Search agents: scanning the prospect pool continuously

Thursday, 8:40 in the morning. A three-person sales team starts its weekly ritual: one of them works through business directories region by region, another goes back over last week's list, the third pastes whatever they find into a spreadsheet. Two hours later the sheet holds 46 rows. Thirty-one of them were on last week's sheet too, nine are already open in another rep's pipeline, and six are genuinely new. Six rows for two hours of combined effort — and next Thursday those same two hours get spent again, because nobody keeps a record of where the search stopped last time.

This article is about turning those two hours into a rule you define once and then let run. In order: what manual prospecting really costs, what a search agent takes over, which criteria belong to the agent and which stay with a person, how to pick a scan frequency, why a fresh match should not become a lead immediately, how agent output enters the working day, what prevents duplicate contact, how many agents are too many, the numbers that show whether it works, and what to do when an agent goes quiet.

What does manual weekly prospecting actually cost?

The cost is not the hours; it is where those hours go. Manual scanning leaks in three places. First, repetition: the same records get reviewed week after week, because "we already looked at this one" lives in the memory of whoever built the list, not in a system. Second, displacement: the search eats the morning hours when a rep is sharpest. Third, latency — a company that added a payment stack to its site on Monday gets noticed four days later at best, and usually not at all.

All three leaks share a cause: prospecting is treated as a habit rather than a process. Nobody has written down who scans, against which criteria, how often. Teams that take the discipline of finding buyers rather than waiting for them seriously run prospecting the way they run quote follow-up: defined input, defined output, measured repeat rate.

There is a quick way to see where you stand. Of the new records you created in the last three weeks, how many belong to a company that had never appeared in your system before? If that share is low, your problem is not a small pool — it is the same pool being scanned over and over, and buying another data source before you fix that is carrying water in a leaking bucket.

What exactly does a search agent automate?

A search agent is not "AI that finds customers". It does something much narrower, and that is precisely why it is dependable. With the Search Agents feature in Opportunity Radar you define your criteria once; the agent re-scans the pool on the interval you set, finds the new businesses that match, and adds them to your list. The load-bearing word is "new": the agent hands you the delta since the last scan, not the whole pool again.

That difference explains why the automation holds. A human search restarts from zero every time; an agent knows what it saw last week. The pool itself is not static either — it draws on Europages, official trade registries, map data and sector sources, with records for the same company merged so you see one card instead of three rows. A team that understands how the discovery pool is built also understands what to expect from an agent.

Two things get confused because the names rhyme. A search agent is a scanning rule: it talks to nobody, writes to nobody, persuades nobody. Voice AI agents that speak on the phone operate at the moment of contact and carry an entirely different risk profile. One builds the list, the other talks to it; putting them in the same mental box guarantees you measure both incorrectly.

Which criteria belong to the agent, and which stay with a person?

A criterion has to clear three bars before it becomes an agent. It must be writable: a value that fits in a filter field, not an impression in your head. It must be stable: its meaning cannot shift from week to week. And it must be selective — a criterion that returns most of the pool is functionally the same as no criterion at all.

In Opportunity Radar the filters that clear those bars are explicit: city, district, category, digital maturity and fit score. On top of them sits Technology Signal Targeting, where you target businesses that use — or conspicuously lack — specific technologies such as e-commerce, payments, live chat, WhatsApp, booking or hiring. When what you sell closes a gap, the "missing" filter is worth more than the "present" one: a restaurant with no booking system is already a defined prospect for anyone selling booking software.

What stays with a person is equally clear: how the decision maker reacts, when budget lands, how warm the relationship with the incumbent vendor is. Those come out of conversations, and forcing them into a filter narrows the agent for no gain.

A practical sequence helps. Save the criterion first as a Saved Search Preset and run it by hand two or three times. Read the top twenty rows it returns, and only when you would genuinely contact them, promote that preset to an agent. Handing an untested criterion to an agent means producing the wrong list reliably every week — and the error stays invisible, because a full list looks like a working one.

How often should the agent run?

Frequency follows the signal, not your patience. Scanning a criterion that changes once a year on a daily schedule just means an empty email every morning; scanning a signal that moves every few days on a monthly schedule means arriving after the window closed. The table below is a reasonable starting layout for common criterion types — shift it as you observe your own market.

Criterion type	How fast the signal moves	Sensible scan	Notification
New job posting, booking tool or live chat appearing	Days	Daily	Daily digest email
Newly registered business, first entry into a category	Weeks	Weekly	Weekly digest email
Change in payment stack or e-commerce platform	Weeks	Weekly	Weekly digest email
Geographic expansion by city, district or country	Months	Monthly	In-app only
Raising the fit-score threshold	Slow but continuous	Weekly	In-app only

On the notification side, Signal Monitoring and Digest does the carrying. Put a company on watch and its high-intent signals accumulate in its history, arriving as a daily or weekly digest email if you want one. That way you are not depending on the habit of opening a dashboard — and any automation that depends on a habit gets abandoned during the first busy week.

Why should a fresh match not become a lead straight away?

Every business the agent surfaces is a candidate, not a lead. The distinction is not administrative, it is measurement: a lead occupies pipeline, enters the forecast, and gets assigned to someone. Dumping a hundred unscreened records in makes every pipeline number unreadable and turns the weekly meeting into a cleanup session.

So agent output lands first in the Saved Candidates Kanban, moving by drag and drop across Saved, Rejected and Converted to Lead, with bulk promotion for the ones you keep. What makes screening fast is the Business Intelligence Card: digital maturity, website and email status, social accounts, the technologies in use and the decision makers sit on one card, so the call gets made without switching tabs.

The Rejected column has a side benefit — it is the best training material you have for your ideal customer profile. When you are working on getting a new rep productive quickly, talking through "why was this one rejected" across thirty real records lands faster than any written profile document.

