Product Intelligence (Ability to Learn) (PI)

1. Abstract

The Product Intelligence (Ability to Learn) (PI) scale is a specialized psychometric instrument formulated to evaluate consumer perceptions of cognitive capabilities embedded within smart, autonomous, and technologically advanced products. Originating from the seminal conceptual and empirical work of Rijsdijk, Hultink, and Diamantopoulos (2007), the instrument operationalizes the “ability to learn” dimension as a foundational pillar within the multidimensional Product Intelligence construct. As consumer durables, internet-of-things (IoT) devices, and artificial intelligence-driven appliances increasingly integrate adaptive algorithms, measuring how users mentally model, experience, and evaluate an artifact’s dynamic learning capacity has become central to engineering, product innovation management, and consumer psychology.

Comprising a five-item measurement battery administered via a 7-point Likert scale (ranging from 1 = “strongly disagree” to 7 = “strongly agree”), the scale assesses the degree to which an interactive device captures user behavioral patterns, preserves contextual information, alters its baseline operations according to historical usage, and autonomously optimizes functional performance over extended deployment periods. Psychometrically, the instrument exhibits strong internal consistency, with composite reliability (CR) and Cronbach’s alpha values consistently exceeding established thresholds (typically α ≥ .85). Rigorous confirmatory factor analytic (CFA) frameworks establish its robust construct, convergent, and discriminant validity against adjacent dimensions such as product autonomy, reactivity, multi-functionality, and human-like interaction. By delineating the cognitive attribution users make toward learning-capable artifacts, the scale provides predictive power regarding technology acceptance, customer perceived value, cognitive trust, and post-adoption consumer satisfaction.

2. Keywords

Product Intelligence, Ability to Learn, Smart Products, Technology Acceptance, Consumer Behavior, Artificial Intelligence, Psychometrics, Scale Validation, Perceived Autonomy, Human-Computer Interaction, Machine Learning Adaptability, Innovation Management, Consumer Satisfaction, Instrument Design, Confirmatory Factor Analysis

3. Authors

The Product Intelligence measurement framework and its constituent “Ability to Learn” scale were conceptualized, operationalized, and psychometrically validated by:

• Serge A. Rijsdijk, Ph.D. — Associate Professor of Innovation Management, Department of Technology and Operations Management, Rotterdam School of Management, Erasmus University Rotterdam, Rotterdam, The Netherlands. Expertise: New product development, consumer evaluation of smart products, technological innovation.
• Erik Jan Hultink, Ph.D. — Full Professor of New Product Marketing, Faculty of Industrial Design Engineering, Delft University of Technology (TU Delft), Delft, The Netherlands. Expertise: Launch strategies for new products, consumer adoption of breakthrough technologies, design-driven innovation.
• Adamantios Diamantopoulos, Ph.D. — Chaired Professor of International Marketing, Faculty of Business, Computing and Economics, University of Vienna, Vienna, Austria. Expertise: Quantitative research methodology, structural equation modeling, psychometrics, scale development in management and marketing.

4. Purpose

The rapid proliferation of algorithmically driven consumer hardware—such as smart thermostats, robotic vacuum cleaners, intelligent wearable health trackers, and autonomous vehicular driving aids—has rendered classic conceptualizations of product performance inadequate. Traditional consumer research historically conceptualized hardware artifacts as static tools whose utility was solely determined by mechanical durability, fixed functional features, and direct ergonomic ease of use. However, the introduction of embedded computational microprocessors, sensor suites, and machine learning models transformed consumer durables into dynamic, evolving entities capable of bidirectional environmental interaction. The primary purpose of the Product Intelligence (Ability to Learn) scale is to empirically quantify consumers’ subjective evaluation of an artifact’s cognitive malleability: its capacity to acquire, store, process, and leverage operational experience to refine future functionality without requiring manual reprogramming by the human user.

In academic research, the instrument addresses a critical measurement void within the fields of Human-Computer Interaction (HCI), marketing science, and organizational innovation. Researchers utilize the scale to investigate how perceived learning capabilities influence cognitive load, perceived risk, perceived competence, and user empowerment. Conventional models like the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT) typically focus on broad utility metrics (e.g., perceived usefulness, perceived ease of use). The Ability to Learn subscale extends these paradigms by uncovering the dynamic, time-dependent attributes of smart hardware that foster longitudinal user engagement and brand attachment.

From an applied industry and design standpoint, the scale serves as an essential diagnostic mechanism during prototype testing, usability audits, and iterative new product development cycles. Product engineers and software developers can implement the scale across alpha and beta deployment phases to verify whether consumers consciously perceive algorithmic learning interventions as intended. If a product incorporates complex adaptive neural networks yet registers low scores on the Ability to Learn instrument, designers are alerted to a fundamental interface gap: the artifact’s autonomous learning behaviors are either invisible, obfuscated, or insufficiently transparent to the consumer, precluding the emergence of user trust, delight, and perceived technological sophistication.

