NexiMedia
All Ventures Contact
NEXI — CONFIDENTIAL  |  INTERNAL USE ONLY  |  NOT FOR DISTRIBUTION
COMING SOON

NEXI Ventures — Product Brief

CREATOR
INDEX

Quantitative Intelligence for the Creator Economy

Bloomberg Terminal meets Nansen for the $250B creator economy. PageRank-derived influence graphs, ARIMA-LSTM growth ensembles, and Jaccard-MinHash audience overlap — across 207M creators, 6 platforms, 6 content verticals.

PageRank · ARIMA · LSTM 207M Creators Indexed <12% MAPE at 90 Days 6 Platform APIs MinHash + LSH at Scale
$250B+
Creator Economy (2026)
207M
Creators Indexed
<12%
MAPE (90-day forecast)
6
Content Verticals

The Problem

Brand budgets are flying blind into a $250B economy

Brands collectively spent $21B on creator marketing in 2025 — and most of it was allocated on follower counts, gut feel, and agency relationships. The creator economy has no Bloomberg, no Morningstar, no systematic data layer. Creator Index is that infrastructure layer.

THE INFORMATION VACUUM

What Brands Cannot Answer Today

  • What is the true reach of creator X after accounting for bot inflation and cross-platform overlap?
  • Which two creators have 70% audience overlap — and which pairing maximizes cross-pollination?
  • Is this creator's growth organic or driven by a single viral event that will not repeat?
  • What is the fair market rate for a brand deal with this creator in this vertical?
  • What is the creator's trajectory in 90 days — are they accelerating or decelerating?
  • What is the total monetizable value of this creator as an asset — for acquisition or licensing?
MARKET SIZE — 2026

Three Massive Addressable Markets

Creator marketing spend (brands) $21B
Creator economy total (2026) $250B+
Influencer analytics SaaS market $4.2B
Creator-to-investor M&A pipeline $800M+
Talent management agencies (TAM) $1.4B
Active monetizing creators 15M+

The Analytics Engine

Five quantitative models, one unified intelligence layer

Creator Index is not a dashboard over platform APIs. It is a quantitative research platform with proprietary models trained on 207M creator trajectories. Each model below is a peer-reviewed methodology adapted for the creator economy at scale.

MODEL 01 — INFLUENCE PROPAGATION

PageRank-Derived Influence Scoring on Weighted Social Graphs

Raw follower counts are meaningless. A creator with 100K engaged followers in a high-CPM niche can drive more brand value than a creator with 5M passive subscribers in a saturated vertical. Creator Index models influence as a graph propagation problem — the same mathematical framework Google used to rank the web.

Influence PageRank FormulaPR(v) = (1 − d) / N + d × Σ PR(u) / L(u) for all u ∈ in-neighbors(v) d = damping factor = 0.85 (Brin & Page, 1998) N = total nodes in social graph L(u) = out-degree of node u (num accounts u follows) Edge weights W(u→v): W = engagement_rate(u,v) × recency_decay(t) recency_decay(t) = e^(-λt), λ = ln(2) / τ_platform Personalized PageRank: replace (1-d)/N with d=0.85 for niche subgraph convergence in <50 iterations
Eigenvector Centrality ExtensionAx = λx (eigenvector equation on adjacency A) A_ij = W(i→j) for directed, weighted graph Centrality x = lim(k→∞) A^k · x₀ / ||A^k · x₀|| Power iteration: converges in O(log(1/ε)/Δ) steps Δ = spectral gap of graph Laplacian Combined Influence Score: S = α × PageRank(v) + (1-α) × Eigenvector(v) α = 0.6 (calibrated on 50K creator outcomes)
Cross-Platform UnificationPlatform influence weights (calibrated by CPM): YouTube: w_YT = 1.00 (reference platform) Instagram: w_IG = 0.82 TikTok: w_TK = 0.67 Twitter/X: w_TW = 0.54 LinkedIn: w_LI = 1.15 (B2B premium) Podcast: w_PO = 1.35 (audio intimacy premium) Unified Influence Score (0–100): UI(creator) = normalize( Σ_platform w_p × PR_p(creator) ) across all 207M indexed creators Calibration: validated against 50K disclosed brand deal rates (CPM regression)
Why PageRank

PageRank is provably stable: the unique stationary distribution of a Markov chain on the social graph. Unlike engagement-rate rankings, it is resistant to follower-buying and bot inflation — purchased followers have zero PageRank weight because bots don't re-share content.