How does agent output enter the sales day?

The most common failure of automation is that its output has no slot in anyone's calendar. The agent runs, the list piles up, nobody opens it. The fix is unglamorous: block a fixed window on the day the digest arrives — twenty minutes of screening, then contact, in that order.

Which channel you use follows what the card shows. For a record with a verified number flagged as mobile, dialing through the list is fastest, and the dialing patterns that raise connect efficiency chew through it while keeping the rep off the keypad.

What comes out of the call has to be recorded too. Tracking calls through disposition codes is the only honest source on what the agent's list actually produced; "we called, nothing came of it" measures nothing and leaves you no way to correct the criterion next week.

On email, Bulk Email Outreach lets you send to selected candidates from one screen, but opening the volume tap at once wrecks deliverability. From a fresh sending domain, warming that domain up in stages is the precondition for carrying the volume an agent produces.

There is a budget detail as well. Contact details arrive masked and are opened through Credit-Based Contact Unlock; once opened they stay open, and viewing the same company again costs nothing extra. So the right order is screen first, unlock second — credits should only follow records you will genuinely contact.

What stops you from touching the same company twice?

A scan that never stops inevitably raises the risk of duplicate contact. Three mechanisms close it. The first sits at conversion: when you move a business across with One-Click Lead Conversion, duplicates are prevented automatically by phone and email matching, so the same company does not land in your list a second time.

The second sits at send time. Everything you add to the Do-Not-Contact List — one by one or by bulk upload, whether email, phone number or company — is checked automatically before every send. Competitors, current customers and anyone who has said "do not write again" belong on it; keeping that list current is less visible than building a campaign and saves more business.

The third is address quality. Unlocked addresses are checked against MX records and marked valid, risky or invalid, and you can see whether a number is mobile or landline. If the person's address is not in the pool, a likely one is derived from the company's known pattern and run through validation, so credits do not go to mailboxes that will bounce.

The more channels you add, the more that check matters. In WhatsApp flows that move from chat to a voice call, reaching the same person on three channels in one week closes doors faster than indifference ever does.

How many agents are too many?

The correct upper bound is the number at which the digest still gets read. In practice three to five active agents cover most teams. Above that, two problems start at once: agents overlap and the same record reappears in several lists, and no agent has an owner because everyone assumes someone else is watching it.

The rule is simple — every agent gets one owner and a purpose you can write in one sentence. "Restaurants in the city centre with no booking system" is a purpose; "potential customers" is not. An agent whose purpose cannot be written is one whose output cannot be judged, and it is the first to switch off during a cleanup.

Campaign agents need an end date. The agent you open while building the sales plan for a new product launch should close when the launch window closes; otherwise it keeps producing lists six months later against a criterion nobody remembers, and those lists quietly erode trust in every other agent.

Before opening yet another agent, try one more move: use Lookalike to pick your best customer and let the system surface companies that resemble it. More often than not, an existing customer describes the profile a sixth agent was struggling to spell out.

Which numbers tell you the agent is working?

Four numbers are enough. Weekly new candidates shows volume. Rejection rate shows criterion quality. Candidate-to-lead conversion shows how well the criterion and the message fit each other. Lead-to-first-meeting shows whether these companies are really your market. Reading them separately beats staring at a single blended conversion figure.

The most commonly misread is the rejection rate. As an illustration: if the large majority of arriving candidates get rejected on sight, the criterion is too loose; if almost none are rejected, it is so tight that it will never produce meaningful volume. A healthy agent produces a list worth looking at in which not everything qualifies.

Measure the message separately. Outreach Analytics shows which segment and which message actually convert, so when the criterion is good but nothing comes back, the problem is the copy rather than the list — and writing to the same list at higher volume only spends your domain reputation.

Do not forget the credit side: meetings produced per unlocked contact summarises the economics in one line. It is the first number to degrade when screening discipline slips, and because volume is usually rising at the same time, the slide looks like progress at first glance.

What do you do when an agent goes quiet?

An agent returning nothing for weeks is not a fault, it is information: the reachable pool for that criterion is exhausted for now. Try three moves in order, each loosening a different constraint.

The first is geography. Multi-Country Targeting lets you narrow the pool by country and category and target businesses abroad, so a criterion that has run dry in one market can fill a list again in another. Map View earns its place here, because you read where density stops on a map with heat and clustering views rather than in a table.

The second is loosening the signal: dropping the fit-score threshold one notch, or removing one of two technology conditions, usually gets the list flowing again. The third is reversing direction — with Need Search you type in the product or service you sell and get the businesses that need it, with the ones showing a digital gap surfacing first. That is especially useful for a new product, while your criterion has not settled.

Over the long run this discipline should not stand alone. The inbound side has to grow too; demand built on the fundamentals of search engine optimisation covers the weeks when an agent goes quiet.

If you want to define your criteria once and let the pool be scanned for you, open a free Rocketly account and set up your first search agent.

FAQ

Will a search agent send messages on my behalf?

No. The agent only scans the pool and adds newly matching businesses to your list. Sending is a separate step: bulk email or calling happens with your approval and after the do-not-contact list has been checked.

What happens if the same business matches several agents?

The record stays single — duplicates are prevented by phone and email matching at the moment of lead conversion. Even so, overlapping agents increase the screening load, so it is worth pulling the criteria apart.

How often should I review an agent?

Once a month is enough. What you review is not whether the agent ran, but the rejection rate of the list it produced; if that rate has drifted to either extreme, update the criterion.

Can I screen candidates without unlocking contact details?

Yes. Digital maturity, the technology list and website status on the business intelligence card are usually enough to screen on, so credits go only to the records you will actually contact.