How a Venture Studio Cut HR Replies From 3 Days to 2 Hours

One small HR team answered policy questions for every company in a venture studio. An AI engine now answers from each handbook, cutting resolution from 3 days to under 2 hours.
3 days → <2 hrs
HR ticket resolution time
< 4 weeks
Implementation Time
Project Cost

the challenge

A venture studio's portfolio companies shared one small HR function. That team fielded policy questions from employees across every company, each with its own handbook and policies. Answering a ticket took about three days, and the queue grew with every new portfolio company the studio added.

what they built

Portfolio Leverage built AIHRPilot, an HR policy engine that answers employee questions from each portfolio company's own handbook and policies. Each question is matched to the most relevant policy passages, and Claude drafts an answer with a category, a confidence score and a citation to the source section. High-confidence answers go straight back to the employee. Mid-confidence answers go to an HR partner as a ready-to-edit draft, and low-confidence questions or sensitive topics such as discrimination, harassment or legal liability go to a person with no draft. The engine sits on top of the studio's existing HR system instead of replacing it.
The build started with the policy corpus. Handbook PDFs were split into roughly 500-word passages with overlap, and each passage was tagged with its source document, section heading and effective date so every answer could cite its policy. Retrieval uses TF-IDF, since the corpus is small enough that a vector database added nothing, and returns the top three to five passages per question. Claude Sonnet reads those passages and returns a category, a confidence score, a draft reply and a citation. Most of the effort went into the system prompt, covering the category taxonomy, confidence rules, reply style and refusal triggers. A three-tier router then decides what happens: auto-send, a draft for an HR partner, or a human-only flag. The interface is an inbox color-coded by confidence, plus an admin page for uploading policies and adjusting thresholds. A first version took a weekend, and production quality took about three more weeks. It went live within weeks of kickoff.

best fit for

Venture studios, holding companies and PE platforms where a small shared HR team supports several companies with different policies, and growing companies of roughly 50 to 500 employees where HR spends its time answering the same policy questions.
Ai ROLE
The AI retrieves the relevant policy passages, drafts an answer with a category, confidence score and citation, and either sends it, hands it to HR as a draft, or flags it for a person based on confidence and refusal rules.

impact

3 days → <2 hrs

HR ticket resolution time across the studio's portfolio companies.

30 days

Time for the system to reach positive ROI.

Commercialized

The engine became a commercial product after the studio deployment.

implementation complexity

Low to medium: a policy corpus pipeline, TF-IDF retrieval, a Claude reasoning layer, three-tier confidence routing and a simple review inbox, built in about four weeks.
Yuri Kruman, Founder, Portfolio Leverage

Yuri Kruman

Founder at Portfolio Leverage | Fractional AI Operating Partner
Portfolio Leverage Company
Founder of Portfolio Leverage, a fractional AI operating partner for PE-backed and mid-market companies. Three-time CHRO with a JD who builds the AI systems himself.
Get an intro
Talk to this team
industry
Financial Services
business organization
HR & People
AI TYpe
Knowledge Management & Search (RAG)
Conversational AI (Chatbot / Agent)
value type
Time Savings
Cost Reduction

frequently asked questions

How did a venture studio cut HR ticket resolution from 3 days to under 2 hours with AI?

It put an AI policy engine in front of its shared HR team. The engine answers employee questions from each portfolio company's own handbook with a citation, sends confident answers directly and hands the rest to HR as drafts, which cut resolution from about three days to under two hours.

What AI was used for the HR policy assistant?

Claude Sonnet via the Claude API, with TF-IDF retrieval over the policy documents, a Python and Flask backend, and confidence-based routing to HR.

What results did the venture studio see?

HR ticket resolution dropped from about three days to under two hours, ROI turned positive within 30 days, and the system went on to become a commercial product.

How long did the AI HR assistant take to build?

A first version took a weekend and production quality about three more weeks, and it went live within weeks of kickoff.

Who is an AI HR policy assistant best for?

Venture studios, holding companies and PE platforms where one small HR team supports several companies with different policies, and growing companies where HR answers the same policy questions all day.

Have a similar challenge?

Ask whether this would work for you, or describe what you're trying to solve.
TELL US WHAT YOU'RE EXPLORING