Digital Performance Management and AI

Introduction

Digital technologies have transformed performance management from an episodic administrative activity into a continuous, data-supported organizational process. Human resource information systems, collaboration platforms, mobile applications, workforce analytics, artificial intelligence (AI), and digital monitoring tools now influence how goals are set, progress is observed, feedback is exchanged, performance is evaluated, and development decisions are made.

The opportunity is substantial. Digital systems can make goals more visible, provide timely coaching, identify learning needs, and support evidence-based workforce decisions. However, technology does not automatically make performance management fair or effective. Poorly designed systems can intensify surveillance, reproduce historical bias, encourage gaming, expose sensitive data, and convert complex human contributions into misleading numerical scores.

Effective digital performance management therefore requires both technical competence and professional judgement. HR practitioners must understand platforms and analytics while also addressing explainability, privacy, cybersecurity, legal compliance, accessibility, employee voice, and human oversight.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain the structure and purposes of digital performance-management systems.
2. Assess HR platforms, mobile feedback tools, and digital monitoring practices.
3. Interpret people-analytics measures and identify common analytical errors.
4. Explain how AI can assist goal setting, feedback, and evaluation.
5. Evaluate algorithmic systems for validity, fairness, explainability, privacy, and security.
6. Design governance arrangements that preserve meaningful human oversight.
7. Apply these principles to an organizational performance-management case.

1. From Annual Appraisal to Digital Performance Management

Traditional performance appraisal commonly involved a formal review once or twice a year. Managers recorded achievements, assigned ratings, discussed development, and sometimes recommended pay or promotion. This model remains relevant in some contexts, but it has several weaknesses: feedback may arrive too late, objectives may become outdated, managers may rely on recent events, and employees may experience the process as a compliance exercise.

Digital performance management is broader than digitizing an appraisal form. It is the coordinated use of digital systems, data, and managerial practices to support:

• organizational and individual goal alignment;
• regular check-ins and developmental feedback;
• documentation of achievements and obstacles;
• coaching and learning recommendations;
• performance evaluation and calibration; and
• workforce-level analysis and planning.

The central design question is not “What data can the organization collect?” It is “What information is necessary and legitimate for better performance conversations and decisions?” A useful system combines evidence with context. Evidence may include outcomes, work products, customer information, peer observations, and learning activity. Context includes resource constraints, role complexity, team interdependence, accessibility needs, market conditions, and factors outside the employee’s control.

A socio-technical perspective

A performance platform is a socio-technical system: its results arise from the interaction of technology, users, rules, incentives, and organizational culture. The same software can support coaching in a trusting culture but become an instrument of surveillance in a punitive culture. Consequently, implementation must address both system configuration and managerial behaviour.

A simplified performance-management cycle is:

1. Translate strategy into role-relevant objectives.
2. Agree on measures, standards, resources, and review dates.
3. Collect proportionate evidence.
4. Discuss progress and provide feedback.
5. Remove barriers and support development.
6. Review outcomes in context.
7. Calibrate decisions and revise goals when conditions change.

Digital tools can accelerate each stage, but they do not replace the conversation between manager and employee.

2. The Digital Performance-Management Architecture

Most organizational systems contain several connected layers.

Integration can reduce duplicate data entry and provide a more complete view of work. It can also increase risk. A single inaccurate job title, incomplete absence record, or biased historical rating may flow across multiple systems and influence important decisions. Data integration should therefore include ownership, validation, access controls, retention rules, and documented definitions for each field.

Platform-selection criteria

HR professionals evaluating a platform should examine:

• Purpose fit: Does the system support the organization’s performance philosophy rather than impose a vendor’s assumptions?
• Usability and accessibility: Can employees and managers use it on appropriate devices and assistive technologies?
• Interoperability: Can it exchange data with payroll, learning, scheduling, and identity systems securely?
• Configurability: Can goals, rating scales, workflows, and approval rules reflect different job families?
• Auditability: Are changes, access, recommendations, and decisions logged?
• Data governance: Are collection, retention, deletion, and secondary-use rules clear?
• Security: Are authentication, encryption, vendor controls, incident response, and backup arrangements adequate?
• Employee experience: Does the system reduce administrative burden and encourage meaningful dialogue?
• Total cost: Have implementation, integration, training, maintenance, and exit costs been considered?

