KWL — Pivot Brief

Updated 28 April 2026 Supersedes 19 April 2026 version. Phase 1 bridge test results are integrated.

TL;DR

KWL is pivoting from music-rights intelligence to music-led cultural intelligence for brands, under the product line Cadence. The hypothesis was that music streaming behaviour would predict consumer commercial outcomes at 30–90 day lags. Phase 1 tested this rigorously and the strict pre-registered criterion failed: music does not improve forecasts of French Consumer Confidence Index over 2021–2025. The data is not silent though — multiple individual music features show TIER 1 / TIER 2 correlations with mood sub-indices, supporting a descriptive cultural intelligence product. The 30–90 day forecast claim is retracted from current marketing. Phase 2 brand-level backtest is locked and in flight — testing whether the bridge holds at brand-category granularity, where the audit suggests it likely does. Pricing today is the Fallback Scenario from the original brief; predictive premium tier reopens conditional on Phase 2.

The Pivot (Unchanged)

From: Music industry data — narrow buyer, slow sales cycle, crowded. To: Cultural Intelligence Platform for brand marketing teams — faster-moving buyer, unique signal, undefended whitespace.

Positioning: “We read how a market behaves and turn it into how a brand should speak.”

How we measure mood: daily Spotify Top 50 + Viral 50 chart data per country, enriched with audio features (valence, energy, tempo, danceability, acousticness, mode, key) and aggregated into weighted market-level signals. Behavioural proxy, not census — Spotify skews 18–35 urban, with composition shaped by playlists and promotion. Validated against population-mood survey data.

What the Science Said vs What We Found

What the science said

• Edmans et al. (2022), Journal of Financial Economics — peer-reviewed evidence that weekly chart valence coincides with same-week equity returns (+8.1 bps per 1-SD), with partial next-week reversal (−7.1 bps). Establishes that music-derived valence is a real, measurable mood signal that moves real financial outcomes.
• The bridge thesis extended this: that the underlying mood signal would persist at 30–90 day lags in consumer confidence and downstream in category and brand demand, where no equity-arbitrage mechanism operates.

What Phase 1 found

What this means honestly

The strict predictive bridge as we tested it does not deliver value at the country/aggregate-CCI/monthly granularity. But the descriptive proxy claim survives. Music chart features do correlate with French population mood — tempo and mode-major most strongly, with effect sizes well above the proxy threshold and tight bootstrap CIs that do not cross zero.

The product reframes from predictive forecasting to descriptive cultural intelligence. The forecasting product reopens conditional on Phase 2 brand-level results.

Why the Bridge Might Still Hold at Brand Level

Three reasons brand-level outcomes are likely more music-responsive than CCI:

1. Less macro absorption. Audit found music signal in raw correlations (mode-major r=+0.43 with CCI) but it dissipates with macro controls. Brand-level outcomes are less correlated with macro variables, so less of the music signal gets eaten.
2. Sub-indices already showed it. Music features × CCI sub-indices (savings intent, major purchases) showed multiple p<0.005 relationships, vs ≈ p=0.01 against headline CCI. Brand-level outcomes are more specific still.
3. Higher cadence (weekly). Phase 2 uses weekly Google Trends — 4× the observations per unit time vs monthly CCI. Same effect size becomes more easily detectable.

Phase 2 is the comeback test — designed independently of (and pre-registered before) we knew Phase 1 outcomes.

Current Status

Phase 1 complete: - Pre-registration locked 28 April 2026 prior to data inspection - All three components (H1 + H2 + forecast backtest) ran end-to-end - Results published in docs/plans/2026-04-28-phase1-bridge-test-results.md - Audit revealed origin classifier was undercounting — fixed in v2 (French-classified rate lifted from 17.6% to 42.8%)

Phase 2 locked, in flight: - 6 categories × 4 brands = 24 brands per market (added Fashion / High-Street post-lock pre-execution) - Brand selection by Rule C — composite z-score of 2019–2021 Google Trends + Wikipedia + GDELT - Outcomes: weekly Google Trends primary; Wikipedia + GDELT supplementary - Held out untouched: Germany, Japan - Pre-registration: docs/plans/2026-04-28-phase2-brand-backtest-preregistration.md (Amendment 1) - ~5 days engineering before Phase 2 backtest fires; ~1 week to result

Data infrastructure: Seven non-music sources piped (Eurostat, Yahoo Finance, Google Trends, Wikipedia, GDELT, Open-Meteo, Netflix) plus daily Apple Music charts across 34 markets. Music data via NPILABS Athena access (france_poc_v1 schema).

The Cadence Product — Today’s Reality

Predictive premium tier (the original £4,800–250K range) reopens only if Phase 2 passes. Today’s pricing reflects what is actually demonstrated: mood-proxy + cultural-narrative descriptive intelligence, not 90-day forecasting.

Pricing benchmarks unchanged: Mintel category report £1,500–4,000, Euromonitor country profile £1,500–3,000, Gartner single research note £500–2,000. Cadence Pro at £1,800–2,400 sits comfortably in line with Euromonitor’s pricing — defensible without the forecasting claim.

Y1 Revenue Target

£100–200K from 50–80 paying accounts (descriptive product economics).

If Phase 2 passes within 90 days: pricing reopens to £4,800/yr Pro tier; Y1 target adjusts to £250–500K.

Exit Thesis

The category remains consolidating:

Honest band updated: - At descriptive product, £500K ARR: realistic exit £5–15M - At predictive product (if Phase 2 passes), £500K–1M ARR: realistic exit £10–30M

WGSN £700M is what compounding produces over 20 years; Stylus £19.4M is the realistic comp for an early-revenue entrant.

Plausible acquirers: Omnicom (Sparks & Honey, Backslash), Sprinklr, Cision, Spotify/SoundCloud, Luminate, a PE-backed trend roll-up, or a holding-group LLM play.

What’s Next

Continuous data collection means by mid-2027 we’ll have ~80–90 monthly observations vs today’s 60. At observed Phase 1 effect sizes, several near-misses would flip to significant. Time itself improves the predictive case.

What I’m Asking of You

1. Acknowledge Phase 1 results. The strict predictive claim does not hold at country/CCI scale. Pricing and positioning have been adjusted accordingly. This brief is the receipt.
2. Stress-test the Phase 2 design if you see flaws. Pre-registration (docs/plans/2026-04-28-phase2-brand-backtest-preregistration.md) locks brand-selection rule, outcome variables, lag windows, and pass criterion. Now is the time to push back if anything looks methodologically weak.
3. Open your network at Phase 2 readout. If Phase 2 passes (~2 weeks from now), the predictive case is back; that’s when introductions to strategic acquirers become useful. If Phase 2 fails, the descriptive product still has a path — but to a smaller acquirer set.
4. Approve the IAM extension to sugr-pulse bucket — needed for the v3 origin classifier with artist-area lookup. 5-minute AWS console action. Not a blocker but tightens our data quality story for Phase 2.

What This Brief Does Not Hide

• Phase 1 hypothesis as originally pitched failed.
• Pre-registration discipline held — the test was honest.
• The descriptive product is the realistic offering today.
• Phase 2 may rescue the predictive offer; may also fail.
• Either outcome is published.

That’s the position.

Source artefacts: - Pre-registrations: docs/plans/2026-04-28-bridge-test-preregistration.md, docs/plans/2026-04-28-phase2-brand-backtest-preregistration.md - Phase 1 results: docs/plans/2026-04-28-phase1-bridge-test-results.md - Forecast plot: data/h1_forecast_backtest.png - Edmans et al. (2022): JFE 145(2), 234–254. SSRN 3776071