How to Automate Prospecting With AI: A Step-by-Step Guide

To automate prospecting with AI, define who qualifies, discover matching companies and people, verify the information needed for outreach, and route eligible prospects into an appropriate follow-up workflow. Keep discovery separate from qualification and content preparation separate from permission to send. Then inspect the actual records, messages, and failures produced by the automation.

This guide turns that process into an operational workflow. It starts with a reviewed list, then explains how to make the repeatable parts run through Unify Plays. The aim is a process whose inputs and outcomes you can inspect, rather than a larger list with unclear ownership. No fixed reply rate, setup time, or pipeline outcome follows from enabling AI.

Step 1: How do you turn your ICP into usable criteria?

Write the account and buyer requirements before searching. Separate required company attributes, relevant buyer responsibilities, and reasons to exclude a record. Decide which sources can establish each requirement and what should happen when the evidence is missing. Keep assumptions visible instead of asking AI to fill them in confidently.

Use the ICP and first-signals workflow to organize positioning and customer evidence into a draft profile. Review which attributes are supported by actual customers and which remain hypotheses. Similarity to a customer is a discovery clue; it does not prove that another company needs the same offer.

Define the intended output, too: companies for research, people for review, or people eligible for outreach. Name the teammate who owns exceptions. Choose one audience and offer for the first workflow so you can explain why each included person belongs in it.

Step 2: How do you discover accounts and contacts without treating them as qualified?

In Unify’s list-building chat workflow, describe the companies or people you want, the criteria, and the evidence required. Start from a saved list or attach an approved account file when appropriate. Ask for a table containing the record identity, relevant attributes, source evidence where available, and missing information.

Review and refine the results before saving. Add columns, investigate uncertain values, and inspect supporting reasoning in enriched cells when present. Save only the subset that meets the stated requirements. Check the resulting list: the documentation distinguishes newly added records from duplicates, existing members, and failed saves.

For contacts, use responsibilities as well as titles. The target-account contact workflow emphasizes matching people to the correct account and keeping unresolved contact details out of outreach. A company domain, subsidiary, current role, and email address need to refer to the intended person and organization; a filled row alone does not establish that match.

Step 3: How do you qualify records with evidence?

Qualification evaluates a candidate against your requirements. Discovery finds candidates. Keep both stages explicit so a search result cannot silently become an approved prospect. In a repeatable play, a research agent can assess a record before a prospecting or enrollment action runs.

Configure research-agent questions for the correct record type: company questions for accounts and person questions for contacts. Set the expected answer types and provide guidance on definitions and sources. Test the answers against records you expect to qualify and records you expect to reject, then inspect the supporting research.

Set the qualifying answer requirements deliberately, including whether all requirements or any requirement must be met. Keep records without sufficient evidence out of automatic outreach while someone resolves them. This is a proposed operating rule; do not assume an interface supplies a dedicated uncertainty branch unless you verify that behavior in your configuration.

Contact verification is another check, not a substitute for qualification. Sequence rulesets govern whether catch-all, unverified, and role-based addresses are allowed, while invalid addresses are blocked. Review these settings for the chosen audience. Verification does not establish inbox placement, commercial interest, or permission to contact someone.

Step 4: How do you prioritize signals without overstating intent?

Attach the observation, its source, and its date to the account. Explain why that observation could relate to the problem your offer addresses. Keep account fit separate: activity from an unsuitable account does not make it suitable, and a relevant business change does not establish budget or a purchase deadline.

The website-visitor workflow illustrates an important distinction: a relevant contact at an identified company is not necessarily the individual who visited. Use account-level activity to prioritize research. Do not tell a person they viewed a page unless the evidence actually establishes that identity and its use is appropriate.

Choose a recency window based on the event and workflow, then record that choice. Missing signals mean the selected sources did not provide an observation; they are not proof that the account has no need. Route accounts with unresolved identity or ownership to review instead of manufacturing a reason to send.

Step 5: How do you set exclusions, ownership, and the next action?

Create and preview outreach exclusions for the groups your prospecting motion should avoid. Apply the relevant exclusions to the selected sequence ruleset. An exclusion definition existing in settings is not enough: its application to the actual workflow matters.

Check engagement policies when they govern the user. For assigned users, policy exclusions replace sequence-ruleset exclusions. Review the policy’s permitted accounts, ownership field, and engagement restrictions rather than assuming both exclusion sets apply together. CRM changes must reach Unify before they can affect data-dependent restrictions.