A vendor demonstration is not sufficient evidence. Organizations should test the system with representative users, including remote workers, frontline employees, employees with disabilities, and managers with different levels of digital skill.

3. Mobile Feedback and Continuous Conversations

Mobile applications allow employees to request feedback, record achievements, recognize colleagues, complete check-ins, and receive reminders. Short, timely exchanges can be more useful than a single annual event, particularly in fast-changing work.

A high-quality digital feedback process has five characteristics:

1. Specificity: It identifies observable behaviour or an outcome.
2. Relevance: It connects behaviour with a goal, customer need, or team standard.
3. Timeliness: It occurs sufficiently near the event to be actionable.
4. Dialogue: The recipient can provide context and respond.
5. Developmental orientation: It identifies how performance can be sustained or improved.

Digital feedback also creates risks. “Always-on” notifications can produce fatigue. Public recognition systems may reward visibility rather than substantive contribution. Emoji reactions and short comments may be misinterpreted across cultures. Employees may feel pressured to respond outside working hours. Managers may substitute frequent digital messages for real coaching.

Organizations should establish norms concerning notification frequency, response expectations, after-hours contact, confidentiality, and the difference between informal feedback and formal evidence. Employees should know who can see feedback and how it may be used.

4. Digital Monitoring and Employee Privacy

Digital monitoring may include activity logs, access records, location data, call statistics, keystroke information, screen captures, customer ratings, wearable-device data, or output measures. Some monitoring is necessary for security, safety, quality, or regulatory reasons. The issue is whether monitoring is necessary, proportionate, transparent, accurate, and connected to a legitimate purpose.

The distinction between output measurement and activity surveillance is important. A completed work product may be relevant evidence; counting keystrokes may be a weak proxy for contribution. Activity indicators can encourage employees to optimize the metric rather than the underlying objective. This is a form of Goodhart’s law: when a measure becomes a target, it may cease to be a good measure.

A privacy-respecting monitoring assessment should ask:

• What specific problem does monitoring address?
• Is a less intrusive method available?
• What data are collected, and at what level of detail?
• Who can access the data?
• How long are they retained?
• Can employees challenge inaccurate data?
• Are workers informed before collection begins?
• Could the data reveal health, religious, union, location, or other sensitive information?
• Are different groups affected differently?

Organizations should separate data needed for safety or system security from data used in performance decisions. A security log should not automatically become a productivity score. Monitoring policies must reflect applicable privacy and employment laws, collective agreements, sector requirements, and organizational policies. Legal compliance is a minimum; ethical legitimacy requires attention to dignity, autonomy, trust, and proportionality.

5. People Analytics for Performance Decisions

People analytics applies descriptive, diagnostic, predictive, or prescriptive analysis to workforce information.

• Descriptive analytics: What happened? For example, average goal completion by department.
• Diagnostic analytics: Why might it have happened? For example, whether missed goals coincide with system outages or staffing shortages.
• Predictive analytics: What may happen? For example, the likelihood that a project will miss a milestone.
• Prescriptive analytics: What action might be considered? For example, recommending additional staffing or training.

The analytical hierarchy should not be confused with certainty. A predictive model identifies a statistical association or probability; it does not prove a cause and does not determine what an organization ought to do.

Basic performance metrics

A completion rate can be represented as:

The numerator and denominator must be defined consistently. An objective should be counted as eligible only if it was active during the relevant period and assigned a clear completion status. The measure is not comparable across teams if one team has many stretch goals and another has routine goals.

A simple weighted score may be expressed as:

where S is the overall score, wᵢ is the weight assigned to criterion i, rᵢ is the normalized result for criterion i, and n is the number of criteria. The weights should sum to 1, and the results must be placed on a common scale. This model assumes that the criteria are sufficiently independent and that a compensatory weighted average is appropriate. Those assumptions may be inappropriate when safety, ethics, or legal compliance is a non-negotiable threshold.

For example, suppose goal achievement is weighted 0.50 and teamwork 0.50. An employee receives normalized results of 0.80 and 0.60. The score is:

The result is 0.70, or 70% on the selected scale. This does not mean that the employee is objectively “70% effective.” It is a constructed indicator whose meaning depends on the validity of the criteria, evidence quality, weights, and context.

