← Home

Partner Basecamp — Programme Findings

About this site
← All reports

A single source of truth for Partner Basecamp programme data

This site publishes the survey findings from every Partner Basecamp cohort — updated after each event and shared with BTS colleagues delivering the programme. It is a single source of truth: a consistent, versioned record that anyone with a BTS login can access and compare across cohorts.

Reports are generated directly from participant survey responses using a pipeline of Python scripts. Everything on the site comes from that data with no manual editing. When a new cohort runs, the reports are regenerated and the site is updated within hours.


Four types of report — each serving a different purpose

Global trends

Audience & reach

Cross-cohort view of who attended, where, NPS over time, org reach, and capability profile. Descriptive — no interpretive commentary. Run after every cohort.

Programme insights

NPS, outcomes & patterns

Interpretive analysis: NPS by persona and proficiency, confidence before vs after, open-text synthesis, and systematic patterns across the programme.

Single cohort

Per-event analysis

Detailed breakdown for one cohort: audience, NPS, D1 ratings, D2 confidence, open text. Generated by analyze.py for each event.

Org profiles

Per-firm survey record

NPS history, participation, persona mix, and benchmark comparisons per partner firm, alongside a feedback log and open action items. Generated for firms meeting the participation threshold.


Key metrics explained

The metrics here serve different analytical purposes and build on each other. Descriptive metrics — NPS, attendance, org reach — tell us what happened and who was there. Diagnostic metrics — persona breakdowns, confidence deltas, proficiency segments — help explain why the programme landed the way it did for different groups. The pipeline scripts produce both. Turning those into insights — interpreting what the patterns mean, what's driving them, and what to do about them — is where the analyst takes over from the script.

Metric What it measures & how to read it
NPS Day 2 recommendation score, −100 to +100. Weighted aggregate across cohorts accounts for different cohort sizes. An open-text follow-up question is asked of all respondents regardless of score — participants say what's behind their rating. This is the primary diagnostic layer for NPS: the scores tell us where we stand; the open text tells us why.
Confidence delta Change in self-rated confidence from Day 1 (before the programme) to Day 2 (after). Positive delta = participants left more confident than they arrived. Reported per competency area and as an overall mean.
Depth / Pace / Relevance Three Day 1 ratings (1–5 scale) on programme design: was the content deep enough, was the pace right, was it relevant to their role? Shown as mean scores with distribution bars. These feed back to facilitation decisions for future cohorts.
Personas These are Anthropic's definitions of the partner ecosystem roles the programme is designed to serve. Participants self-select from four function groups in the Day 1 survey; those map to three personas used throughout the reports. NPS and outcomes differ meaningfully across all three — the splits are more useful than the programme average alone.

Developer (Engineering) — Engineers and developers who build, integrate, and deploy AI solutions. They work at the implementation layer: writing code, integrating APIs, and putting AI capabilities into production.

Architect (Architecture) — Technical architects responsible for AI strategy, system design, and the frameworks that guide how AI is adopted across an organisation or client engagement. They operate at the intersection of technical depth and strategic direction.

Transformation Lead (Business Leadership · Project/Engagement Management) — Commercial and delivery leads who connect AI work to client outcomes. They operate across business development, engagement management, and strategic leadership — the people who decide whether AI initiatives get resourced, scoped, and delivered.
AI proficiency Self-reported on Day 1: Learning and exploring, Applying in practice, Delivering independently, or Operating at the frontier. Used to segment NPS and confidence data — Frontier participants typically start with higher confidence but show smaller deltas; those still learning and exploring show the largest shifts.
D1 / D2 respondents D1 = Day 1 Pulse survey respondents (programme design feedback). D2 = Day 2 End of Session survey respondents (NPS, confidence). D2 is typically smaller as some participants leave before the closing session. NPS and confidence figures come from D2.

How data gets from survey to report

It helps to distinguish three levels of output the pipeline produces. Data is what analyze.py extracts — numbers, counts, distributions. NPS is data. Findings are patterns that emerge when data is read across cohorts and segments — "NPS varies significantly by persona" is a finding, surfaced by insights.py. Insights are interpretive claims that explain findings and point toward action — "the Architect NPS gap reflects a content calibration issue, not audience quality" is an insight, and it requires analyst judgment to reach. The pipeline handles the first two; the third requires a human in the loop.

1
Participant completes the survey
Two Microsoft Forms surveys per cohort — Day 1 Pulse (design feedback) and Day 2 End of Session (NPS, confidence, open text). Participants access via QR code displayed in the room. Questions are standardised across all v2 cohorts (Cohort 6 onwards).
2
Forms exported and combined (merge_forms.py)
After the event, the two Excel exports from MS Forms are downloaded and merged into a single workbook. During this step, email addresses are pseudonymised before the file is stored.
Privacy step
3
Metrics extracted (analyze.py)
The combined workbook is processed by analyze.py, which calculates all metrics — NPS, confidence deltas, persona breakdowns, org distribution — and writes them to a JSON file stored in Box. No identifiable information is written to the JSON; email addresses are not included in any output. Facilitator notes from the session — observations on room dynamics, content delivery, and participant engagement — are also incorporated here. They provide context the survey data alone cannot capture and feed into the interpretive layer.
PII stripped
4
Analyst review and open-text interrogation (ask.py)
This is where findings become insights. analyze.py produces a Tier-2 analyst review alongside each cohort report — a set of flags for segments that deviate from programme benchmarks, with suggested ask.py queries for each one. The analyst works through those flags using ask.py, a script that routes structured questions about the open-text responses through Claude. This interrogation layer surfaces themes, tests hypotheses, and builds the interpretive narrative that cannot be derived from quantitative metrics alone. Conclusions can be published back into the cohort report and the trends Field Notes log. The Trends report, by contrast, is purely descriptive and requires no analyst interpretation — it runs straight from the data.
5
Reports regenerated
Three report types are published to the site from the cohort JSON files: the single-cohort findings report (analyze.py, one per event), the cross-cohort trends report (trends.py, updated after every cohort), and the interpretive insights report (insights.py, updated following analyst review). Org profiles are also generated per partner firm (org_profile.py) for firms meeting the participation threshold. No raw survey data is embedded in any HTML output — only aggregated metrics.
6
Site updated
Updated HTML reports are copied to the git repository's docs/ folder and pushed. The site updates within about one minute. Raw survey data never enters the git repo or this site.
What's stored and what isn't. Data security for this programme is our responsibility to build and govern — there is no platform handling it for us. Email addresses are pseudonymised (the local part is replaced with a one-way hash) before the merged workbook is saved to Box. Organisation names are derived from the email domain — e.g. accenture.com — and are retained because they're programme-level data, not personal identifiers. No names or email addresses appear in any report, JSON metrics file, or on this site. Raw data files will not be made available. If you have older raw survey data from earlier cohorts on your local machine, delete it.

Planned additions to the pipeline and site

These are deferred until the underlying infrastructure is ready. The current pipeline — Box for data, git for scripts, Cloudflare Pages for hosting — is the stable foundation they'll build on.

Interactive chart filtering Architect signal cross-cohort analysis Facilitator query tool SQL data store

Questions, gaps, or things that look wrong?

This site is only useful if the data it shows is trustworthy and the reports answer the right questions. If you notice something that looks off, want to see a metric that isn't here, or have a question about how something is calculated, reach out to Ted.