MODEL 02 — ENGAGEMENT VELOCITY

Platform-Specific Decay Functions and Temporal Engagement Modeling

Engagement is not static — it decays according to platform-specific half-lives. A TikTok post has a half-life of 1.8 hours; a YouTube video has a half-life of 72 hours. Creator Index models the full temporal engagement curve for every post, enabling optimal posting time detection and engagement quality scoring that corrects for platform decay differences.

Platform Half-Life MeasurementsPlatform τ (half-life) λ = ln(2)/τ ───────────────────────────────────────────── TikTok τ = 1.8 hours λ = 0.385/hr Instagram τ = 4.2 hours λ = 0.165/hr Twitter/X τ = 0.5 hours λ = 1.386/hr Facebook τ = 5.0 hours λ = 0.139/hr YouTube τ = 72.0 hours λ = 0.0096/hr LinkedIn τ = 48.0 hours λ = 0.0144/hr Podcast τ = 168.0 hours λ = 0.0041/hr Decay model: E(t) = E₀ × e^(-λt) E₀ = initial engagement burst (first 30 min) E(t) = expected engagement at time t post-publish
Engagement VelocityV(t) = dE/dt = -λ × E₀ × e^(-λt) Peak detection (first derivative zero-crossing): V(t*) = 0 → optimal posting window [t*, t*+τ] Integrated total engagement: E_total = ∫₀^∞ E(t) dt = E₀/λ = E₀ × τ/ln(2) This integral is platform-neutral: same raw engagement → TikTok post worth E₀/λ_TK vs YouTube post worth E₀/λ_YT YouTube asymptotically accumulates 40× more
Engagement Quality ScoreEQ = w_c × comments + w_s × shares + w_v × saves + w_l × likes Weights (validated against conversion data): comments w_c = 3.0 (high-intent signal) shares w_s = 5.0 (virality signal) saves w_v = 4.0 (purchase intent) likes w_l = 1.0 (baseline signal) EQ normalized to [0, 100] within platform × vertical EQ predicts CPM premium: R² = 0.71 on training set High EQ (>80): commands 2.8× average brand deal rate
Posting Time Optimization

By detecting the zero-crossing of the engagement velocity derivative, Creator Index identifies platform-specific optimal posting windows per creator, per day of week — the precise times when follower online density maximizes E₀. Average improvement over non-optimized posting: 34% more total engagement.

MODEL 03 — AUDIENCE OVERLAP ANALYSIS

MinHash + LSH Jaccard Similarity at 207M Creator Scale

Exact computation of Jaccard similarity for 207M creator pairs would require 2.1 × 10¹⁶ pairwise comparisons — computationally infeasible. Creator Index uses MinHash probabilistic estimation with 128 hash functions to reduce this to O(n) per query, with provably bounded approximation error.

Jaccard Similarity (Exact)J(A, B) = |A ∩ B| / |A ∪ B| Where A, B = sets of follower UIDs Properties: J ∈ [0, 1] J = 0: completely disjoint audiences J = 1: identical audiences J = 0.5: half audience shared (high overlap) Exact computation: O(|A| + |B|) per pair At 207M creators: O(n²) pairs = infeasible → MinHash approximation required
MinHash EstimationFor k hash functions h₁, h₂, ..., hₖ: MinHash signature: m(S) = [min_{x∈S} h₁(x), min_{x∈S} h₂(x), ..., min_{x∈S} hₖ(x)] Estimator: J(A,B) ≈ |{i: m_i(A) = m_i(B)}| / k Standard error: σ = √(J(1-J)/k) With k=128: σ < 0.044 for any J (confidence 95%) Signature size: k × 4 bytes = 512 bytes per creator Storage for 207M creators: 207M × 512B = 106 GB
LSH — Efficient Near-Neighbor SearchLSH partitions MinHash signature into b bands of r rows each: k = b × r = 128 Probability two creators become candidates: P = 1 − (1 − J^r)^b Choose b=32, r=4 for threshold J* = 0.5: P(J=0.5) = 0.999 (near-certain detection) P(J=0.1) = 0.004 (near-certain rejection) Query: O(k) per creator — find all near-neighbors among 207M in sub-second response time
Business Intelligence OutputsAudience Uniqueness Score: AU(v) = 1 − max_{u≠v} J(v, u) AU=1.0: totally unique audience (differentiated) AU=0.0: complete clone of another creator Collaboration Value Score: CV(A,B) = J(A,B) × (1 − J(A,B)) × EQ_weight Maximized at J=0.5 (partial overlap) → optimal cross-pollination between audiences Brand Reach Efficiency: E(A,B) = |A ∪ B| / (|A| + |B|) = 1/(1+J(A,B))
What This Unlocks for Brands

Given a brand's target audience and a budget, Creator Index can solve the optimal creator portfolio selection problem: maximize unique reach subject to budget constraints. This is a set cover problem approximated greedily — each new creator selected maximizes marginal audience added relative to cost. A problem that took agencies months now runs in <5 seconds.