Choose who receives manual tasks and who handles replies. The record owner, sending mailbox owner, and task assignee may differ. Record the intended routing and inspect the actual assignment before expanding the workflow. Keep opt-outs and do-not-email preferences intact; automation should not treat them as inconvenient missing data.

Step 6: How do you draft outreach and choose the right approval path?

Use Unify Sequencing to carry the reviewed audience and research into outreach preparation. Define the offer, purpose, tone, steps, and information the copy may use. Review the sequence structure and recipient content separately: a template preview is not the same as an actual person’s generated draft.

The agentic enrollment documentation distinguishes these entry paths:

For reviewed agentic outreach, generate drafts in chat, open each person’s content, make corrections, and use the enrollment approval controls when ready. Do not assume every AI-generated message always waits for a rep. Do not assume a play’s Sequence action is an agentic draft-review queue either; current documentation describes these as distinct workflows.

Step 7: How do you make the repeatable workflow run as a play?

Once the criteria and routing are clear, build a play by connecting a trigger to compatible actions. Check whether each step receives a company, person, or list of people. Connect qualifying and non-qualifying paths intentionally. Draft changes are separate from the published workflow.

Select the trigger that matches your intent. A manual trigger processes records explicitly sent to it. Audience and criteria triggers check matching records over time, including records already matching when published. They are not automatically limited to future events. If you need a future window, configure an appropriate time condition and preview the matching records.

Choose the correct prospecting action. Prospect can find people at a company using ordered personas and a per-company limit. Its methods distinguish prioritizing existing records, finding the best matches, and finding new people only. Get People returns people already in Unify. Agent qualification evaluates supplied records; it does not discover a fresh contact list.

For automatic template outreach, connect qualifying people to a Sequence action. Configure persona routes, the published template sequence, and sender selection. Review fallback mailboxes for owner-based routing and the Everyone else route. A missing route or unavailable mailbox can skip enrollment. Set the company enrollment limit with other active workspace outreach in mind.

Finally, review play rerun settings. Automated plays normally suppress a record after it starts a run; supported rerun settings can allow another run after a cooldown. A manual trigger is controlled execution, not a dry-run mode: connected actions can still send, enroll, or write. Test only the actions and records you have authorized.

Step 8: What should you inspect before expanding automation?

Use the blank acceptance record below for an approved check. Record what actually happened; leave unknown observations blank until reviewed. Include allowed and excluded records in your permitted test set so you inspect both successful routing and prevention.

Open the play logs to inspect the path and action results for each record. Then inspect the resulting sequence enrollment. Queued, started, blocked, and completed describe different states. A completed play does not prove an email was sent or read. Resolve blocked content or mailbox access and check the enrollment again.

Review incoming replies and their full context. Replies can end outreach under the sequence’s ending conditions; out-of-office responses have separate handling. Assign a person to review requests, referrals, and objections. If incorrect targeting or copy appears, pause the affected workflow and inspect existing enrollments separately rather than assuming a paused play cancels prior outreach.

Use sequence results for ongoing review, keeping audience, version, and observation period comparable. Read the documented denominator before reporting a rate; sequence and step metrics differ. Keep failures, skips, and unresolved records visible. Expand only when the team understands why records entered, what happened, and who owns the next action.

Frequently asked questions

What parts of prospecting can AI automate?

AI can assist with research, candidate discovery, enrichment, qualification questions, and drafting. Configured workflows can route records and enroll eligible people. Verify each task’s output and exceptions; a completed automation is not proof of qualified demand.

What should stay under human control?

Keep someone responsible for audience criteria, evidence quality, outreach eligibility, unusual messages, and buyer conversations. Choose review requirements deliberately and verify the actual entry path instead of assuming all generated content waits for approval.

Can a Unify play generate agentic outreach for approval?

The documented Sequence action enrolls people into published template sequences. For per-person AI-generated drafts and review, use the separate agentic enrollment flow. Chat and list entry paths support review; person-record enrollment can start without separate draft approval.

How do I check AI-sourced contact data?

Review identity, current role, account association, source evidence, and verification status separately. Hold unresolved required fields out of outreach. Multiple sources can improve retrieval opportunities, but neither enrichment nor a populated email field guarantees accuracy or inbox placement.

Do I need one platform or several tools?

Either can support a workflow if record identity, permissions, handoffs, and outcome checks are clear. A consolidated platform reduces some handoffs, while connected tools require explicit mapping and error handling. Evaluate the complete path from discovery to a reviewed next action.