Q: Why does Architect consistently score +28 pts above programme NPS across 4 cohorts — is this audience fit, content relevance, or delivery?

The data points to audience fit as the primary driver, but content relevance is a co-factor — delivery cannot be assessed from available data.

Architects score +67 (C6), +38 (C7), +88 (C8), and +75 (C9) versus programme NPS of +43, +32, +43, and +38 respectively — a consistent gap averaging approximately +28 pts. The most plausible explanation is audience fit: Architects are pre-selected for system-design thinking and arrive with sufficient technical grounding to engage with content pitched at an 'Applying in practice' to 'Delivering independently' level, whereas Transformation Leads and Developers show more variance (Transformation Lead ranges from +25 to +80; Developer from +17 to +62). Content relevance is a co-factor: the single highest Architect verbatim explicitly cites 'well matched against the product capabilities and how to think about the larger set of questions for clients' (C8 Architect, frontier proficiency). Detractor verbatims from non-Architect personas repeatedly cite the programme being 'too foundational' or 'not beginner friendly,' implying the calibration suits Architects better than other groups. Delivery as a distinct driver cannot be isolated — no facilitator ratings, session-level scores, or delivery-specific survey items are present in the data. Key data points: · Architect NPS by cohort: C6 +67, C7 +38, C8 +88, C9 +75 — all above cohort programme NPS · Architect n is small across all cohorts: C6 n=6, C7 n=8, C8 n=8, C9 n=4 — findings are directionally consistent but not statistically robust · Key Architect promoter verbatim (C8): 'Well matched against the product capabilities and how to think about the larger set of questions for clients' — directly citing content-to-role fit Caveats: Architect n ranges from 4 to 8 per cohort — below the 20-respondent threshold for confident segment conclusions; treat as directional signal only. To distinguish audience fit from content relevance from delivery, you would need session-level ratings broken out by persona, facilitator scoring, and a pre-programme role-relevance self-assessment.

2026-07-28 18:20 UTC

Q: Why do Delivering independently AI proficiency participants rate the programme +17 pts above average — what does the D1 calibration data tell us?

The D1 calibration data does not include NPS broken out by AI proficiency segment, so the +17 pt advantage cannot be directly explained from the available data.

The data brief confirms that 'Delivering independently' participants rate the programme +17 pts above the programme average NPS of +37, but NPS scores are not disaggregated by AI proficiency band in any cohort — only headcounts per band are provided (C6: n=8, C7: n=14, C8: n=11, C9: n=13, totalling n=46 across four cohorts). D1 calibration scores are available at cohort level only (ranging from 3.95 in C9 to 4.08 in C7), not segmented by proficiency. Verbatim evidence points in a consistent direction: detractors and passives are disproportionately advanced practitioners frustrated by content pitched too foundationally, while 'Delivering independently' participants may sit in a sweet spot — experienced enough to apply the material immediately, but still gaining structured frameworks they lack. The one 'Delivering independently' promoter verbatim explicitly values 'fundamental AI engineering principles,' suggesting the programme's technical grounding resonates at this level without feeling remedial. Key data points: · 'Delivering independently' n across C6–C9: 8, 14, 11, 13 — total n=46, sufficient for a segment-level cut but not currently reported · D1 confidence scores range 3.95–4.08 across cohorts, but are not broken out by proficiency band — no D1 calibration comparison is possible by segment · Detractor verbatims skew toward 'Operating at the frontier' (e.g., valantic Architect, nps=6; Horváth Developer, nps=4) and passives toward 'Applying in practice', suggesting 'Delivering independently' avoids both ceiling and floor dissatisfaction effects Caveats: To answer this question properly, NPS and D1/D2 confidence scores need to be cross-tabulated by AI proficiency band. Specifically: NPS by proficiency segment per cohort, and mean D1 confidence score and D1→D2 confidence delta by proficiency band would isolate whether the effect is driven by pre-existing confidence level, confidence gain, or content-fit perception.

2026-07-28 18:20 UTC

Q: What explains PwC's persistent below-programme NPS score across 2 cohorts, and what would change it?

PwC's NPS averaged -23 pts across C7 (-45, n=11) and C8 (+0, n=5), driven by a consistent signal that the programme is perceived as too basic for their attendees' existing skill levels.