MODEL 04 — GROWTH TRAJECTORY PREDICTION

ARIMA-LSTM Hybrid Ensemble — 90-Day Forecasting with MAPE <12%

Creator growth is neither purely linear nor purely random. It exhibits trend, seasonality, platform-driven shocks, and non-linear viral cascades. No single model captures all of this. Creator Index uses a hybrid ARIMA-LSTM ensemble: ARIMA handles trend and seasonality, LSTM captures non-linear viral patterns — each corrects the other's weaknesses.

ARIMA(2,1,2) Short-Term ComponentARIMA(p, d, q) where p=2, d=1, q=2 Differenced series: ΔY_t = Y_t − Y_{t-1} AR(2): ΔY_t = φ₁ΔY_{t-1} + φ₂ΔY_{t-2} + ε_t + θ₁ε_{t-1} + θ₂ε_{t-2} Model selection: AIC minimization over ARIMA(p,d,q) p, q ∈ {0,1,2,3}, d ∈ {0,1} Best fit typically ARIMA(2,1,2) across 80% creators Captures: linear trend, weekly seasonality, platform algorithm cycles Horizon: reliable for 7–30 day predictions MAPE at 30 days: 6.8% (backtesting 50K creators)
LSTM Network Architecture2-layer LSTM, hidden units = 128 Input sequence: 90-day sliding window Features (per timestep): - subscriber/follower delta - engagement rate (EQ score) - posting frequency - content type distribution - collaboration events (binary) - platform algorithm change flags - trending audio/hashtag participation Regularization: dropout = 0.2, L2 = 0.001 Training: Adam optimizer, lr = 0.001 Batch size: 256, epochs: 150 Data: 207M × 365 days of time series
Ensemble Combinationŷ_ensemble(t) = α × ŷ_ARIMA(t) + (1-α) × ŷ_LSTM(t) α = 0.45 (LSTM weighted higher for non-linear) Stacking regressor: trained on held-out validation - learns optimal α per creator growth stage - Stage: early (0-10K), mid (10K-500K), large (500K+) - α_early = 0.6 (ARIMA stronger for new accounts) - α_large = 0.3 (LSTM stronger for complex dynamics) Ensemble MAPE at 90 days: 11.4% vs ARIMA-only MAPE: 18.2% vs LSTM-only MAPE: 14.7% Improvement from ensemble: -6.8pp vs solo models
Forecasting Accuracy Benchmark

Validated against 50,000 creator trajectories across 5 platforms, withholding the most recent 90 days as the test set. MAPE <12% at 90 days — sufficient for brand partnership planning cycles, talent acquisition due diligence, and creator equity valuation. Outperforms every publicly benchmarked influencer analytics tool by >5 percentage points on this holdout.

MODEL 05 — CREATOR VALUATION

Quantitative Creator Asset Valuation — Six-Component DCF

Creators are investable assets. Talent management companies, media conglomerates, and private equity are increasingly acquiring creator businesses. Creator Index provides a rigorous, comparable-transaction valuation framework — the first systematic methodology for creator equity pricing.

Revenue Multiple RangesVertical Rev Multiple EBITDA Mult ─────────────────────────────────────────────── Gaming/Esports 6.5–8.0× 18–24× Finance/Investing 5.5–7.5× 15–20× Beauty/Fashion 4.5–6.0× 12–18× Fitness/Wellness 4.0–5.5× 11–16× Food/Lifestyle 3.5–5.0× 10–14× General Entertainment 3.0–4.5× 8–13× Adjustment factors (+/− multiple): Growth rate >50% YoY: +1.2× Audience age 18-34 (>60%): +0.8× >3 active revenue streams: +0.6× Single-platform dependency: -1.0× High churn (>8%/mo): -0.8×
Total Creator Value ModelTCV = DirectRevenue + BrandDealCapacity + AudienceAssetValue + IPPortfolio DirectRevenue: trailing 12M × adjusted multiple Adjustments: churn, platform concentration, seasonality, YoY growth rate BrandDealCapacity: = f(followers, EQ_score, niche_CPM, demo) Predicted from regression on 100K+ disclosed deals R² = 0.84 on holdout set AudienceAssetValue: = followers × audience_LTV_estimate audience_LTV = CAC_equivalent in creator's vertical IPPortfolio: merchandise, licensing, course IP = revenue × IP_multiple (vertical-specific)
Sponsorship Rate Prediction — Regression on 100K+ Disclosed Deals