Analytical validity

Before using a metric, HR should test:

• Construct validity: Does it measure the intended concept?
• Criterion validity: Does it relate meaningfully to a relevant outcome?
• Reliability: Would similar evidence produce a similar result?
• Contextual validity: Does it remain meaningful across roles and conditions?
• Fairness: Does it disadvantage groups for reasons unrelated to performance?

Correlation does not establish causation. If high performers receive more training, training participation and performance may correlate because strong employees are selected for development, not because the training caused the performance. Analysts should consider role differences, opportunity, manager effects, workload, tenure, and missing data before drawing conclusions.

6. AI-Assisted Goals, Feedback, and Evaluation

AI systems can assist performance management in several ways:

• suggest objectives based on organizational priorities and role profiles;
• identify whether goals are specific, measurable, achievable, relevant, and time-bound;
• summarize documented feedback;
• identify repeated development themes;
• recommend learning resources;
• detect incomplete review workflows;
• provide draft coaching questions; and
• flag unusual patterns for human investigation.

These uses differ in risk. Drafting a coaching question is generally lower risk than ranking employees for promotion. An AI-generated summary may omit context or reproduce an offensive statement. A recommendation trained on historical ratings may encode past managerial bias. Generative AI may produce plausible but false statements, commonly called hallucinations. Therefore, AI output must be treated as an input to professional judgement, not as an authoritative evaluation.

AI-assisted goal setting

AI can help convert broad strategic priorities into draft objectives, but goals must be negotiated with the employee and manager. A well-designed process checks:

• whether the employee has control or substantial influence over the result;
• whether the target is realistic given resources and constraints;
• whether the goal creates harmful incentives;
• whether quality and ethical standards are included;
• whether the goal is accessible and role-appropriate; and
• how changing conditions will trigger revision.

For example, an AI tool might suggest “reduce average customer response time by 15%.” The manager and employee must examine whether speed could undermine accuracy, whether the employee controls the relevant workflow, and whether customer complexity varies across cases.

AI-assisted feedback and evaluation

AI may organize evidence, but it should not silently convert language frequency, sentiment, or platform activity into a final judgement. Text sentiment is especially risky because communication styles differ across cultures, languages, disabilities, and roles. Concise communication is not necessarily poor communication, and positive language is not necessarily high performance.

High-impact decisions should include disclosure that AI was used, an understandable explanation of its role, access to relevant evidence, an opportunity to correct errors, and review by a qualified human decision-maker. Human oversight must be substantive. A manager who merely approves an algorithmic recommendation without examining it provides nominal rather than meaningful oversight.

7. Explainability, Bias, and Fairness

Explainability concerns whether people can understand how a system reached or supported an output. A technically complex model may be accurate but difficult to explain. In employment settings, affected individuals need practical explanations: what information was considered, what the output means, what it does not mean, and how to challenge it.

Bias can enter through:

• biased historical ratings;
• unrepresentative training data;
• proxy variables, such as location or school attended;
• inconsistent labels;
• missing data;
• measurement practices that favour visible work;
• feedback disparities; and
• different error rates across groups.

Removing a protected characteristic from a dataset does not guarantee fairness because other variables may act as proxies. Conversely, examining protected-group information may be necessary for auditing disparate outcomes, subject to lawful and ethical safeguards.

A practical fairness review compares selection, recommendation, or error patterns across relevant groups. The appropriate fairness criterion depends on the decision context. Equal selection rates, equal false-positive rates, and equal calibration cannot generally all be achieved simultaneously when base rates differ. Fairness is therefore not a single technical number; it requires a justified relationship between the measure, the decision, the law, and organizational values.

Bias mitigation should occur across the lifecycle:

1. Define a legitimate purpose and affected population.
2. Examine data provenance, quality, and representativeness.
3. Test outputs before deployment.
4. Monitor outcomes and complaints after deployment.
5. Provide correction and appeal mechanisms.
6. Retire or redesign the system when harms cannot be controlled.

8. Privacy, Cybersecurity, and Data Governance

Performance data can include identifiable, confidential, and sensitive information. A sound governance framework addresses the data lifecycle: collection, use, sharing, retention, access, correction, deletion, and secure disposal.

