Competitor Intent Intelligence

Don't just detect what a competitor changed. Reconstruct what they may be trying to do — and test that hypothesis against contradictory evidence and later outcomes.

This Actor monitors ecommerce pages over time and turns raw changes into decision-ready competitor intelligence. It keeps persistent history, connects related commercial events inside a temporal window, compares them with the store's own behavior, produces an intent hypothesis with supporting evidence, counter-evidence, alternatives, confidence, and recommended action, and can track whether earlier hypotheses were later confirmed by observed follow-on events.

The EDGE: intent reconstruction with falsification and calibration

Most ecommerce monitors stop at alerts such as “price dropped”, “stock changed”, or “new products appeared”. This Actor is designed to go one level higher:

Observed changes → temporal evidence → intent hypothesis → supporting evidence → counter-evidence → alternative hypotheses → confidence → later outcome validation → observed precision → decision brief.

For example, a price drop combined with catalog expansion and a new merchandising signal may support a promotion or launch hypothesis. A simultaneous restock pattern can weaken a clearance hypothesis. The Actor keeps observed facts separate from inferred intent and does not present correlation as proven causation.

When a hypothesis is created, the Actor can persist that prediction, evaluate later events after the configured evidence window, and build a calibration profile showing confirmed vs. not-confirmed outcomes and observed precision by hypothesis type. Small samples are explicitly marked as early evidence rather than overclaimed.

Best for

• Competitor strategy and promotion monitoring
• Ecommerce competitive intelligence
• Pricing and assortment teams
• Agencies monitoring client competitors
• Merchandising and category intelligence
• Feeding decision-ready signals to agents, BI, Make, Zapier, n8n, or custom workflows

What it detects and reconstructs

The monitoring layer detects price, availability, product identity, sale/discount, shipping, merchandising, and broader content changes. The intelligence layer can add:

• commercialEvent — structured interpretation of an observed commercial change
• intentHypothesis — the likely commercial intent reconstructed from related events
• supporting evidence and counter-evidence
• alternative hypotheses
• confidence and confidence trend
• historical deviation / strategy fingerprint
• causalIntelligence — competing hypotheses, falsifiers, anomaly scoring, evidence quality, and decision support
• prediction — a persisted current hypothesis awaiting later outcome validation
• calibrationProfile — confirmed/not-confirmed outcomes and observed precision overall and by hypothesis type
• opportunity — potential response opportunity
• decisionBrief — concise decision support and recommended action

These fields are evidence-driven. When evidence is insufficient, the Actor should preserve uncertainty rather than invent a confident narrative.

How it works

1. Add one or more product, category, collection, or landing-page URLs.
2. The first successful check stores a persistent baseline.
3. Later runs detect meaningful commercial changes.
4. Event history is retained per monitor/URL.
5. Related events are clustered across a temporal window.
6. The intelligence engine proposes competing intent hypotheses and actively checks for contradictory evidence, alternatives, and falsifiers.
7. A leading hypothesis can be persisted as a prediction.
8. Later runs validate expired predictions against observed follow-on events and update a calibration profile.
9. Structured results are written to the dataset for human review or downstream automation.

Input example

{
  "urls": ["https://www.example.com/product/example"],
  "monitorId": "competitor-a",
  "compareMode": "commerce",
  "outputMode": "all",
  "changeThreshold": 1,
  "maxPages": 10
}

Output example

{
  "status": "changed",
  "changeType": "commercial_change",
  "commercialEvent": {"eventType": "price_cut"},
  "leadingHypothesis": "seasonal_or_campaign_promotion",
  "evidenceQuality": 68,
  "prediction": {
    "hypothesis": "seasonal_or_campaign_promotion",
    "status": "pending",
    "horizonHours": 168
  },
  "calibrationProfile": {
    "resolvedPredictions": 6,
    "pendingPredictions": 1,
    "observedPrecision": 0.667,
    "byHypothesis": {
      "seasonal_or_campaign_promotion": {
        "confirmed": 3,
        "notConfirmed": 1,
        "total": 4,
        "observedPrecision": 0.75
      }
    }
  }
}

The exact fields present depend on the evidence available for that run. A calibration profile is observational and historical; it does not convert an intent hypothesis into proof of causation.

HTTP-first economics

The Actor is designed to stay lightweight and HTTP-first. It uses public page/structured commerce signals and persistent Apify storage rather than requiring a browser, proxy, or paid external intelligence API by default. This keeps monitoring suitable for repeated scheduled runs and leaves room for sustainable Pay-per-event economics.

Recommended workflow

Save the input as an Apify Task and schedule repeated runs. Use outputMode: changesOnly when downstream systems should receive only baselines, meaningful changes, calibration resolutions, intelligence events, and errors. Send high-confidence decision briefs to dashboards or workflow automation; retain lower-confidence hypotheses for review instead of automatically acting on them.

Pricing

The Actor uses Apify Pay per event pricing. The current price displayed on the Apify Store is authoritative. Pricing should be evaluated against the value of intelligence events and actual platform run costs rather than raw scraped rows.

Verified behavior

The monitoring core has been tested across separate Actor runs: an initial run creates a baseline and later runs retrieve the previous snapshot and compare new observations against it. The intent reconstruction, falsification, prediction persistence primitives, later-outcome validation, and precision calculation are covered by automated tests. Store-publish verification also checks that the deployed Actor returns the calibration profile on live runs.

Limitations

Intent is an inference, not a fact. The Actor explicitly separates observed evidence from hypotheses and can expose counter-evidence and alternatives. Calibration is based only on observed events that the monitor could actually see, and small samples should not be treated as statistically stable. Some sites block automated requests or render data only after JavaScript execution; those pages may return limited signals or an error in the lightweight HTTP path. Failures are reported per URL so one blocked page does not stop the rest of a run.

Apify workflow

Use the Actor with Apify Tasks, Schedules, API calls, datasets, webhooks/integrations, and downstream automation. Persistent state is what allows repeated observations to become temporal intelligence instead of isolated scraping results.