In C7, PwC scored -45 vs. a cohort NPS of +32 — a 77-point gap — and in C8 scored +0 vs. a cohort NPS of +43, a 43-point gap. Verbatims from PwC participants point to a single recurring theme: the content is pitched below their experience level. C7 PwC detractors explicitly said 'good for entry level with no Claude code experience' and called for 'more dialogue and discussion' rather than heads-down coding they felt could be done independently. The C8 PwC detractor felt 'lost' and 'behind' — the opposite direction — suggesting PwC may be sending a mixed-seniority group without sufficient pre-segmentation. Notably, the one C7 PwC promoter ('very helpful for a beginner') was a Practitioner-level Transformation Lead 'Applying in practice', confirming the content lands well for less experienced attendees but not for more seasoned ones. To change this, PwC would need to either send attendees better matched to the programme's current level, or Anthropic would need to create a streamed or advanced track for participants already delivering independently. Key data points: · PwC C7 NPS: -45 vs. cohort +32 (77-point gap), n=11 · PwC C8 NPS: +0 vs. cohort +43 (43-point gap), n=5 · PwC C7 detractors (3 verbatims) all cited content being too basic or format mismatched to in-person value · Only PwC promoter in C7 was entry-level ('Applying in practice', Practitioner seniority) — confirming the content works for newer practitioners · PwC is absent from C6 and C9 org-level data, so the pattern is confined to two cohorts Caveats: C8 PwC n=5 is well below the 20-respondent threshold for confident conclusions — treat as directional only. A breakdown of PwC attendees' AI proficiency levels across cohorts would confirm whether PwC consistently sends more experienced practitioners, which is the most actionable diagnostic available.

2026-07-28 18:20 UTC

Q: What is driving the high passive rate in C9 London May 20-21, 2026 — what do the verbatims say and what would convert them to promoters?

C9 London's 45.8% passive rate is driven by experienced participants finding content insufficiently advanced — verbatims point to depth and instructor capability as the two conversion levers.

C9 London has the highest passive rate across all four cohorts at 45.8% (n=24), compared to 25–35% in the San Francisco cohorts. Three passive verbatims are available and all score 7–8; none express strong dissatisfaction, but none articulate a clear reason to recommend. The detractor verbatims are more revealing: a frontier-level Developer from Horváth (NPS 4) explicitly states 'Few new things but mainly existing topics you mostly know when working with claude. Instructors did not seem to be capable of answering deeper technical questions.' A second detractor (NPS 5) notes the content is 'More for less experienced recipients.' This pattern mirrors C6 London passives, where a frontier-level SFEIR Developer (NPS 7) said content 'was already kind of known' for advanced users. C9's cohort skews experienced: 13 participants are 'Delivering independently' and 13 are 'Operating at the frontier' out of 55 day-one attendees, meaning roughly half the room may have outgrown the current content level. Converting passives to promoters likely requires either a streamed advanced track or deeper technical modules — and demonstrably stronger instructor capability on complex questions. Key data points: · C9 passive rate: 45.8% — highest across C6–C9 (next highest is C8 at 35.1%) · 26 of 55 C9 day-one participants (47%) are at 'Delivering independently' or 'Operating at the frontier' proficiency · Detractor NPS 4 verbatim explicitly cites shallow content depth and instructor technical limitations as the failure points Caveats: Only 3 passive verbatims are available for C9 (n=24 NPS respondents), which is insufficient to generalise with confidence. The passive verbatims themselves are thin ('Great content', 'It helped a lot') and do not explain the 7–8 scores — a follow-up open question targeting passives specifically would sharpen the conversion diagnosis considerably.

2026-07-28 18:21 UTC

Q: What do promoters and passives say about the programme, and what specifically separates them?

Promoters value practical depth and clear applicability; passives liked the content but felt it was too broad or too introductory for their level — the gap is calibration to experience, not content quality.

Promoters consistently cite specific, actionable gains: technical depth on Claude ('very useful technical dive into Claude'), foundational AI engineering principles, client framing ('well matched against product capabilities and how to think about larger questions for clients'), and structured coverage of important topics. Their language signals confidence and utility. Passives, by contrast, express general satisfaction — 'great content,' 'good course,' 'it was very good' — but their criticism, where present, centres on level-fit: one C6 passive (Operating at the frontier, Developer, SFEIR) explicitly noted 'most of the content was already kind of known,' and a C7 passive (PwC) wanted 'more build guidance instead of working off pre-built plumbing.' The separating factor is not whether participants enjoyed the programme, but whether it delivered new capability or stretch relative to their existing proficiency. Detractors reinforce the same pattern at greater intensity: advanced participants across C6, C7, and C9 flag insufficient depth, while less experienced participants in C8 flag being left behind — confirming a level-calibration problem in both directions. Passives sit in the middle: satisfied enough not to detract, but not stretched enough to advocate. Key data points: · Passives represent 31.7% of all respondents across C6–C9 (n=142); C9 London has the highest passive rate at 45.8% (n=24), flagged as high-risk for churn · Promoter verbatims consistently reference specific content value: 'technical dive into Claude,' 'fundamental AI engineering,' 'how to think about larger questions for clients' — all outcome-oriented · Passive verbatims where substantive (C6 SFEIR, C7 PwC) cite level-mismatch: content already known to more advanced participants, or desire for more hands-on building vs. pre-built scaffolding — not dissatisfaction with delivery Caveats: Verbatim coverage is sparse — only 3 promoter, 3 passive, and 3 detractor quotes per cohort are available, limiting confidence in thematic conclusions. A full verbatim export with NPS scores and proficiency tags would allow quantitative theme coding to validate whether level-calibration is the dominant passive driver or one of several.

2026-07-28 18:21 UTC