Important controls include:

• role-based access and least-privilege permissions;
• strong authentication and appropriate multi-factor authentication;
• encryption in transit and at rest;
• secure application programming interfaces between systems;
• vendor due diligence and contractual controls;
• audit logs that cannot be casually altered;
• staff training against phishing and social engineering;
• tested incident-response and business-continuity plans;
• data minimization and retention schedules; and
• regular security and algorithmic audits.

Cybersecurity is not only an IT responsibility. HR determines what data are collected, how they are interpreted, and who should access them. Managers must not export employee data into unapproved generative-AI tools. Employees should be told what information is processed and how to report a suspected breach or inaccurate record.

Privacy impact assessments and algorithmic impact assessments are useful before deployment. They should document purpose, data sources, affected groups, foreseeable harms, safeguards, residual risks, and decision owners. Depending on jurisdiction and context, organizations may also need to address data-protection law, automated-decision rights, discrimination law, labour law, occupational safety, accessibility obligations, and collective bargaining requirements.

9. Human Oversight and Governance by Design

Human oversight should be designed into the workflow rather than added after a problem occurs. A governance model can assign responsibilities as follows:

A risk-tiered approach is appropriate. Low-risk uses may include reminders or formatting assistance. Medium-risk uses may include draft summaries or learning recommendations. High-risk uses include ranking employees, recommending discipline, determining promotion, or making termination-related predictions. High-risk uses require stronger validation, transparency, human review, documentation, appeal, and ongoing monitoring. Some uses may be unacceptable because they are excessively intrusive, unreliable, or incompatible with employee rights.

A useful decision rule is that no consequential employment decision should rely solely on an opaque automated output. The decision-maker should be able to explain the evidence, identify uncertainty, consider information outside the model, and record why the final decision was reached.

10. Illustrative Case: Northstar Customer Solutions

Background

Northstar Customer Solutions is a fictional, illustrative company with 600 customer-support employees working in offices and remotely. It introduces a digital performance platform that combines customer ratings, ticket volume, response time, attendance records, peer recognition, and manager feedback. An AI service produces a monthly “performance potential” score and recommends employees for leadership development.

Problem

Within six months, remote employees receive lower average potential scores than office-based employees. Employees handling complex technical cases also score below those handling routine requests because their response times are longer. Several employees report that the system does not account for system outages, difficult customers, language differences, or accessibility-related work patterns. Managers begin treating the AI score as an objective ranking.

Analysis

HR conducts an audit. It finds that:

• response time has a large influence on the score;
• case complexity is not included;
• customer ratings vary by language and customer segment;
• peer recognition is more frequent for employees whose work is highly visible in office channels;
• historical manager ratings contain inconsistent standards; and
• employees were not clearly informed that the score would influence development access.

The score therefore has questionable construct validity. It measures a mixture of speed, visibility, customer sentiment, and historical managerial judgement rather than leadership potential. The system also creates fairness risks because job assignment and work location affect the indicators.

Approach and solution

Northstar pauses the use of the score for promotion and development decisions. It establishes a cross-functional review team including HR, managers, employees, accessibility specialists, analytics staff, and legal advisers. The team:

1. publishes a plain-language explanation of the system;
2. removes raw activity measures that have weak relevance to contribution;
3. adds case complexity and documented service quality;
4. separates security and attendance data from developmental evaluation;
5. gives employees access to the evidence used and a correction process;
6. requires managers to discuss context before recording an evaluation;
7. tests outcomes across work location, job type, language group, and disability status where lawful and appropriate;
8. introduces a rule that AI recommendations are advisory only; and
9. reviews the model quarterly and reports material changes to employees.

The revised process uses structured human review rather than one composite ranking. Employees receive a balanced discussion of customer outcomes, quality, teamwork, learning, and improvement opportunities. Leadership-development nominations require written managerial justification and review by a calibration panel.

Outcome

After two review cycles, employees report greater understanding of the process, and differences associated with work location decrease. Northstar still uses analytics for staffing and coaching, but it stops presenting a single AI score as a measure of human potential. The organization also discovers that several performance problems were caused by outdated software and staffing shortages rather than employee effort.

Learning points

The case illustrates that data-rich systems are not necessarily valid systems. Fairness requires attention to job design, opportunity, context, and process—not merely adjustment of an algorithm. Transparency, employee participation, and appeal mechanisms improve both legitimacy and data quality. Most importantly, AI should support performance conversations rather than replace them.

