How ONX uses artificial intelligence, what models are involved, and what human oversight is in place.
Last reviewed: May 2026 · Contact compliance@optimalnexus.com
EU AI Act — Risk Classification
Most ONX features are commercial decision-support tools outside the EU AI Act's high-risk categories (Annex III). One exception: the Hiring module's candidate–job match scoring supports recruitment and falls under the EU AI Act's employment-AI category (Annex III 4(a)) — it is therefore operated under the Act's strictest control framework: mandatory human review before a score is used, per-decision audit logging with model version, stated reasoning (strengths and gaps) alongside every score, and ongoing accuracy monitoring. Candidates are informed when AI assists the evaluation of their application. This assessment is reviewed annually and when processing materially changes.
ICP Profile Generation
Analyses your product, market, and win/loss signals to surface ideal customer profile recommendations.
Output: Recommendations only — customer reviews and confirms
Company Discovery Scoring
Ranks discovered companies against your ICP criteria using firmographic and signal data.
Output: Priority ranking — sales team decides which to pursue
Deal Risk & Stall Detection
Identifies patterns in deal room activity indicating pipeline risk or buyer disengagement.
Output: Risk indicators and nudge suggestions — rep decides action
Asset & Outreach Suggestions
Recommends relevant sales assets or messaging based on deal stage and buyer signals.
Output: Suggestions only — rep selects and sends
Revenue Intelligence Summaries
Generates natural-language summaries of pipeline health, forecast, and team activity.
Output: Informational summaries — no automated forecast adjustments
Outreach Email Drafting
Drafts personalised outreach copy from a contact’s role, company context, and signals.
Output: Draft only — a person reviews, edits, and sends
Resume Parsing (Hiring)
Extracts skills, titles, and history from uploaded resumes with per-field confidence; sensitive identifiers are detected and redacted before processing.
Output: Structured profile — low-confidence extractions require human review before use
Candidate–Job Match Scoring (Hiring)
Scores a candidate’s fit against a job description, with stated strengths and gaps. This supports recruitment and falls under the EU AI Act’s employment-AI category (Annex III 4(a)), so it runs under the Act’s strictest control framework.
Output: Score with reasoning — a recruiter must review and accept it before it influences the pipeline; every run is decision-logged
Anthropic — Claude (Sonnet / Haiku)
Privacy policy ↗Purpose
ICP generation, signal analysis, deal recommendations, asset suggestions, outreach drafting, resume parsing and candidate matching (Hiring)
Data shared
Account context, company data, pipeline signals. Outreach drafting includes contact first name and job title; Hiring features include resume text (sensitive identifiers redacted first). Never used for model training.
Region / Transfer
United States (SCC covered)
Perplexity AI — Sonar (live web research)
Privacy policy ↗Purpose
Real-time market and macro-signal research
Data shared
Company names, domains, industry terms — no personal data
Region / Transfer
United States (SCC covered)
OpenAI — GPT-4o (optional enrichment flows)
Privacy policy ↗Purpose
Supplementary content generation where enabled
Data shared
Company-level context only
Region / Transfer
United States (SCC covered)
Human in the loop
Every AI output requires a human to review and act. No commercial action is automated without rep approval.
Explainability
AI recommendations surface the signals and reasoning behind them so reps can evaluate quality.
Confidence limitations
AI outputs are surfaced as indicators and suggestions, not deterministic facts. Outputs should be verified.
Minimal personal data in model prompts
Prompts are company-level wherever possible. Where a feature needs individual data (outreach drafting: first name and job title; Hiring: resume text), only what the feature requires is sent, sensitive identifiers are redacted, and the use is disclosed above.
No model training on your data
Customer data is never used to train or fine-tune AI models by ONX or its AI providers.
Bias monitoring
ONX monitors AI outputs for systematic bias in scoring or recommendation patterns. Issues are reviewed and corrected.
ONX AI must not be used to make or inform the following types of decisions:
Human Oversight Statement
All commercial actions within ONX remain under customer control. ONX provides decision-support insights, workflow recommendations, and pipeline signals only. No output from ONX AI constitutes a final decision. Users are responsible for evaluating AI recommendations before acting on them.