5. Psychological Construct

The construct measured by this instrument is situated within the broader multidimensional domain of Product Intelligence (PI). Rijsdijk and colleagues operationalize product intelligence as the degree to which a product exhibits behaviors that, if observed in human beings, would be designated as intelligent. Within this macro-construct, intelligence is parsed into six distinct behavioral dimensions: autonomy, ability to learn, reactivity, cooperation, humanlike interaction, and personality. The Ability to Learn dimension specifically isolates the perceived cognitive plasticity, knowledge acquisition, and contextual memorization displayed by the physical or digital product.

To fully comprehend this construct, it is necessary to examine its key cognitive and operational components:

• Behavioral Pattern Recognition: The artifact’s capacity to monitor repetitive user interventions, environmental shifts, and operational schedules, detecting recurring sequences without explicit procedural instruction.
• Longitudinal Memory Retention: The ability of the device to store prior states, preference settings, and historical interaction data across discrete usage episodes, resisting ephemeral erasure when power cycles or contextual tasks conclude.
• Autonomous Operational Modification: The product’s functional execution alters over time. Instead of relying indefinitely on factory-installed static heuristics, the product self-modifies its decision rules to match the idiographic preferences of its primary operator.
• Performance Optimization Through Usage: The explicit perception by the user that the product functions with greater efficacy, speed, or precision at time t₂ compared to time t₁ solely as a consequence of accumulated behavioral interaction.

For example, consider a non-intelligent heating system versus a smart thermostat evaluated via this construct. A conventional programmable thermostat executes a rigid timer protocol: if programmed to heat at 06:30, it triggers the furnace at 06:30 regardless of ambient meteorological fluctuations or user sleep rhythms. Conversely, an intelligent thermostat demonstrating high “Ability to Learn” senses that the user consistently overrides settings on colder mornings, tracks how long the residential thermal envelope takes to reach target temperatures under varying humidity levels, and shifts its activation time autonomously to 06:12. The user identifies that the machine has extracted insights from behavioral discrepancies and adjusted its protocol accordingly. The Ability to Learn scale captures this precise psychological attribution of experiential knowledge acquisition.

6. Theoretical Framework

The conceptual foundation of the Product Intelligence (Ability to Learn) scale draws upon multiple intersecting theories across cognitive psychology, cybernetics, computer science, and consumer behavior:

Cybernetics and Dynamic Systems Theory

At its core, the scale is rooted in cybernetics, initially conceptualized by Norbert Wiener (1948). Cybernetics defines intelligence in non-biological systems as the capacity for closed-loop feedback control, self-regulation, and adaptive response to environmental inputs. Rijsdijk et al. advanced this framework by arguing that modern consumer electronics have graduated from first-order cybernetics (maintaining a fixed equilibrium through error correction, such as a traditional thermostat maintaining 20°C) to second-order cybernetic learning systems. In second-order cybernetics, the internal goals, reference criteria, and operational matrices of the system are themselves modified by the system’s ongoing interaction with the external environment. The Ability to Learn scale assesses whether consumers recognize and validate this higher-order adaptive loop.

Distributed Cognition and Mental Models

The scale integrates Edwin Hutchins’ theory of Distributed Cognition and Donald Norman’s cognitive engineering principles regarding user mental models. When interacting with technology, humans continuously construct mental models to predict system responses. When artifacts demonstrate machine learning, the boundary between human cognition and artifact functionality blurs. The consumer no longer treats the product as a passive cognitive offloading vessel (e.g., a paper notebook or static calculator), but as an active cognitive partner. The scale captures the psychological shift wherein the consumer attributes cognitive agency, learning memory, and inductive reasoning directly to the non-biological apparatus.

Anthropomorphism and Social Cognitive Theories

Furthermore, the instrument interfaces with the psychology of anthropomorphism. As articulated by Epley, Waytz, and Cacioppo (2007), human beings possess an innate dispositional tendency to project human psychological capacities—such as intentionality, learning capability, and emotional states—onto non-human entities. When an electronic appliance alters its behavior in alignment with subtle user routines, consumers deploy social-cognitive schemas, interpreting algorithmic parameter updates through the familiar lens of human experiential learning. Rijsdijk and Hultink capitalized on this sociocognitive reality, constructing items that map how people perceive artificial learning without requiring consumers to understand the underlying mathematical code, neural networks, or Bayesian updating equations that produce the observed adaptations.

