Outbound prospecting is research work. Someone has to find the companies that fit, find the right people inside them, notice what changed this month, and write it down where sales will see it. Percy is the outbound prospecting agent in AgentNava's sales library. This post shows how Percy builds a prospect list, what it looks for, where it stops for your approval, and what it will never do.
Meet Percy, an AI agent for outbound prospecting
Percy is a starter agent: a ready-made agent whose instructions and workflows you can read before you start and change after. Its instructions open with the job:
“You keep the top of the funnel filled with warm, well-researched prospects that fit the team’s ideal customer profile, so sales spends its time on conversations.”
From the instructions Percy follows
The ideal customer profile (ICP) is your description of a good customer: industry, company size, the roles you sell to, geography, and the problem you solve. Percy treats it as the filter for everything it does.
Its four workflows are written procedures that it follows step by step:
- Build an ICP list: turn a campaign brief into a filtered, deduplicated list of companies and contacts.
- Enrich prospects: research each company and contact, score them, and write a short brief.
- Daily signal scan: check the active list each morning for news that makes now a good time.
- Disqualify and clean: move stale or poor-fit prospects out of the active list, with a reason.
What Percy connects to, and what each connection can do
A connection is an account you link to AgentNava so an agent can work in it. Percy’s instructions name three:
- Apollo: searches for companies and people by industry, size, geography, seniority and title, enriches records, and searches news articles about companies.
- HubSpot: your CRM. Percy checks it for existing customers, open deals and opted-out contacts, and writes enriched records and briefs back to it.
- LinkedIn: the connection works on your own LinkedIn account. It can read a person’s profile when Percy already has their LinkedIn ID, and it cannot search members or list someone’s recent posts.
All three are listed as Beta in AgentNava’s connections catalog, which means usable today with documented limitations.
That LinkedIn limit matters, because Percy’s instructions ask it to check each contact’s recent posts. In practice, Percy finds company news, funding, hiring and launches through Apollo’s news search and through web search, which every AgentNava agent has by default. If recent LinkedIn activity is central to how your team qualifies prospects, plan for a person to check it, and edit Percy’s instructions to say so.
Step 1: build a list that fits your ideal customer
The list workflow starts with a brief, and Percy will not start without one. From its instructions: “Never start a prospecting run without knowing the campaign goal, target segment, and what a good outcome looks like.”
- Collect the brief: industry, company size, geography, target titles or seniority, any must-have technology, and how many prospects you want.
- Check HubSpot for accounts to leave out: current customers, open deals, and contacts who asked not to be contacted.
- Search Apollo for companies with the industry, size and geography filters, then tighten one filter at a time if the result is too broad, explaining each change.
- Search for contacts inside those companies, one or two per company.
- Remove duplicates and the accounts from step 2, and note why each was removed.
- Show you the totals, the filters, and three to five examples, and wait for your approval.
- Once you approve, pass the list to enrichment.
The workflow’s own example:
“You get 180 matching companies and 210 contacts. You cross-reference HubSpot and remove 12 existing customers and 3 suppressed contacts. You surface the filtered set to the user: ‘Filtered to 168 companies, 195 contacts. Sample: [three company names with title examples]. Ready to proceed to enrichment?’”
Example run, from the instructions Percy follows. Numbers are illustrative.
Step 2: research each prospect and score it
Enrichment works through the approved list one record at a time. For each company, Percy looks up headcount, funding history, news from the last 90 days, technology signals and public statements about priorities. For each contact, it records title, tenure and earlier roles.
Then it scores each prospect on three questions, each rated high, medium or low:
| Score | The question |
|---|---|
| Fit | Do the company and the role match the ideal customer profile? |
| Signal | Is there a specific, recent event that makes now a good time to reach out? |
| Timing | Is there a trigger, such as funding, a hire, a launch or a new fiscal year, that makes urgency plausible? |
A prospect does not score high on signal for looking like a good fit. From Percy’s instructions: “A signal must be a specific, recent event.”
For each prospect, Percy writes a brief of two or three sentences: who they are, why they fit, the strongest signal, and a suggested opening angle. It adds the brief as a note in HubSpot, fills empty fields, and tags the record “Prospected”. The workflow’s example brief:
“RevOps Director at 200-person SaaS, Series B eight weeks ago. Posted three weeks ago about pipeline visibility pain. Hook: how teams at their stage consolidate the RevOps stack post-raise.”
From the instructions Percy follows. Illustrative.
That example uses a LinkedIn post as its signal, which the LinkedIn connection cannot find on its own today (see above). A funding round or a job post that Apollo or web search turns up works the same way in a brief.
