All Shipped by HR

Shipped by HR

Building OILSmith to transform performance & feedback at Juro

Thomas Forstner · Juro

In this episode Adam is joined by Thomas Forstner, VP People & Talent at Juro who walks us through the problem they saw with performance management, and how they built OILSmith with Claude to underpin their new approach to performance & feedback which ultimately led to faster, more continuous performance conversations and a reduction in how long it too to identify and action underperformance too.

What they built

Thomas built OILSmith - a Claude project (AI assistant) that automates and improves Juro's bi-monthly performance review process, which they call "Oil Checks."

OILSmith handles five distinct jobs:

  • Feedback coaching - guides managers and employees to give structured feedback using the OILS framework (Observation, Impact, Listen, Suggestion)
  • Fact-finding / dossier prep - pulls visible performance data from connected tools to help employees self-prepare
  • Scheduling support - reduces friction around finding time during busy periods
  • Summarisation and upload - reduces admin after conversations
  • Cycle-level data analysis - surfaces aggregate talent density trends for leadership

OILSmith is organised around named "Skills" (modular instruction sets), including:

  • Oil Prep - helps the reviewee build their own performance dossier before the conversation
  • Oil Up - helps managers upload completed Oil Check outputs to the HR system

Why they built it

  • A six-monthly review cycle was too slow for a fast-moving AI-first company where priorities shift constantly
  • Previous attempts at continuous performance (monthly cadence) failed because the admin burden remained too high for managers
  • Wanted performance data to be predictive and actionable — correlating with real outcomes like promotions and performance improvement cases - not just a periodic HR exercise
  • Upcoming UK employment law changes (unfair dismissal rights from 6 months, effective Jan 2027) made a reliable, frequent, documented performance record a legal and operational priority

Tools used

  • Claude (Anthropic) - core AI model; OILSmith is built as a Claude Project with custom Skills
  • Claude Opus - used for richer data synthesis tasks
  • Granola - AI meeting note-taker; primary source of visible performance behaviour data
  • Slack - secondary data source for visible contributions
  • Google - connected via Claude connector
  • Linear - connected for product and engineering team performance data
  • HubSpot - connected for sales team quota performance data
  • Notion - stores the performance playbook and company principles, uploaded as files into OILSmith
  • Humaans - destination for completed Oil Check uploads

How you could build this yourself

  1. Define your feedback framework - pick or adapt a simple model (Thomas uses OILS). Make sure your CEO already knows and endorses it; that buy-in matters more than the framework itself.

  2. Choose your AI platform - Claude Projects, ChatGPT Custom GPTs, or Gemini Gems all work similarly. Pick whichever your team already uses.

  3. Write your project instructions with AI help - don't start from a blank page. Ask Claude or ChatGPT: "I'm building a performance review assistant. Help me write the instructions it needs." Iterate from there.

  4. Connect your data sources - identify where visible work actually lives (meeting notes, project tools, CRM, Slack). Connect only what the reviewee themselves can access - this avoids "big brother" concerns and gives a more complete picture than a manager's view alone.

  5. Upload your context documents - add your performance playbook and company values/principles as files so the assistant can assess against your actual dimensions (e.g. results, skills, culture).

  6. Build modular Skills - create separate instruction sets for each job (prep, upload, coaching). Ask the AI to help you write each skill prompt. Name them consistently so users know which to invoke.

  7. Flip ownership to the reviewee - have the employee build their own dossier using Oil Prep and share it with their manager. This distributes workload more evenly and avoids surveillance concerns.

  8. Set cadence at every two months, not monthly - monthly proved too frequent even with reduced admin. Every two months gives enough signal while keeping iteration cycles short.

  9. Get explicit exec sponsorship - have your CEO send a message making 100% completion a clear expectation with a named deadline. Thomas identified this as the single biggest driver of completion rates.

  10. Track three things in your leadership reporting - consistency (time to 100% completion), reliability (does overall score correlate with results score?), and actionability (does the data actually drive formal next steps like PIPs or promotion cases?).

Source

Built by Thomas Forstner, VP of People and Talent at Juro, and presented as part of an Open Org "Shipped by HR" session.

Related playbooks

Useful resources