7. Validity

The psychometric integrity of the Product Intelligence (Ability to Learn) scale has been substantiated through rigorous quantitative validation procedures across consumer samples evaluating diverse product categories (e.g., smart home appliances, mobile devices, advanced digital cameras, and autonomous maintenance systems):

Construct and Convergent Validity

Convergent validity evaluates whether individual items reliably converge on the intended underlying construct. In the initial and subsequent structural equation modeling investigations conducted by Rijsdijk et al. (2007), all standardized factor loadings for the items constituting the Ability to Learn subscale were statistically significant (p < .001) and substantial, consistently exceeding the .70 threshold (loadings ranged from .72 to .88). Furthermore, the Average Variance Extracted (AVE) surpassed the conventional .50 benchmark recommended by Fornell and Larcker (1981), demonstrating that the latent construct accounts for the majority of variance observed in its measurement indicators rather than measurement error.

Discriminant Validity

To demonstrate that “Ability to Learn” is distinct from other dimensions of product intelligence and broader technological traits, comprehensive discriminant validation testing was executed:

• Fornell-Larcker Criterion: The square root of the AVE for the Ability to Learn factor proved markedly greater than the inter-factor correlations between Ability to Learn and all adjacent dimensions (including Product Autonomy, Reactivity, Multi-functionality, and Humanlike Interaction).
• Nested Model Chi-Square Difference Tests: Setting the correlation between the latent factor of Ability to Learn and neighboring constructs to unity (φ = 1.0) yielded a statistically significant deterioration in model fit (Δχ² test, p < .001), corroborating that perceived learning is not merely a manifestation of autonomy or reactive execution, but an empirically unique cognitive dimension.

Predictive and Nomological Validity

The nomological network of the scale has been repeatedly validated within structural equation models testing consumer adoption trajectories. The Ability to Learn construct exhibits robust, positive structural paths toward key downstream psychological variables:

• Perceived Relative Advantage: Significant positive associations (β ≈ .31 to .44, p < .01), showing that consumers view self-learning hardware as delivering superior incremental utility.
• Consumer Satisfaction: Directly and indirectly enhances post-purchase satisfaction by mitigating repetitive manual configuration stress.
• Perceived Complexity / Cognitive Burden: Interestingly, research indicates a dual-effect in nomological networks: while perceived learning capability enhances perceptions of product sophistication, if algorithmic adaptability lacks transparency, it can occasionally introduce slight increases in perceived user complexity, highlighting the scale’s sensitivity in capturing nuanced consumer evaluations.

8. Reliability

The reliability of the Product Intelligence (Ability to Learn) scale has been established via classic and modern psychometric indicators:

• Internal Consistency (Cronbach’s Alpha): Across empirical scale purification and validation phases involving diverse technological product categories, the five-item subscale routinely demonstrates exceptional internal consistency coefficients. Rijsdijk et al. (2007) reported a Cronbach’s alpha (α) of .88 for the Ability to Learn dimension, safely above the widely accepted .70 exploratory and .80 confirmatory thresholds (Nunnally & Bernstein, 1994). Subsequent cross-national replications examining autonomous IoT products have registered internal consistency coefficients spanning from .84 to .91.
• Composite Reliability (CR): While Cronbach’s alpha assumes tau-equivalence (equal factor loadings), Composite Reliability provides a more precise parameterization within a confirmatory structural equation modeling framework. The CR for the scale consistently meets or exceeds .87, demonstrating that the reflective indicator set exhibits minimal measurement error variance.
• Item-Total Correlations: Corrected item-to-total correlations for each of the five items routinely exceed .65, well above the standard .35 cut-off, confirming that each specific item is deeply aligned with the overarching learning construct and contributes meaningful variance to the aggregate score.

9. Factor Analysis

During the systematic scale development process, Rijsdijk, Hultink, and Diamantopoulos (2007) subjected an initial broad pool of candidate items to successive Exploratory Factor Analyses (EFA) and Confirmatory Factor Analyses (CFA):

Exploratory Factor Analysis (EFA)

During preliminary exploratory stages, items measuring product intelligence dimensions were submitted to Principal Axis Factoring with oblique rotation (Promax), reflecting the theoretical expectation that sub-dimensions of intelligence covary in real-world artifacts. The items designated for “Ability to Learn” cleanly resolved onto a single discrete factor, accounting for substantial common variance. Factor loadings on the primary learning construct exceeded .70, while cross-loadings onto alternative dimensions (such as reactivity or autonomy) remained minimal and well below the .30 threshold.

