TL;DR: Copilot is the fastest way to do anything in Unstuck Engine. It's not just Q&A — it can build ICPs, define Personas, launch Signals, assemble Audiences, research leads, and draft outreach, all in plain English.
What Copilot is
Open the app and Copilot is the first thing you see — a dedicated workspace at the root URL, with a composer that asks "What do you want to find?". Copilot owns its own sidebar entry at the top of the navigation, above Signals, Qualification, Leads, Accounts, Audiences, and Insights. That ordering is intentional: most of what an operator does in Unstuck Engine starts with telling Copilot what you want.
Copilot is context-aware on every other page too. The Qualification page has an Ask Copilot · Tune your ICP button in the top-right. Audiences has Ask Copilot · Build me an audience. Signals has Ask Copilot · Tune your signals. Open a single Lead and you'll see Ask Copilot · Tell me about this lead. The same Copilot, but the page tells it what's currently on screen so you don't have to.
What you can ask it to do
Five capability clusters cover most of the day-to-day:
- Build the targeting model. Describe your customer in plain English and Copilot generates a complete ICP — segments, parameter modes (Filter / Score / Exclude), per-parameter weights. Same for Personas inside each ICP. Try:
> build me an ICP for mid-market SaaS companies with a security-first culture. - Define signals and autoqualify. Tell Copilot which buying behaviors you care about and it configures the relevant signals — keyword tracking, job changes, demo views, ad views — and wires them into your qualification rules. Try:
> track when CISOs at Fortune 500 financial services post about SOC2 on LinkedIn. - Assemble audiences. Describe an audience as a sentence and Copilot turns it into a multi-condition filter, including engagement, recency, and source weighting. Try:
> show me VPs of marketing at Series B SaaS companies who replied to a demo invite in the last 14 days. - Research and write for a single lead. From a lead detail page, ask Copilot for a one-page brief, talking points for a discovery call, or a custom-tailored outreach email — grounded in the signals that flagged this lead. Try (from any lead):
> tell me about this lead. - Answer questions about your own data. Copilot reads your workspace state in real time. Ask
> how many warm leads do I have from the Mid-market SaaS ICP?or> which signal contributed most to last week's qualifications?and it answers from your own pipeline, not a generic knowledge base.
How it's grounded
Copilot's answers cite the entities they reference. When it tells you a lead is hot, it shows you which signals tied them in. When it suggests a Persona refinement, it cites the historical reply-rate data behind the recommendation. The citation surface is part of the message UI — a row of entity chips (lead names, company names, signal names) beneath any assistant turn that referenced specific records.
If Copilot can't ground an answer in your data, it says so. You won't see fabricated metrics or made-up account names. That's a load-bearing rule for the assistant — grounding is first-class, not a footnote.
Where Copilot does NOT belong
A few intentional gaps:
- Irreversible destination pushes. Copilot drafts and stages a push to HubSpot, Salesforce, or Apollo. Final confirmation is always a click from you. The handoff is in your control.
- Credit-significant runs. Bulk enrichment over thousands of records prompts a credit-cost preview before execution. You decide whether the cost is worth it; Copilot doesn't decide for you.
- Sensitive credentials. Copilot never asks for or stores credentials. API keys and integration secrets live in Settings, edited by you directly.