Research Methodology
How peer groups are formed, how percentiles are calculated, and what the score thresholds mean
Benchmarking places your scores in context. Knowing that your AI Adoption composite is 62 is useful; knowing that 62 puts you in the 71st percentile among professionals in your industry and role level is more actionable. Peer comparison converts an absolute number into a relative position.
Benchmarks only appear when there is enough data to make the comparison statistically meaningful. When a threshold has not been met, the comparison section does not appear in your results. There are no placeholders, estimated figures, or "coming soon" sections.
Peer groups are determined by the demographic information you provide at the end of the assessment. Three fields define your peer group:
Your peer group label is the combination of these three fields. For example: Financial Services / Senior Manager / 1,000-5,000 employees. When comparing your results, you are compared first against this specific peer group, and then against the broader population of all respondents for the same study.
If you have not provided demographic information, your peer group is "All respondents." You will still receive a percentile comparison against the overall distribution, but narrower peer group comparisons will not appear.
Narrow peer groups only appear when enough participants have provided matching demographics. Comparisons based on too few data points would be misleading.
| Benchmark type | Minimum completions required |
|---|---|
| Overall profile distribution | 200 completions per study |
| Industry benchmark | 30 completions per industry per study |
| Role-level benchmark | 30 completions per role type per study |
| Organisation size comparison | 50 completions per size band |
| Cross-study correlation | 100 participants with two or more studies completed |
| Longitudinal trends | 50 retakes per study |
As the dataset grows, more specific peer group comparisons will become available. Early participants may see only the overall distribution. Later participants will see finer-grained comparisons as their industry and role groups accumulate sufficient data.
Percentiles are calculated as the percentage of respondents who scored lower than you on the same study:
Percentile = (number of respondents with a lower score / total respondents) × 100
A percentile of 70 means your score is higher than 70% of all respondents who completed that study. It does not mean you answered 70% of questions correctly: there are no correct answers in these assessments. The percentile is purely a relative position.
Percentiles are calculated for:
If fewer than five total respondents have completed a study, the system returns a placeholder showing the 50th percentile with an "insufficient data" note. This threshold is low because early-access participants should not receive a false peer comparison that could change once real data accumulates.
Percentile ranking means different things depending on the study and which direction is favourable:
Each study's composite score is divided into four bands, each with a label and an interpretation. These labels describe your current position; they are not an assessment of your potential or your trajectory.
| Score | Label | What it means |
|---|---|---|
| 0-29 | Low Vulnerability | Your role has strong natural defences against AI displacement. The work you do sits at the intersection of tacit knowledge, coordination complexity, and novel judgement, where AI tools augment rather than substitute. Focus on maintaining and extending this edge. |
| 30-54 | Moderate Vulnerability | Parts of your role overlap with AI capabilities, but your overall position is not acutely exposed. Strategic adaptation will strengthen your position. Understanding which specific dimensions are lower will indicate where to focus development. |
| 55-74 | Elevated Vulnerability | Significant portions of your work are within AI's current or near-term reach. Active repositioning is recommended: moving toward coordination, evaluation, and synthesis responsibilities where AI assistance amplifies rather than replaces your contribution. |
| 75-100 | High Vulnerability | Your current role has substantial overlap with AI capabilities across multiple dimensions. Diversifying into skills that AI cannot easily replicate, and building coordination responsibilities, is advisable as a near-term priority. |
| Score | Label | What it means |
|---|---|---|
| 0-29 | Early Stage | AI adoption is nascent. Most use is confined to built-in features or single-purpose tools. Foundational investments in tools, skills, and shared practices are the priority. |
| 30-54 | Developing | AI is gaining traction but use is inconsistent across tasks or the team. Some high-value applications exist. The priority is scaling what already works and addressing the barriers that limit broader adoption. |
| 55-74 | Advancing | AI is well-integrated across several areas of work. Adoption has moved beyond basic use cases. The priority is consistency across the team and closing the gaps in specific dimensions where adoption is thinner. |
| 75-100 | Leading | AI adoption is mature and strategic. Use is deep, broad, and adaptive. The priority is maintaining momentum, managing the governance implications of advanced use, and exploring frontier applications. |
The composite for Study 3 is the maximum of the three friction dimension scores, not the average. The label describes the severity of the dominant friction type.
| Score | Label | What it means |
|---|---|---|
| 0-29 | Low Friction | Few structural barriers to AI adoption. Your work environment is well-designed for knowledge work, or you have developed effective strategies for managing friction. AI can reach its potential impact here. |
| 30-54 | Moderate Friction | Some friction points exist. Targeted interventions addressing the dominant friction type can unlock meaningful gains in AI effectiveness without requiring large-scale organisational change. |
| 55-74 | High Friction | Significant barriers are limiting AI effectiveness. The problem is structural, not personal: individual AI skill will not resolve barriers at this level. Systemic changes to how work is organised are needed. |
| 75-100 | Severe Friction | Pervasive friction is blocking AI adoption at nearly every turn. Even well-designed AI tools will underperform in this environment. Structural reform, not incremental improvement, is required. |
Benchmarks are not shown at all until the relevant threshold is met. This is an intentional design choice. An industry benchmark based on eight respondents would be statistically meaningless and could actively mislead participants about their standing.
As the dataset grows, benchmarks appear progressively: the overall distribution first, then industry and role-level comparisons, then cross-study correlations and longitudinal trends. Participants who return to their results pages will see richer comparisons as the thresholds are crossed.
If you completed an assessment early in the research program and saw limited benchmarking, signing back into your account will show updated comparisons as the dataset has grown.
Every benchmark comparison displays the number of respondents it is based on. This allows you to judge the reliability of the comparison yourself. An industry comparison based on 31 respondents (just above the 30-person threshold) is worth treating with more caution than one based on 300.
Convenience sample. Participants self-select into the studies. The sample over-represents people who are actively thinking about AI and their work. Comparisons against a broader population of all knowledge workers would likely show different distributions.
Peer group labels are descriptive. The peer group is assembled from the demographics you self-report. Percentile calculations use all respondents for the same study. As the dataset grows and peer group thresholds are met, peer group comparisons will narrow to genuinely matching participants.
Scores reflect self-report. Benchmarks compare self-reported responses, not externally measured outcomes. The comparisons are most useful for identifying relative patterns within the dataset, not for making absolute claims about performance or readiness.
The dataset is growing. Early benchmarks will shift as more participants complete the studies. A percentile of 80 when 50 people have completed a study may move when 5,000 have completed it. Treat early benchmarks as indicative rather than definitive.