Confirmatory Factor Analysis (CFA)

To confirm the dimensional purity and structural stability of the latent construct, Confirmatory Factor Analysis was estimated using maximum likelihood techniques via structural equation modeling software (e.g., LISREL / AMOS). The five-item reflective measurement model demonstrated exceptional goodness-of-fit indices across empirical trials, conforming to Hu and Bentler’s (1999) rigorous cut-off criteria:

• Chi-Square to Degrees of Freedom Ratio (χ²/df): Ranged between 1.45 and 2.10, comfortably below the conservative upper threshold of 3.0.
• Comparative Fit Index (CFI): Consistently reached or exceeded .97 (standard benchmark ≥ .95).
• Tucker-Lewis Index (TLI / NNFI): Maintained values ≥ .96, confirming structural integrity relative to the null baseline model.
• Root Mean Square Error of Approximation (RMSEA): Exhibited tight estimates spanning .035 to .058, accompanied by a 90% confidence interval upper bound below .08 and a non-significant test of close fit (pclose > .05).
• Standardized Root Mean Square Residual (SRMR): Documented at values ≤ .041, indicating very low discrepancy between observed and model-implied sample covariance matrices.

These robust global and local fit metrics confirm that the five reflective items function as a strictly unidimensional measurement block representing perceived product learning capability.

10. Instrument / Measurement Tool

The Product Intelligence (Ability to Learn) instrument is structured as follows:

• Test Type: Multi-item self-report questionnaire measuring consumer subjective perception of product technological capabilities.
• Measurement Paradigm: Reflective measurement model (latent variable causes the indicator responses).
• Target Respondent: End-users, consumers, product evaluators, or test participants who have interacted with or evaluated a targeted smart/interactive product.
• Item Count: 5 reflective survey items.
• Response Scale: 7-point Likert response format:
• 1 = Strongly Disagree
• 2 = Disagree
• 3 = Somewhat Disagree
• 4 = Neither Agree nor Disagree (Neutral)
• 5 = Somewhat Agree
• 6 = Agree
• 7 = Strongly Agree
• Scoring Rules:
• All five items are positively keyed (no reverse-scored items are present in the purified scale).
• Mean Index Scoring: Sum the numerical values of the 5 responses and divide by 5 to generate a composite score ranging from 1.00 to 7.00. Higher aggregate scores indicate higher perceived learning capacity.
• Latent Variable Modeling: Alternatively, researchers utilizing Structural Equation Modeling (SEM) preserve individual item indicators as observed variables loading onto a single latent construct (“Ability to Learn”) to account explicitly for measurement error.

11. Permissions & Fee and Test Year

The Product Intelligence measurement framework and its constituent subscales were formally published in the peer-reviewed academic literature in 2007 by the Journal of the Academy of Marketing Science (Volume 35, Issue 3).

• Copyright Status: The conceptual and empirical publication is copyrighted by the Academy of Marketing Science and published by Springer Nature.
• Academic Research Usage: Consistent with standard scientific conventions, the scale items may be utilized, adapted, and operationalized by non-commercial researchers, university scholars, and postgraduate students for academic, educational, and scientific inquiry without financial remuneration, provided complete formal attribution is given to the original authors (Rijsdijk, Hultink, & Diamantopoulos, 2007).
• Commercial and Industrial Application: Organizations, market research agencies, or corporate entities seeking to embed the scale within commercial diagnostic suites, proprietary benchmarking audits, or fee-for-service enterprise software platforms should review copyright permissions via the publisher (Springer Nature / RightsLink) or establish formal contact with the authors.

12. References

The theoretical and empirical foundations of this psychometric instrument are documented in the following scholarly references:

• Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864
• Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104
• Hu, L. t., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. https://doi.org/10.1080/10705519909540118
• Norman, D. A. (2013). The Design of Everyday Things: Revised and Expanded Edition. Basic Books.
• Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric Theory (3rd ed.). McGraw-Hill.
• Rijsdijk, S. A., & Hultink, E. J. (2003). “Honey, have you seen the hamsters?” Consumer evaluations of autonomous products. Journal of Product Innovation Management, 20(3), 204–216. https://doi.org/10.1111/1540-5885.2003004
• Rijsdijk, S. A., & Hultink, E. J. (2009). How smart products create customer value: Investigating the effects of smartness on value-in-use. Journal of Product Innovation Management, 26(1), 24–40. https://doi.org/10.1111/j.1540-5885.2009.00332.x
• Rijsdijk, S. A., Hultink, E. J., & Diamantopoulos, A. (2007). Product intelligence: Its conceptualization, measurement and impact on consumer satisfaction. Journal of the Academy of Marketing Science, 35(3), 340–356. https://doi.org/10.1007/s11747-007-0040-6
• Wiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press.

13. Items of the Scale