Privacy & Benchmarking Explainer

1. Privacy-First Architecture

GigAnalytics is built on a data minimization principle: we collect only what's necessary to compute your ROI dashboard, and we store it in a user-partitioned database where your data is isolated by Row Level Security (RLS) from every other user's data.

2. What Data GigAnalytics Stores

Here is a complete inventory of data stored in your GigAnalytics account:

3. Benchmark Feature: Full Data Flow

The benchmark feature allows Pro users to opt in to contributing anonymized aggregate metrics that power the "how do I compare?" insights in the dashboard. Here is the complete data flow, step by step:

1. 1

   You enable "Contribute to benchmarks" in Settings → Privacy

   This toggle is OFF by default. Enabling it starts the contribution pipeline for your account. You can disable it at any time.

2. 2

   Our backend computes aggregate metrics from your raw data

   The pipeline runs nightly. It reads your transactions and time entries and computes: hourly rate percentile bucket, revenue range bucket, experience range bucket, platform category, and region (country-level). It does NOT read transaction descriptions, client names, or exact amounts.

3. 3

   Buckets are assigned (not raw values)

   Raw hourly rate of $87/hr is bucketed to "$80–$100/hr". Revenue of $4,200/mo is bucketed to "$4,000–$5,000/mo". This prevents precise individual inference.

4. 4

   K-anonymity check

   Before your bucket is included in any aggregate, the pipeline checks: does this (platform, rate_bucket, region) combination have ≥ 25 other contributors? If not, your data is suppressed for that segment until the pool grows.

5. 5

   Differential privacy noise is added

   Even for qualifying buckets, we apply Laplace noise (ε = 0.5) to the aggregate counts before storing them. This means the published percentile is statistically indistinguishable from what it would be if any single user were removed from the pool.

6. 6

   Contribution is decoupled from user ID

   The final aggregated values in the benchmark store are not linked to your user_id. The pipeline uses a one-way hash of (user_id + contribution_date) as a deduplication key that cannot be reversed.

7. 7

   Benchmark values are served to Pro users

   The published p25/median/p75 rates are served to all Pro users querying their platform + region segment. Your personal data never appears — only the anonymized aggregate.

4. K-Anonymity: The Math

K-anonymity is a formal privacy guarantee. A dataset satisfies k-anonymity if, for every record, at least k−1 other records share the same quasi-identifying attributes.

In GigAnalytics's benchmark dataset, the quasi-identifying attributes are:

• platform_category — bucketed (design, development, writing, consulting, other)
• hourly_rate_bucket — $10 increments up to $200, then $50 increments
• region — country-level (US, UK, CA, AU, DE, other)
• experience_range — <1yr, 1–3yr, 3–7yr, 7yr+

What this means in practice: If you're the only freelance copywriter in New Zealand earning $90–$100/hr with 3–7 years of experience, your data won't appear in the benchmark for that segment. You'll see "insufficient data" for that specific combination.

5. Differential Privacy: The Math

K-anonymity alone is vulnerable to attacks where an adversary knows about a specific individual. Differential privacy (DP) provides a stronger guarantee: the probability of any inference about an individual changes by at most e^ε whether or not that individual's data is in the dataset.

GigAnalytics applies the Laplace mechanism to aggregate counts and statistics before publication:

Why ε = 0.5? Lower ε = stronger privacy guarantee but noisier statistics. We chose ε = 0.5 as a balance: it satisfies "strong" DP by most academic standards (ε ≤ 1) while keeping the benchmark accuracy high enough to be useful (±$10 noise on a $60–$150 range is acceptable).

The combination of k=25 anonymity and ε=0.5 differential privacy provides defense-in-depth: k-anonymity protects against record linkage attacks; differential privacy protects against membership inference attacks.

6. Benchmark Contribution Pipeline Architecture

7. What You See vs. What You Share

8. Data Deletion and Opt-Out

9. Third-Party Data Processors

10. Security Measures