Egret AI System — Search Quality Input
Egret AI System — Recruiter Input Guide: what this is, why your input matters, and how to fill out the form so the AI learns from our best work — not generic recruiting logic. Tips and examples appear under the most important questions below.
What we’re building and why your knowledge is the key ingredient
We’re building an AI assistant that searches our 121,000-contact database and surfaces the best candidates for any new search. Think of it as a very fast junior researcher who never forgets a contact — but only as good as what it has been taught about what good looks like at Egret.
The problem: no off-the-shelf AI understands the difference between a manufacturer-side candidate and a distributor-side candidate. It doesn’t know why an LC credential matters for a lighting controls role, or why someone from a large corporate manufacturer might be wrong for a scrappy startup client. Only we know that — and we need to teach it.
Your past placements and strong presentations are the training data. The form captures your recruiting instincts in a structured way so the AI can learn from them. There is no shortcut here — this is the most valuable thing you can do for the system.
- 5–10 searches to document per recruiter
- 2–4 candidates documented per search
- ~30 min estimated time per search entry
- 1 day total investment from the whole team
Which searches to document
Pick searches where you remember the candidates clearly. These work best:
Placements that stuck past 6 months
These are your highest-quality examples. You know the candidate was right, the client was happy, and the placement held. Document everything you remember about why it worked.
Strong presentations that went to interview
Even if the search didn’t close, candidates who made it to interview represent real quality judgment. What made you confident enough to present them?
Candidates who looked good on paper but were wrong
These are just as valuable as the placements. The AI needs to learn what not to surface, not just what to surface. If you remember a candidate who had the right resume but failed for a non-obvious reason, document that too.
Aim for variety across market segments — don’t only document lighting searches if you also work distribution or automation. Each recruiter’s specialty is irreplaceable input for the segments you cover.
How to fill out the form — step by step
- Pick your searches before you open the form. Spend 5 minutes writing down 5-10 searches you remember clearly before you start. Placements are ideal, strong presentations work too. Having them in mind first makes the form much faster to fill out. The very first question asks how you would describe the search in one or two sentences — having your answer ready before you open the form saves time and produces a better answer.
- Pull up Crelate for each search as you go. Having the candidate record open while you fill out the form helps. You don’t need to copy fields verbatim — you’re telling the story of why they were right, not transcribing their resume.
- Answer the why questions in plain language. Write the way you’d explain it to a colleague, not the way you’d write it in a presentation. Specific and conversational beats formal and vague every time. If you find yourself writing “strong background” or “excellent communicator” without backing it up with specifics, add the detail.
- Submit once per search, then submit another for the next one. The form is designed for one search per submission. When you finish a search entry, it will ask if you have another. This keeps responses clean and easier to process.
- Aim for 5-10 submissions; prioritize quality over speed. Five complete, specific entries are worth more than ten rushed ones. If you only have time for three really good ones this week, submit three and come back to the rest.
How many searches and candidates do you need?
Across the whole team, we need 20-30 search entries with 2-4 candidates each. That gives the AI enough examples to learn meaningful patterns. Here is what we are aiming for:
| Category | Minimum | Ideal |
|---|---|---|
| Total searches documented | 20 | 30 |
| Placed candidates | 15 | 20 |
| Presented but not placed | 5 | 8 |
| Strong but wrong candidates | 3 | 5 |
| Market segments covered | 4 | 6 |
| Functions covered (sales, eng, ops…) | 3 | 5 |
| Geographies covered | 4 | 6 |
Common questions
- What if I don’t remember all the details?
- Write what you remember. Partial information is still useful. The “why were they a strong fit” answer from memory is often richer than pulling it from notes because you are capturing what actually mattered, not what got written down at the time.
- Do I need to look up Crelate IDs for each candidate?
- No — just use the candidate’s full name. We will match names to database records on the back end. Crelate IDs are helpful if you have them but not required.
- What if a search had only one strong candidate?
- That’s fine — document that one candidate fully in the Candidate 1 section and leave Candidates 2 and 3 blank. One complete entry is better than three thin ones.
- Should I document searches that didn’t result in a placement?
- Yes, if you got to the presentation or interview stage. Candidates who were presented but not placed — and the reasons why — are valuable training data. Document what you know.
- How long will this take?
- About 20-30 minutes per search once you are in the rhythm. The first entry will take longer as you get familiar with the form. Budget 2-3 hours total for 5-8 searches, which can be done across a few sessions.
- What happens with the information I submit?
- Responses go into a secure internal system used only to train the Egret AI. Nothing is shared externally. Candidate and client information is used solely for improving our internal search quality.
Thank you for taking the time to do this right.
Every answer you put into this form directly shapes what the AI learns about what good looks like at Egret. The more specific and honest your answers, the better the system will perform — and the more time it will save you finding the right candidates in future searches.

