Talent Research

---
name: metaview-talent-research
description: >
  Commission talent-market research from Metaview's research agent and turn it into decisions.
  Trigger on "map the talent market for X", "where should we open our next engineering hub",
  "how big is the pool of Y in Z", "salary benchmarks for this role in these cities", "which
  companies should we hire from", "who's laying off in our space", "build a target-company
  list", "what does the market look like for this search" — or when a hiring plan, location
  decision, or difficult role needs outside-in market evidence. Runs Metaview research mode
  (async), optionally enriches it with what your own interviews reveal (comp expectations,
  competitor mentions, decline reasons), and delivers a decision-ready brief.
---

# Metaview Talent Research

Every hard hiring decision has a market question under it: is this role fillable here, at this comp, from these companies? This skill commissions that research from Metaview's research agent — talent pools, feeder companies, comp benchmarks, market events — and can cross-check the outside-in view against what candidates are actually telling you in your own interviews.

**Who it's for:** TA leads planning hiring, sourcers scoping hard roles, leaders making location/comp decisions, agencies doing client and market intelligence.
**Metaview products needed:** Sourcing (research mode). The inside-view enrichment additionally uses the AI Notetaker + Reports.

## Why this is powerful

Market research you'd normally buy or skip — pool sizes, comp ranges, who's shedding talent — comes back in under half an hour, commissioned with a decision-grade brief instead of a vague question. And it gets compared against your own interviews: what candidates actually ask for, where they actually come from, who they're actually talking to. The output isn't a PDF to file; it's "add these four companies to the search and raise the band 10%".

## How research mode behaves (set expectations)

- `send_sourcing_message` with `mode="research"` and no `search_id` starts a research thread. It's a general research assistant: web research, synthesis, data analysis, charts, reports — no candidate profiles, no outreach (that's source mode / metaview-sourcing-copilot).
- It is **asynchronous**: quick questions take a couple of minutes; a full structured market report typically takes **5-30 minutes**. Tell the user this upfront, and offer to check back rather than blocking.
- Poll `get_sourcing_messages` for the thread. Two quirks to expect: `phase` can read `idle` while the report is still being assembled, and thread messages can materialize with a lag — the reliable done-signal is the report itself appearing in the thread. If nothing has appeared after ~20-30 minutes, send a short status-check message; it gets a fast reply.
- Finished reports often arrive as a **file attachment** (e.g. an HTML report) viewable in the Metaview app, not as chat text. Include in your brief: "alongside the report file, post the key findings as a plain message in this thread" — so you can relay them without leaving Claude.
- Research threads persist: `list_sourcing_searches` shows them (`mode: "research"`). Reuse a thread for follow-ups on the same topic — it keeps context.

## Operating rules

- Tool names may carry a connector prefix; match on the trailing name.
- Market numbers (pool sizes, salaries) are estimates. Keep the research agent's confidence framing when relaying — and label anything you compute yourself as Observed (from your data) / Reported (from the research) / Inferred (your synthesis).
- Don't re-run research that exists: check `list_sourcing_searches` for a recent thread on the same market first.

## Workflow

### 1. Frame the question before commissioning

`get_user_context` once per session. Then sharpen the ask — research quality tracks brief quality. Establish: the decision this informs (hire here vs there? adjust comp? build target list?), role + seniority, geography(ies), and what "done" looks like (comparison table? shortlist of companies? report for leadership?).

### 2. Commission with a structured brief

`send_sourcing_message(mode="research")` with a brief that names the deliverable explicitly. Strong pattern:

> Research [role] talent in [geo(s)]. I want: (1) approximate pool size and how it's trending, (2) the 5-10 employers with the largest concentrations, (3) typical comp ranges (base, and total where meaningful), (4) companies likely to be good sources right now — layoffs, post-acquisition, slowed growth — with the evidence, (5) [decision-specific question]. Present as a structured report with a summary table. Where numbers are estimates, say so and give the basis.

Proven research shapes to offer users:

- **Talent map** — where a role's talent concentrates, by city/company, with pool sizes.
- **Target/feeder company list** — who to hire from (and who to avoid), with reasoning per company; feeds straight into a sourcing search ICP.
- **Comp benchmark** — ranges by geo/seniority for a role; ask for the basis of each estimate.
- **Location strategy** — comparing 2-4 cities for a new team: pool, comp, competition for talent, notable employers.
- **Market events monitor** — layoffs, acquisitions, funding news creating hiring windows in a defined space.
- **Company deep-dive** — one company's org, hiring, attrition signals (agencies: pitch prep; in-house: competitive intel).

While it runs, tell the user it's underway and either check back on a cadence or continue other work.

### 3. Enrich with the inside view from your own interviews

Your interviews are a live market dataset. Where relevant, run the inside-view queries in parallel with the commissioned research (using the discipline from metaview-interview-intelligence — reuse existing AI fields where they exist):

- **Comp reality check**: candidates' stated comp expectations from recent screens for this role family vs the research benchmark. One currency at a time.
- **Where candidates actually come from**: current-company mentions across recent interviews for the role — your observed feeder list vs the researched one.
- **Why people are moving**: reasons-for-leaving and pull factors from screens — sharpens outreach messaging for the target list.
- **Who else they're talking to**: competitor processes mentioned in interviews — the live competitive set.

If the workspace has little relevant interview data (new customer, new role family), skip this quietly — don't pad.

### 4. Synthesise and deliver

Don't just forward the agent's report. Produce the decision brief:

- **Answer first**: the recommendation or read, in 2-3 sentences, tied to the decision from step 1.
- **The evidence**: the market findings (condensed, with the agent's confidence markers) side-by-side with your inside view where you built one — flag agreements and gaps ("research says €70-85K; your last 12 screens averaged €78K asked — consistent" / "research names X as top feeder; your interviews say X's people go to Y instead").
- **So-what actions**: e.g. "add these 4 companies to the search ICP", "comp band needs +10% for this geo", "hub decision favours city B on pool-to-competition ratio".
- Format: chat brief by default; artifact/one-pager or deck on request; for agencies, a client-shareable version with internal notes stripped.

### 5. Close the loop

Offer the natural next step: feed the target list into a sourcing search (metaview-sourcing-copilot), set a recurring re-run for market monitoring (scheduled task, if supported), or save the brief as a document.

## Edge cases

- **User asks a research-shaped question that their own data answers better** ("what comp are candidates asking for?") — that's an interview-intelligence question; answer from interviews and offer the external benchmark as the add-on, not the other way round.
- **Tiny/niche markets**: pool estimates get unreliable; ask the research agent to say how it derived numbers and present ranges, not points.
- **The user needs candidates, not research**: hand to metaview-sourcing-copilot — research mode never returns contactable profiles.

## Example invocations

- "Map the senior product designer market in Amsterdam — pool, top employers, comp, and which companies to approach right now."
- "We're choosing between Warsaw and Lisbon for a 15-person platform team. Build the comparison."
- "Build a target-company list for the Head of Finance search, then check who our candidates actually come from."
- "Are we paying under market for mid-level AEs in London? Compare the benchmark with what candidates have told us."