AI prospecting: better lists don't mean more meetings

AI for sales prospecting is the use of models to decide who to contact: watching for hiring, funding, tech installs and job changes; scoring accounts against the shape of deals that closed; and assembling the phone numbers and emails to reach them. It replaces the research half of prospecting, and it does it well.

The research half was never the expensive half. What follows is the arithmetic that makes that concrete, because it is the reason so many teams buy good AI prospecting tools and see no movement in booked meetings.

What AI does to a list, briefly

Three capabilities, all real and all now table stakes.

Signal aggregation. Rather than filtering on firmographics that were true last quarter, the tool watches for events — a funding round, a spike in engineering hires, a competitor’s tag appearing on the site — and surfaces the account when several land together.

Adaptive scoring. Instead of fixed points per job title, the model learns from closed-won history what a buyer looks like, and updates as more deals close.

Job-change monitoring. A champion who moves takes their opinion of your product with them. Catching that within days is one of the highest-converting triggers in outbound, and it is genuinely hard to do manually at any scale.

None of this is in dispute, and none of it is where teams get stuck. The problem starts one step later.

The arithmetic: a queue that outruns the team

Set up a plausible month and work it through. Every number here is an input you should replace with your own; nothing is a benchmark.

One rep. Five hours a day on the phone, twenty-five dials an hour dialling manually — call it 125 contact attempts a day. Your prospecting tool flags 400 accounts as high-intent this month. Reaching a decision-maker on a cold line reliably takes more than one attempt; use three.

• 400 accounts × 3 attempts = 1,200 attempts
• 1,200 ÷ 125 per day = about ten working days

Ten working days is two weeks. So an account flagged on day one is finished inside the fortnight, and an account flagged on day twenty has not had its first attempt yet — by which time the hiring spike that flagged it has been filled, and the funding announcement has been read by every vendor in the category.

The list is decaying faster than the team can work it. Better targeting does not help here. Better targeting raises the conversion rate on each attempt while leaving the number of attempts exactly where it was, and the binding constraint is the number of attempts.

Change the constraint and the shape changes. A dialer running four lines makes four attempts per slot rather than one — that is arithmetic on the line count, not a performance claim — and the same 1,200 attempts fit into roughly two and a half days instead of ten. The signal is still warm when the call lands, which is the entire point of having generated the signal.

This is why connect rate and dials per hour belong on the same dashboard as your intent scores. They are not separate concerns. One of them decides whether the other one ever gets used.

Three ways an AI prospecting stack stalls

Every signal is treated as equally fresh. Most tools surface a flagged account without a half-life attached, so a job posting from yesterday and a funding round from last quarter arrive in the same queue looking identical. Give each signal type a decay window and work them in that order; it costs nothing and it changes which accounts get the scarce attempts.

The model never hears what happened on the phone. Scoring learns from closed-won records, which are the deals that reached the CRM. It does not learn from the four hundred calls where a rep found out in ninety seconds that the budget was frozen, because “no budget until Q3” is a call disposition rather than a closed-lost opportunity. Everything the model most needs to know is discovered on calls it cannot see.

Success is measured in activity. Emails sent, accounts flagged, sequences launched. All three go up when you buy the tool, by construction. The number that tests whether targeting improved is dial-to-meeting rate — meetings per attempt, not meetings per month — because that is the one that holds when attempt volume also changes.

Closing the loop

The fix for the second failure is structural rather than clever. Dispositions from live conversations have to reach the scoring model, which means they have to be written back rather than typed into a notes field.

Concretely: a rep works a flagged account, learns in the first minute that the company outsources the function entirely, and marks it. That outcome flows to the CRM, the model reads it, and accounts that look like that one drop down the queue. Do that a few hundred times and the model has learned something no amount of firmographic data could have told it — because it happened in a conversation, and conversations are the only place it exists.

The prerequisite is volume. A loop that closes forty times a month teaches the model almost nothing; one that closes four hundred times teaches it a great deal. Which returns to the same constraint the arithmetic started with.

What to do first

Before evaluating a single prospecting tool, run one calculation.

Count how many contact attempts your team can actually make in a week. Then ask each vendor on your shortlist how many accounts their tool would flag in the same week. If the second number is larger than the first, targeting is not your constraint and a better list will produce a longer queue rather than a fuller calendar.

If it is smaller, buy the prospecting tool — that is the right purchase and it will work.

Most teams find the second number is larger, often by a lot, and it is a genuinely uncomfortable finding because the prospecting tool is the more interesting product to buy. Pharo sits on the other side of it: a parallel dialer that dials two, four or eight lines at once, drops the dead numbers before a rep hears them, and writes every outcome back so the loop above actually closes. Enrichment runs through a multi-provider waterfall passed through at cost, with no margin on the credits — what that means on a real run is worked through in our Clay alternatives comparison, and the published rates across the category are in the dialer pricing index.

The list was never the hard part. Getting through it while it is still true is.