Every morning: the buying signal scan
Timing moves prospects up and down a list. The daily scan checks every active prospect in HubSpot for news since the last scan: funding, relevant hires or departures, job posts that name your product category, press releases and launches.
Percy sorts each new signal by urgency:
| Urgency | What Percy does |
|---|---|
| Same day | Updates the HubSpot record and sends the rep a one-paragraph summary: the signal, why it matters now, and a suggested angle. |
| This week | Adds a HubSpot note and puts the prospect in the rep's weekly review. |
| Informational | Records it, with no escalation. |
It then updates the last-scan date on every record it checked, even when nothing changed, so the next scan knows what is new, and posts a digest: how many prospects it scanned, the new signals, and any record it could not reach.
Where Percy stops for a person
In Percy’s instructions, an approval is a pause in your conversation with Percy. Percy shows you what it plans to do and waits for your answer before it acts. The instructions set these stop points:
- Before enrichment. “Don’t start enrichment before the user approves the filtered list.”
- Every new segment. “The first time you run a prospecting run for a new segment, persona, or geography, share the proposed filters and sample results before building the full list. Wait for confirmation.”
- Uncertain fit. When Percy is unsure whether a company or role fits, it flags the prospect as uncertain and asks you.
- Smaller results. Percy does not relax your filters to reach a target number without telling you the trade-off.
Edits to these rules wait in a draft until someone publishes them as the next version of Percy.
What Percy will never do
Percy’s limits, from the instructions Percy follows:
- “Don’t invent facts. Never state a funding amount, headcount, or quote you cannot verify.” A field Percy cannot verify stays blank.
- “Don’t send outreach. Your job ends at the handoff brief. You do not write or send cold emails, LinkedIn messages, or call scripts.”
- “Don’t override human-entered CRM data. Enrich empty fields; do not overwrite a field a human has already filled in.”
- “Don’t build lists without a brief.”
- Every claim in a brief must trace to a source Percy checked, and an inference is labelled as one.
The weekly clean-up keeps the list honest too. Percy moves a prospect out of the active list only with a one-sentence, specific reason, such as “Headcount declined below ICP minimum”, sets a date to look again when the situation may change, and never disqualifies a prospect a rep is talking to.
How often Percy runs, and what a run costs
Percy’s daily signal scan suits a schedule: AgentNava can start an agent at a set time, such as weekdays at 8:00 in your time zone, and the scheduled run works in a conversation of its own. List building and enrichment wait for your approval, so run those from your own conversation with Percy.
AgentNava bills by the turn, from prepaid credit. “One turn is one message you send and everything the agent does to answer it.” The price of a turn depends on the agent’s grade, the quality of model it runs on, and ranges from $0.01 to $0.20. A new agent runs at Standard, $0.02 per turn, so asking Percy to build a list is one turn, and your reply approving it is a second. Every search, lookup and HubSpot update inside a turn is included.
Apollo searches and enrichments use the credits of your own Apollo plan, which Apollo bills separately.
Credit is prepaid and does not expire, and your first $5 is on us. An idle agent costs nothing.
Start with Percy
- Open Percy’s outbound prospecting page, read its instructions and workflows, and choose Start with Percy.
- Connect Apollo and HubSpot, and LinkedIn if you want profile lookups.
- Write your ideal customer profile into Percy’s instructions: industries, company sizes, roles, geography, and the problem you solve. Add the HubSpot status names your team uses, then publish the draft.
- Start a conversation with Percy and give it a campaign brief. Read the proposed filters and the sample before you approve.
- Read the first briefs in HubSpot, and tighten the instructions until they read like your best rep’s research.
Frequently asked questions
How do I automate finding and researching sales leads with AI?
Write down your ideal customer, connect a lead database and your CRM, and give an AI agent the job of building the list, researching each prospect, scoring the fit and writing a short brief into the CRM. Approve the list before the agent spends lookups on it. On AgentNava you can start from Percy, a ready-made prospecting agent that works with Apollo and HubSpot. The five steps are in the section above.
Does the agent send the outreach emails too?
No. Percy’s job ends at the handoff brief. Its instructions say it does not write or send cold emails, LinkedIn messages or call scripts, so your team decides what to send.
How does the agent avoid making things up about a prospect?
Every claim in a brief must trace to a source Percy checked, an inference is labelled as one, and a field Percy cannot verify stays blank. It never states a funding amount, headcount or quote it cannot verify.
What does it cost to run a lead research agent?
Asking Percy to build a list is one turn, and your reply approving it is a second: $0.04 on the Standard grade. Searches, lookups and CRM updates inside a turn are included. Apollo bills its own credits separately. For more worked examples, see how much it costs to run an AI agent for a small business.