Why Lisa

Built by HR professionals. Tuned on thousands of real HR scenarios.

HR is one of the most nuanced domains in business. The same termination scenario has different legal requirements in California, New York, and Texas. A "performance conversation" at a startup in month six of a PIP is strategically different than one at a 5,000-person enterprise. "My employee is struggling" could mean burnout, a personal crisis, an accommodation request, or something that needs immediate HR escalation — and the wrong response can cause irreversible harm.

Generic LLMs trained on the internet produce generic HR advice: cover your bases, consult your HR team, document everything. That's not coaching. Lisa's system is purpose-built around a discovery-first methodology — she asks the right questions before she gives any guidance, because the right guidance depends entirely on the specifics.

Five specialists. Two with live data today. Three more coming.

When a manager asks about a compensation dispute, Lisa activates her Compensation Specialist, which queries the BLS Occupational Employment and Wage Statistics API in real time, searches live market sources (Glassdoor, Levels.fyi, LinkedIn Salary), cross-references the company's own salary bands, and pulls from anonymized peer data in the Lisa Network. The manager gets a specific, multi-source answer — not a guess from training data.

When the situation involves employment law, the Legal Specialist retrieves verified provisions from the EEOC, DOL, and the applicable state labor department, detects whether the manager is in a jurisdiction where they may face personal liability, and passes all of this structured context into Lisa's response before she replies. The manager never knows a consultation happened — they just get a better answer.

The L&D, Recruiting, and Benefits specialists are active today — providing expert guidance through deep domain prompts and domain-specific workflows. Live external data integrations (e.g., O*NET for Recruiting, ECI for L&D benchmarking) are on the roadmap and will bring these specialists to the same data-grounded depth as Legal and Compensation.

Jurisdiction-aware risk detection. Deterministic escalation gates. Hard stops.

When Lisa detects any of seven defined high-risk trigger categories — harassment allegations, ADA accommodation requests, FMLA issues, workplace safety concerns, retaliation claims, legal threats, or whistleblower situations — she stops advising and escalates. This is not a prompt instruction the system can override; it's deterministic logic that evaluates every message against a hardcoded trigger list before any response is generated.

Lisa also knows when she's approaching the edge of her competence. Her system is designed to acknowledge uncertainty and stop before making things worse — a design philosophy that runs counter to the incentive of most AI systems, which optimize for appearing helpful even when they shouldn't be.

Full audit trail. Searchable transcripts with risk metadata. Export-ready for legal proceedings.

Every Lisa conversation is logged, timestamped, and stored with structured metadata — including the risk level classification, which specialist sub-agents were consulted, and what escalation actions (if any) were triggered. The result is an automatic paper trail that replaces the scattered email threads, hallway conversations, and mental notes that currently constitute most companies' HR documentation.

HR administrators and legal teams can search transcripts by employee, manager, date range, topic, or risk level. Exports are available in role-gated JSON and CSV formats suitable for legal discovery requests. The audit trail is a byproduct of using Lisa — managers don't have to do anything extra to create it.

Rigorous evaluation infrastructure. Not just vibes — p-values.

Lisa runs under a 6-category expert evaluation rubric — scoring every response on Legal Accuracy, Strategic Thinking, Discovery Quality, Actionable Guidance, Context Awareness, and Professional Tone. The evaluation runs continuously in the background on sampled live conversations, so quality is measured in production — not just in lab conditions. Responses that fall below threshold are flagged for human review automatically, and performance drops surface immediately for fast diagnosis.

Admin dashboard. Escalation queue. Risk distribution trends. ROI without reading transcripts.

The admin dashboard gives HR leaders a real-time view of how Lisa is being used across the organization: which managers are engaging, which topics are generating the most activity, where risk is concentrating, and how AI quality scores are trending over time. The escalation queue surfaces all conversations where Lisa triggered a hard stop, with full context, so HR can follow up appropriately.

For CHROs making the case for continued investment, the dashboard provides the usage, engagement, and risk-reduction data needed to demonstrate ROI — without requiring anyone to read individual conversations or compromise manager privacy.

Remembers specific employees, prior advice, and how situations evolved. Gets better over time.

When a manager returns to discuss an employee they raised three weeks ago, Lisa recalls the prior conversation, the advice given, and any actions the manager mentioned taking. This enables longitudinal coaching — tracking whether a performance conversation led to improvement, whether a communication plan worked, whether the situation escalated or resolved. This is what transforms Lisa from a Q&A tool into a trusted advisor.

Lisa also builds an organizational knowledge base as she's used — learning the company's culture, the specific challenges in different departments, and the individual style of each manager. This context makes every subsequent conversation richer and more specific.

Tools library. Anonymized template sharing. Export to .doc and .pptx.

Every time Lisa helps a manager draft a PIP, write a difficult email, create a training plan, or build an Individual Development Plan, that artifact goes into a searchable tools library — available for that manager to reference later, and (with appropriate anonymization) shareable across the organization. Over time, the organization builds a library of real-world HR artifacts that reflect its actual culture, style, and patterns — not generic templates from the internet.

Exports to .doc and .pptx make it easy to take Lisa's work product into existing workflows — no friction, no reformatting, no copying and pasting. Institutional knowledge persists beyond the conversation and beyond the manager who generated it.

Multi-model router. Not every problem needs the largest, most expensive model.

Lisa's architecture routes each query to the most accurate, fastest, or cost-effective model available for that specific task. A quick tone question doesn't need GPT-4o. A nuanced termination analysis does. A document classification task might be best served by a smaller, faster model. By routing intelligently, Lisa maintains quality while managing costs — and because she's not tied to a single provider, she moves with the pace of AI development rather than waiting for one vendor's next release.

When a new frontier model ships that outperforms the current routing on a specific task, it can be swapped in — validated by Lisa's A/B testing infrastructure — without rebuilding the product. The architecture is designed to absorb model improvements automatically.