Most tools tell you a CEO bought shares and call it a signal. We go further — because the difference between a real conviction signal and a misleading one matters enormously.
Every day, retail investors see "CEO buys $2M" and treat it as a buy signal. The problem: context is everything, and raw filings strip it away entirely.
When executives buy stock in the open market — not options or RSUs — they're putting personal capital at risk based on private knowledge of their own company. Research consistently shows open-market purchases outperform the market, but the edge concentrates in specific patterns:
A single insider buying a small amount after a stock has already rallied 40% tells you very little. Thirty insiders buying across a 30-day window at 52-week lows, with analysts raising targets — that's a very different conversation.
Insider buying
is not a crystal ball
Even the best insider signals fail regularly — insiders share the same macro uncertainty as everyone else. The goal isn't certainty; it's identifying high-quality probabilistic edges and weighting them correctly.
Not all insider buying is equal — the motivation changes how much weight it deserves.
The insider believes the stock is undervalued and is putting real capital behind that view — typically after a sharp drawdown or during broad pessimism. The signal most correlated with forward returns.
Boards buy to signal confidence after a volatile quarter or strategic pivot. Real, but partly performative — more meaningful when other insiders act independently.
Purchases to rebalance, avoid wash-sale rules, or execute a gifting plan. Real transactions, but made for mechanical reasons — forward-looking information content is low.
Plans set up in advance that execute automatically, regardless of current conditions. The decision reflects past conviction, not current sentiment — yet they appear in filings identically to discretionary buys.
Four scenarios where raw insider data most commonly produces false signals — and where our AI analysis is designed to flag them.
Insiders can anchor on a past price and mistake a structural decline for a bargain. A CEO who paid $40 buying at $18 isn't necessarily getting a deal.
$50K from a billionaire CEO is a rounding error; the same from a CFO on a $300K salary is a genuine commitment. Most tools don't normalise for this.
A lone insider buying against analyst downgrades is a contrarian bet, not a high-conviction read. Operational knowledge doesn't override macro headwinds or earnings revision cycles.
Post-announcement buys often reflect delayed enthusiasm, not forward-looking conviction. The information is already public — the edge is already priced in.
Our AI Signal Score automates several of these checks — but understanding the framework helps you interpret it correctly.
Most scoring systems use fixed weights and call it a day. Ours tracks what actually happens after every signal is issued — so the engine can be tuned against reality, not theory.
Every AI Signal issued today is matched against real price returns at 30, 60, and 90 days — and compared to the S&P 500 at each horizon. We record not just the return, but whether the signal beat the market. All six factor scores are stored alongside the outcome, building a ground-truth dataset from day one.
The optimisation pipeline is built. We're waiting on ~300 matured signal outcomes. The first 30-day returns land mid-September 2026; by November 2026 we expect enough records to retrain. At that point, factor weights are updated against real return data: factors that consistently predicted gains earn more weight, weaker ones get trimmed. Every change is shown in full — no black box.
The six factors and their current weights — insider activity (25%), analyst consensus (25%), price upside (20%), AI sentiment (20%), track record (15%), buying into weakness (10%) — reflect our best hypothesis today. Real return data will validate or refine them. This is honest about where we are, and intentional about where we're going.
Raw filing alerts are easy to generate. Turning them into genuine investment signals is hard — it requires weighting buying quality, insider credibility, macro context, and analyst consensus simultaneously. That's what our AI Signal Engine does.
We treat insider data as one of several inputs that, weighted correctly, produces a meaningful probabilistic edge. Every weight, every factor, every score is visible and explained — so you always know exactly why a signal is rated the way it is. And as real outcome data accumulates, those weights will be refined against reality, not just theory.
Click any factor to explore its contribution. Auto-cycles every 4 seconds.
We check EDGAR every 30 minutes and parse every new Form 4 automatically.
Claude analyses the full filing picture and returns a structured assessment.
Six factors merge into a single 1–5 score with a named signal tier.
Every ticker in the dashboard shows a live AI Signal Score with full factor transparency — exactly what you've been reading about, applied to thousands of active tickers.
We truly appreciate you taking the time to help us! Every bug reported and every suggestion shared helps us on our mission to make the Insider Intel platform faster, smarter and more valuable for everyone.