

Plan sponsors should implement phased rollout strategies based on our segmentation findings. Figure 2 reveals that simpler tools like income estimation and goal tracking show broader appeal across segments (45%–60% interest) compared to complex functions like investment advice (11%––61% range). This suggests sponsors should start with basic AI tools as entry points, then gradually introduce sophisticated features for early adopter segments. The identification of Employer-Driven AI Users (27.7% of employees) as a distinct bridge segment indicates that workplace-sponsored AI tools can convert employees who might resist independent AI exploration. Communications should be segment-specific: emphasize convenience and innovation for early adopters, security and simplicity for sceptical users, and employer endorsement for the bridge segment.

The findings reveal important design considerations for AI retirement planning tools. The education paradox—where highly educated Late Majority segments show low AI adoption suggests that technical sophistication alone does not drive acceptance. Instead, developers should prioritize user experience, transparency, and intuitive interfaces over complex features. The complementary relationship between AI and human advisors indicates opportunities for hybrid platforms that facilitate rather than replace advisor-client relationships, such as tools that help advisors provide more personalized recommendations or platforms that seamlessly connect users to human experts when needed.

This research suggests that AI tools might provide an alternative source of guidance for individuals who do not frequently consult financial professionals or have limited access to expert advice. Even if they have concerns about technology, they may still engage with an employer-sponsored platform that offers low-cost, on-demand financial recommendations. Educational resources or brief training sessions could improve confidence and familiarity with AI-based applications, empowering users to make more informed decisions. By lowering barriers to entry, these platforms have the potential to broaden financial advice access, particularly for employees in lower-income brackets or those who hesitate to approach traditional advisors due to cost or perceived complexity.