# How to Build an AI Agent Prospecting Workflow You Can Run Today

> A practical, step-by-step setup where an AI assistant finds, qualifies and drafts outreach for local business leads, while you keep the final say on what gets sent.

Prospecting eats the hours you should spend on paid work. You search for businesses, copy details into a sheet, hunt for an email, guess at the owner's name, write something personal, then forget to follow up. An AI agent can take over the repetitive parts, but only if you give it a clear workflow. Without one, you get a pile of generic emails sent to the wrong people.

This guide walks through a workflow you can set up today. The AI does the grunt work. You make the judgment calls.

## Decide what the agent is allowed to do

Before touching any tool, write down the jobs you will hand over and the ones you keep. A sensible split for a small business or agency:

- **Agent does:** search for leads, pull public business details, score how well each lead fits, research each website, draft a first line and a short email.
- **You do:** approve the target list, review drafts before they go out, decide who gets a call, reply to interested people.

Keeping a human on the send button matters. AI drafts can be confidently wrong about a business, and one bad assumption in a cold email costs you the lead.

## Step 1: Write a one-paragraph targeting brief

Agents work best with tight instructions. Write a brief you can paste into any AI tool:

- **Who:** business type and location, such as "independent house painters in Leeds and Bradford".
- **Fit signals:** has a website, has reviews, no recent social posts, runs ads, whatever matters for your offer.
- **Disqualifiers:** chains, franchises, businesses outside your service area.
- **Offer:** one sentence on what you sell and the result it gets.

If you are unsure which niche to start with, pick one where you can name a specific problem. Our pages on [painters](https://getclient.pro/industries/painters) and [law firms](https://getclient.pro/industries/law-firms) show how the angle changes by industry.

## Step 2: Give the agent a real lead source

An AI chatbot asked to "find 50 plumbers in Austin" will often invent names or give stale results. Feed it real data instead.

GetClient's Google Maps scraper searches live by keyword and location, and you can run bulk keywords across several cities at once. Each lead can include business name, category, address, phone, website, emails found, the owner's name and title with a confidence score, and social profiles, where these are public. GetClient also has a REST API and an MCP server, so an AI tool that supports MCP can request and work with leads directly rather than guessing.

Already have a list from elsewhere? [Import leads from CSV](https://getclient.pro/blog/import-leads-from-csv) into My Leads and add custom fields for the agent's output, such as fit score and first line.

## Step 3: Let the agent qualify and research

Now give the agent each lead with your brief and a fixed set of questions. Ask for structured output so you can sort it:

1. Fit score from 1 to 5, with one sentence explaining why.
2. One specific, verifiable observation from their website or listing.
3. A suggested first line using that observation.
4. Any reason to skip this lead.

Tell it plainly: "If you cannot find something, say so. Do not guess." This single instruction cuts most made-up details.

Then spot-check. Open five leads at random and confirm the observations are true. If the agent is wrong on more than one or two, tighten the brief before continuing. Drop anything scored 1 or 2.

## Step 4: Clean the list before anything is sent

Run email verification on the qualified list. Bounces hurt your sending reputation, and an agent has no idea whether an address still works. Also confirm your domain is set up correctly; if SPF, DKIM and DMARC are unfamiliar, read [SPF, DKIM and DMARC explained](https://getclient.pro/learn/spf-dkim-dmarc) first.

Cold email rules differ by country, so check the ones that apply to you and your recipients. At minimum, identify yourself honestly and give people an easy way to opt out.

## Step 5: Turn drafts into a short sequence

Don't let the agent write a unique 300-word essay per lead. It is slow to review and tends to drift. Instead, use a template with merge tags and one personal line from the research step:

> Hi {{params.firstname}}, [agent's first line]. I help {{params.category}} businesses in {{params.city}} get more booked jobs from their website. Worth a quick chat about {{params.website}}?

Store the agent's personal line in a custom field and review it lead by lead.

Build two or three follow-ups. A [three step email sequence](https://getclient.pro/blog/three-step-email-sequence) is a good starting shape. Preview each step before scheduling, and set weekday send limits so you stay at a modest pace from your own Gmail, Outlook or SMTP mailboxes. Every email includes a one-click unsubscribe link, and unsubscribed addresses are skipped automatically.

## Step 6: Keep a human on replies

This is where you step back in. Read replies in the Inbox, answer interested people yourself, and use the agent only to suggest drafts. A prospect who replies wants a person.

For leads who clicked but never answered, a short call can work well. Our [cold call script after email](https://getclient.pro/resources/cold-call-script-after-email) gives you wording that references the email without sounding pushy.

## Review the workflow weekly

Once a week, look at campaign analytics and ask:

- Which fit scores turned into replies? Adjust the brief to favour them.
- Were any first lines wrong or awkward? Add rules to the agent prompt.
- Are opens fine but replies low? The offer or ask probably needs work. See [how to fix a low reply rate](https://getclient.pro/blog/fix-a-low-reply-rate).

Treat the prompt like a living document. Small edits each week add up to an agent that understands your market.

You can try this workflow with the 100 free credits new GetClient accounts get, no card needed. Plans and credit packs are on the [pricing page](https://getclient.pro/pricing).

## Frequently asked questions

### Can an AI agent send cold emails completely on its own?

Technically some setups allow it, but it is risky. Agents can misread a website or invent details, and those errors go straight to a prospect. Let the agent draft and score, and approve sends yourself, at least until you trust its output.

### Which AI tool should I use as the agent?

Any capable assistant can handle qualification and drafting if you give it clear instructions and real data. If your AI tool supports MCP, it can connect to a lead source like GetClient's MCP server instead of relying on its own memory.

### How many leads should I start with?

A small batch, say 30 to 50 as a rule of thumb, is enough to check the agent's accuracy and see whether your offer gets replies before you scale up.

### How do I stop the AI from making things up about a business?

Tell it explicitly to say when it cannot find information, ask for observations it can point to on the website, and spot-check a random sample of leads before sending anything.

### Do I need a separate domain for AI-assisted outreach?

Many senders use a separate domain to protect their main one, but it is a choice, not a requirement. Whatever you use, set up SPF, DKIM and DMARC and keep daily volumes modest.

Source: https://getclient.pro/blog/ai-agent-prospecting-workflow