Creator Index maintains a database of 100,000+ disclosed brand deal rates, scraped from FTC disclosure filings, creator media kits, and agency rate cards. A multivariate regression predicts fair-market sponsorship rates with R²=0.84:

Rate = β₀ + β₁×log(followers) + β₂×EQ_score + β₃×niche_CPM + β₄×demo_age_score + β₅×platform_weight + ε

This eliminates the 200–400% variance in brand deal pricing — the most expensive inefficiency in creator marketing.

Data Architecture

From platform APIs to actionable intelligence

Six platform data sources, five processing pipelines, three output surfaces — running on Supabase Edge Functions with Postgres vector search and scheduled batch jobs.

// Creator Index Data Pipeline — Full Stack INGESTION (per creator, per platform, near-real-time): ├── YouTube Data API v3 → subscribers, views, upload_frequency, EQ metrics ├── Instagram Graph API → followers, reach, impressions, story_metrics ├── TikTok Research API → followers, video_views, completion_rate, EQ metrics ├── Twitter/X API v2 → followers, impressions, engagement_rate, repost_rate ├── LinkedIn Creator API → followers, impressions, professional demographics └── Podcast (RSS + Chartable) → downloads, listener demographics, episode EQ │ └── Cross-validation: scraped public data vs. API data → bot detection flag bot_score = sigmoid(w₁×follow_ratio + w₂×eng_delta + w₃×growth_spike) PROCESSING (Supabase Edge Functions + scheduled jobs): ├── influence_graph_job (daily) → PageRank + eigenvector centrality update │ Spark-style: chunk 207M nodes into 1M batches, iterate to convergence ├── decay_fitting_job (per post) → fit E(t) = E₀·e^(-λt) on 96-hour engagement window │ Log-linear regression: log(E(t)) = log(E₀) − λt → OLS in O(n) per post ├── overlap_job (weekly) → MinHash signature update + LSH bucket rebuild │ 128 hash functions × 207M creators → 106GB signature matrix in Postgres ├── forecast_job (daily) → ARIMA re-fit + LSTM inference on 90-day horizon │ LSTM runs on GPU (A100, batched): 207M × 90-step inference in 4 hours/day └── valuation_job (weekly) → TCV recomputation with updated comp transactions OUTPUT: ├── Creator Profile Pages → public + authenticated (Creator tier) ├── Brand Dashboard → campaign planning, audience overlap, optimal creator sets ├── Creator Portal → personal analytics, brand deal rate card, growth forecast └── REST API → B2B integration (talent agencies, investment platforms)
SCALE CONSIDERATIONS

Engineering at 207M Creator Scale

  • PageRank requires sparse matrix multiplication on 207M × 207M adjacency — chunked batching with in-memory Spark-style reduce
  • MinHash signatures: 106GB stored in Postgres with HNSW vector index for nearest-neighbor queries
  • LSTM inference: batched GPU (NVIDIA A100), 207M creators in ~4 hours/day, ONNX export for efficient serving
  • ARIMA fits: distributed across 512 Supabase Edge Function workers — one ARIMA per creator per day
  • Cold-path: new creator indexed within 15 minutes of first API call; warm-path: cached profiles <50ms
DATA MOAT

Why This Is Hard to Replicate

  • API access accumulates proprietary historical data that cannot be backfilled — first-mover advantage compounds daily
  • Bot-inflation correction model trained on verified human accounts — proprietary training signal from creator partnerships
  • Brand deal rate database: 100K+ disclosed deals, curated and normalized — requires years to compile
  • Cross-platform audience matching: identity resolution without personally identifiable data — requires device graph or behavioral fingerprint partnership
  • Backtested forecast accuracy claims require 365+ days of historical data to validate — moat deepens with time
01
Search or Import a Creator
Brand types a creator handle or uploads a list. Creator Index cross-references all 6 platforms, resolves identity, and returns a unified profile within 15 minutes for new creators (cache hit: <200ms for known creators).
02
View Quantitative Profile
Unified Influence Score (0-100), engagement quality score, platform breakdown, audience demographics, bot-adjusted reach, 90-day growth forecast with confidence intervals, engagement decay curves, and predicted brand deal rate.
03
Build Optimal Campaign Portfolio
Brand inputs target demographics, budget, and content vertical. Creator Index runs the greedy set-cover algorithm to select the creator portfolio that maximizes unique reach at minimum CPM. Runs in <5 seconds over 207M creators.
04
Negotiate with Data
Brand enters negotiation with predicted fair-market rate range based on regression on 100K+ comparable deals. Creator enters negotiation knowing their quantified value. Creator Index eliminates the information asymmetry that currently costs creators and brands millions annually.