11. Implementation Best Practices

A responsible implementation can follow these stages:

Diagnose the performance problem

Clarify whether the organization needs better goal alignment, more timely coaching, improved documentation, development planning, or workforce planning. Do not purchase technology before identifying the problem.

Design measures with users

Involve employees and managers in defining contribution, quality, constraints, and acceptable evidence. Test whether the proposed measures can be influenced by the employee and whether they create harmful incentives.

Classify risks

Map each use from low to high impact. The more a system affects pay, promotion, discipline, job security, or access to opportunity, the stronger the safeguards must be.

Pilot and validate

Use a representative pilot. Compare system outputs with expert review, examine false positives and false negatives, test accessibility, and invite users to identify missing context.

Train managers and employees

Training should cover goal quality, feedback, interpretation of dashboards, bias awareness, privacy, cybersecurity, AI limitations, and appeal procedures. Technical training alone is insufficient.

Monitor continuously

Track adoption, data quality, disparate outcomes, complaints, overrides, security events, and unintended incentives. Establish thresholds that trigger investigation or suspension.

Preserve alternatives and appeals

Employees should have a meaningful route to challenge inaccurate data, inappropriate goals, or consequential decisions. An appeal should be reviewed by someone with authority and sufficient independence from the original decision.

Contemporary Developments

Several developments are shaping the field. Generative AI is making automated drafting and summarization widely available, but its fluency can disguise error. Skills-based systems are increasingly used to infer capabilities from work experience and learning records, creating opportunities for internal mobility while raising questions about how skills are defined. Digital work also increases the importance of asynchronous contribution, making presence-based measures less defensible. Regulatory attention is growing around automated employment decisions, transparency, discrimination, and high-risk AI applications.

The likely future is not fully automated performance management. It is augmented management: systems handle administrative tasks and identify patterns, while managers and employees interpret evidence, negotiate goals, exercise judgement, and remain accountable for decisions.

Chapter Summary

Digital performance management integrates goals, feedback, evidence, development, evaluation, and analytics through connected technologies. Its value depends on valid measures, appropriate context, responsible implementation, and constructive organizational culture.

HR platforms should be evaluated for purpose fit, usability, accessibility, interoperability, auditability, data governance, security, and employee experience. Mobile feedback can improve timeliness but may create notification fatigue, privacy concerns, and superficial interaction. Digital monitoring should be necessary, proportionate, transparent, and distinguished from intrusive activity surveillance.

People analytics ranges from descriptive to prescriptive analysis. Metrics and weighted scores are constructed indicators, not complete truths. AI can assist with drafting goals, summarizing feedback, recommending development, and identifying patterns, but it may reproduce bias, omit context, or generate false information. High-impact employment decisions require explainability, fairness testing, privacy protection, cybersecurity, correction rights, appeal, and meaningful human oversight.

Responsible governance is socio-technical and lifecycle-based. It includes clear purpose, data minimization, access controls, impact assessments, employee participation, validation, monitoring, and the willingness to suspend or redesign harmful systems. The central professional principle is that technology should strengthen fair and useful performance conversations, not reduce employees to opaque scores.

Key Terms

Algorithmic bias: Systematic and unfair differences in automated outputs or impacts.

Artificial intelligence: Computational systems that perform tasks associated with human cognitive activities, such as prediction, classification, or language generation.

Auditability: The ability to reconstruct and review data access, system changes, outputs, and decisions.

Construct validity: The extent to which a measure represents the concept it claims to measure.

Continuous feedback: Regular performance communication occurring throughout the work cycle.

Data minimization: Collecting and retaining only data necessary for a legitimate purpose.

Digital monitoring: Use of digital records or technologies to observe work, activity, outcomes, or behaviour.

Explainability: The extent to which the operation or output of a system can be understood and communicated.

Generative AI: AI that produces new text, images, code, or other content from learned patterns.

Goodhart’s law: The idea that a measure may become less useful when it is turned into a target.

Human oversight: Meaningful human review, judgement, accountability, and intervention in automated processes.

People analytics: Analysis of workforce data to support organizational and HR decisions.

Performance calibration: A structured process for reviewing and improving consistency in performance judgements.

Predictive analytics: Use of data and statistical or computational models to estimate likely future outcomes.

Privacy impact assessment: A structured examination of how a proposed activity may affect personal privacy.

