🤖 AI Insights & Limitations
GigAnalytics uses Anthropic Claude (via Vercel AI Gateway) to generate income insights. This page explains exactly what the AI does, what it doesn't do, and how to interpret its output responsibly.
Which AI Model?
| Use case | Model | Notes |
|---|---|---|
| Weekly summaries | claude-haiku-4-5 | Fast, cost-effective narrative summaries |
| Price suggestions | claude-haiku-4-5 | Structured output (price + rationale) |
| Schedule suggestions | claude-haiku-4-5 | Best hours/days recommendation |
| Fallback (no AI) | Deterministic rules | Used when AI is unavailable or data quality is low |
What the AI Can Do ✅
Summarize your income trends in plain language
Given your last 30 days of net revenue, transaction counts, and hourly rates, it writes a 2–3 sentence summary of what happened.
Suggest a rate adjustment based on your data
It analyzes your current true hourly rate, your income target gap, and your historical volume to recommend a rate change with a specific rationale.
Recommend your best hours and days to work
Based on your revenue heatmap data, it identifies your top-performing time slots and explains the pattern in plain language.
Flag anomalies and data gaps
If your data shows unusual spikes, a long gap with no income, or a sudden rate drop, the AI will call it out explicitly.
What the AI Cannot Do ❌
Predict future income
All insights are backward-looking. The AI does not forecast what you will earn next month.
Account for market conditions outside your data
The AI only sees your transactions and time logs. It has no knowledge of platform algorithm changes, seasonal demand, or economic conditions.
Give financial or tax advice
GigAnalytics AI is not a financial advisor. Do not use its output for tax planning, investment decisions, or loan applications.
Learn from your feedback
The model is stateless. Dismissing or acting on an insight does not change future suggestions. Each generation is independent.
Benchmark against other users
AI insights are generated from your data only. Benchmark comparisons (percentile rankings) come from the separate, opt-in benchmark layer — not the AI.
Data Quality & Fallback Behavior
When your data is below these thresholds, GigAnalytics falls back to deterministic rule-based insights instead of calling the AI model. The insight card will show a "Based on rules (not AI)" label.
| Condition | Threshold | Fallback |
|---|---|---|
| Transaction count | < 5 transactions | Rule-based summary only |
| Time data | 0 billable hours | No hourly rate, no schedule suggestion |
| Date range | < 7 days of data | No weekly summary |
| Net revenue | $0 or negative | Alert insight only (no rate suggestion) |
| AI gateway error | Timeout / unavailable | Deterministic fallback for all insight types |
Confidence Levels
Each insight includes a confidence field (low / medium / high) based on sample size and data completeness:
| Level | Criteria | Meaning |
|---|---|---|
| 🟢 high | ≥ 20 transactions AND billable hours logged | Strong data foundation; insight is reliable |
| 🟡 medium | 5–19 transactions OR some hours missing | Reasonable estimate; treat as directional |
| 🔴 low | < 5 transactions OR no time data | Sparse data; insight may be misleading |
Privacy: What the AI Sees
The AI receives only aggregated statistics from your data — totals, averages, transaction counts, percentile ranks, and date ranges. It does not receive raw transaction descriptions, client names, or individual line items. See the Privacy Policy for full details.