Pricing Model

Separate tiers for brands and creators

Two distinct buyer personas with different willingness-to-pay. Brands pay for campaign intelligence. Creators pay for personal analytics and rate card justification.

Brand Plans

Starter
$99
/month
  • 50 creator searches/month
  • Influence score + EQ score
  • Basic audience overlap (2 creators)
  • 30-day growth forecast
  • 5 verticals
Start Trial
Growth
$299
/month
  • Unlimited creator searches
  • Full quantitative profile
  • Audience overlap (up to 20)
  • 90-day ARIMA-LSTM forecast
  • Campaign portfolio optimizer
  • Sponsorship rate predictor
  • API access — 1K calls/month
Start Trial
Enterprise
$699
/month
  • Everything in Growth
  • Creator valuation (TCV model)
  • Bulk creator list analysis
  • Custom vertical weighting
  • API — 20K calls/month
  • Slack/CRM integration
  • Dedicated success manager
Contact Sales
Creator Pro
$29
/month
  • Your personal analytics dashboard
  • Unified Influence Score
  • Engagement decay modeling
  • 90-day growth forecast
  • Brand deal rate card
  • Audience overlap with competitors
  • Optimal posting time calendar
Join Waitlist

Revenue Projections

High-ACV brand tier drives the model

Milestone Brand Accts Creator Accts Monthly Revenue
Month 1 — Launch1580$6,845
Month 350300$23,450
Month 6150900$71,700
Month 93502,500$169,250
Month 127005,000$340,500
ARR at Month 12——$4,086,000
Assumptions: blended brand ARPU $400/mo, creator ARPU $29/mo
UNIT ECONOMICS

Brand Tier Economics

Average brand ACV $4,800
Brand trial-to-paid conversion 32%
Brand churn rate (monthly) 2.8%
Brand LTV (28-month avg retention) $11,200
CAC — brand (content + SDR) $1,200
Brand LTV : CAC 9.3 : 1
Gross margin (SaaS, no GPU amortized) 78%

Build Plan

Phased build — data first, analytics second, monetization third

PHASE 01 — WEEKS 1-4

Data Foundation

  • API integrations: YouTube, Instagram, TikTok, Twitter/X
  • Ingestion pipeline: normalized schema, 1M creator seed dataset
  • Basic Supabase schema: creator profiles, time series, engagement records
  • MinHash signatures: batch compute for seed dataset
  • PageRank: initial influence score computation
PHASE 02 — WEEKS 5-10

Analytics Models

  • Engagement decay fitting: log-linear regression per post
  • ARIMA training: fit per-creator models on 90-day history
  • LSTM training: GPU batch training on seed dataset
  • Valuation model: comp transaction database build, regression training
  • Sponsorship rate predictor: regression on disclosed deal database
PHASE 03 — WEEKS 11-14

Product + Monetization

  • Brand dashboard: creator search, profile pages, campaign optimizer
  • Creator portal: personal analytics, forecast display, rate card
  • Stripe billing: brand and creator subscription tiers
  • REST API: B2B integration endpoints, API key management
  • Waitlist → closed beta → public launch
MVP BUDGET

14-Week Build Cost

Engineering (3 engineers × 14 weeks) $168,000
GPU compute (A100, LSTM training) $8,000
API access fees (6 platforms) $3,200/mo
Supabase + Vercel infrastructure $450/mo
Brand deal comp database acquisition $12,000
Total MVP (14 weeks) ~$235,000
Why This Is the Right Moment

The creator economy crossed $250B in 2025 with no systematic data infrastructure. Platform APIs are now mature enough to support large-scale ingestion. GPU compute for LSTM training is commodity. The combination of timing, technical maturity, and market size creates a narrow window for a first-mover data moat. This is the Bloomberg Terminal moment for the creator economy.

Strategic Acquirers

CAA, WME, Wasserman, Dentsu, Publicis, and every major holding company has an active creator marketing practice with no systematic data. Creator Index is a natural acquisition target at 8–12× ARR within 24 months of demonstrating product-market fit. Series A at $4M ARR is the primary exit scenario.

NEXI — CONFIDENTIAL  |  INTERNAL USE ONLY  |  NOT FOR DISTRIBUTION