Proportionality: The principle that the intrusiveness of a practice should be justified by its legitimate purpose.

Role-based access control: Restricting system access according to a person’s authorized organizational role.

Socio-technical system: A system shaped by the interaction of technology, people, processes, incentives, and institutions.

Transparency: Providing understandable information about data use, system operation, and decision processes.

Review Questions

Short Answer Questions

1. What distinguishes digital performance management from simply placing an annual appraisal form online?
2. Identify three criteria for evaluating an HR performance platform.
3. Why can activity monitoring be a weak measure of employee contribution?
4. Distinguish descriptive analytics from predictive analytics.
5. What does meaningful human oversight require in an AI-assisted evaluation process?

Long Answer Questions

1. Explain the benefits and risks of mobile feedback systems and propose principles for responsible implementation.
2. Discuss how bias can enter an AI-assisted performance-evaluation system and explain lifecycle approaches to mitigation.
3. Evaluate the privacy and cybersecurity issues associated with digital employee monitoring.
4. Explain why validity, reliability, and context matter when interpreting people-analytics measures.
5. Design a governance framework for an organization using AI to recommend employees for development opportunities.

Analytical / Application Questions

1. A firm ranks warehouse employees primarily by units processed per hour. Analyse the validity and fairness risks of this measure and propose a balanced alternative.
2. An AI tool recommends a manager for promotion because of highly positive language in peer comments. What additional evidence, safeguards, and human-review steps should HR require?
3. Using the Northstar case, identify two technical problems and two governance problems. Recommend indicators that would show whether the revised system is improving performance management.

Multiple Choice Questions with Answers

Question 1

Which statement best describes digital performance management?

A. A digital replacement for all managerial judgement

B. The coordinated use of digital tools, data, and practices to support performance processes

C. A system that measures employee activity continuously

D. An annual appraisal completed through software

Correct Answer: B. The coordinated use of digital tools, data, and practices to support performance processes

Question 2

Which measure is most likely to be a weak proxy for meaningful contribution in many knowledge-work roles?

A. Quality of a completed work product

B. Achievement of an agreed customer outcome

C. Number of keystrokes recorded

D. Evidence of solving a documented problem

Correct Answer: C. Number of keystrokes recorded

Question 3

If an employee completes 18 of 24 eligible objectives, what is the completion rate?

A. 60%

B. 70%

C. 75%

D. 80%

Correct Answer: C. 75%

Question 4

What is a primary limitation of using an AI-generated feedback summary without human review?

A. It cannot store any information

B. It may omit context or reproduce inaccurate statements

C. It always produces identical summaries

D. It prevents employees from setting goals

Correct Answer: B. It may omit context or reproduce inaccurate statements

Question 5

Which principle requires collecting only information necessary for a legitimate purpose?

A. Data minimization

B. Calibration

C. Interoperability

D. Gamification

Correct Answer: A. Data minimization

Question 6

Why does removing a protected characteristic from an AI dataset not necessarily eliminate bias?

A. AI systems cannot use numerical data

B. Other variables may act as proxies for the characteristic

C. Protected characteristics are never relevant to fairness audits

D. Removing variables always increases accuracy

Correct Answer: B. Other variables may act as proxies for the characteristic

Question 7

Which use generally requires the strongest safeguards?

A. Formatting a draft goal

B. Sending a review reminder

C. Recommending employees for termination

D. Suggesting a coaching question

Correct Answer: C. Recommending employees for termination

Question 8

A weighted score uses weights of 0.6 and 0.4, with results of 0.8 and 0.5. What is the score?

A. 0.62

B. 0.64

C. 0.68

D. 0.72

Correct Answer: C. 0.68

Question 9

What is the best example of meaningful human oversight?

A. Automatically accepting every AI recommendation

B. Disabling all analytics permanently

C. Reviewing evidence, considering context, explaining the decision, and allowing correction

D. Allowing only the software vendor to change decisions

Correct Answer: C. Reviewing evidence, considering context, explaining the decision, and allowing correction

Question 10

Why should security logs not automatically become productivity scores?

A. Security logs are always inaccurate

B. Their original purpose and meaning may differ from performance evaluation

C. Security data cannot be encrypted

D. Productivity can never be measured

Correct Answer: B. Their original purpose and meaning may differ from performance evaluation