Skip to main content

How Keyplay Uses AI

We use LLMs to improve ICP models, research and enrich accounts, and generate account intelligence.

Lauren Hayes avatar
Written by Lauren Hayes
Updated over 2 weeks ago

AI Overview & Nutrition Facts

Keyplay uses LLMs to enhance its account intelligence platform, helping to deliver richer insights, clearer segmentation, and more advanced ICP modeling. Keyplay does not process PII, never sends sensitive customer data to LLMs, and maintains privacy and security best practices when working with AI.

Category

Details

AI Use Cases

ICP definition, account scoring, custom agents for account research & enrichment, data cleaning.

Data Types

Public web data (websites, job posts, social media), Keyplay-collected company-level signals.

Models Used

Production: OpenAI (latest GPT model families, embeddings).

Experimental/Roadmap: Gemini, Claude.

Privacy

No customer 1P data is used or shared.

Keyplay opts out of OpenAI model training.Only public company data is processed.

Processing

API calls to OpenAI are server-to-server; no client IPs or personal data transmitted.

Update Cadence

Public data refreshed weekly–quarterly depending on type.

Agents schedules are set by users.

AI models and orchestration updated continuously with product development.

Limitations

Outputs may contain inaccuracies or reflect bias in public data; coverage varies by company footprint.

AI Capabilities

AI Agents for Account Research & Enrichment

  • Customers can define prompts and customize agents for a wide range of account intelligence use cases.

  • Agents crawl and analyze public web data to generate structured and unstructured account enrichment. More info here.

  • Agents are orchestrated by Keyplay; customer 1P data is never used or accessed.

Lookalike Model & Similarity Scores

  • Uses AI-powered vector analysis to find accounts with business models similar to your best-fit customers.

  • Scores accounts based on similarity to help you prioritize and build targeted plays around your ICP.

Industry Classification

  • AI-driven industry categorization provides more nuanced and reliable classifications than traditional frameworks like NAICS or self-reported data.

  • Supports primary and secondary categories for precise filtering and segmentation. For example, AI is a sub-category under SaaS & Cloud.

Account Research

  • AI analyzes large amounts of public data to produce detailed company descriptions, including:

    • Business model and operating activities

    • Market positioning

    • History and current status

    • Personality and culture

Data Quality

  • Uses AI to clean, validate, and enrich data for accuracy and consistency.

AI Risk Assessment

  • Keyplay never trains models on customer data.

  • All LLM inputs are limited to publicly available company information.

  • API calls to OpenAI are server-to-server; no client data or IP addresses are transmitted.

  • Keyplay opts out of OpenAI model training.

Risk Category

Details

Mitigation Strategies

Data & Privacy

No customer 1P data used or shared. External AI only processes public company information. API calls are server-to-server; no personal data shared.

Enforced opt-out for OpenAI model training; strict architectural separation; privacy-centric design.

Hallucination / Accuracy

Agents may generate inaccurate or speculative information, especially on ambiguous queries.

Human QA, preview on subset of accounts, customer feedback loop, backtesting against known lists. (docs.keyplay.io)

Bias in Public Data

Outputs could reflect bias or gaps present in scraped public sources.

Multi-source collection, bias awareness in model design, continuous iteration. (docs.keyplay.io)

Coverage Gaps

Companies with sparse public footprints may yield limited insights.

Signal layering, similarity scoring, enrichment cadence, fallback to higher-level signals. (docs.keyplay.io)

Compliance & Security

Use of external models may raise governance concerns in regulated industries.

SOC 2 controls (audit process), internal output monitoring, no PII usage, security-first customer data handling.

Operational Risk

System downtime or model drift could impact scoring and enrichment reliability.

Continuous monitoring, versioning, and rollback capabilities in pipeline orchestration.

Did this answer your question?