Short answer: Formula started by targeting freelancers who were just starting out. Our call-to-close conversion sat at 3%.
We pivoted our own ideal customer profile (ICP) to established SMBs and small firms with existing revenue and a validated offer. Conversion jumped to 18% — a 6x improvement — with no change to the product, the pricing, or the agent itself.
Why did we start by targeting beginner freelancers?
It felt logical at launch. Freelancers just starting out are everywhere, they are easy to reach on LinkedIn, and €97 a month sounds affordable when you are picturing someone with a tight budget.
We also assumed the people who need help selling the most are the ones who have never sold much. That assumption was wrong, and it cost us a year of mediocre numbers before we admitted it.
What actually went wrong with that ICP?
Our agent did its job: it built the list, sent the LinkedIn invitations, wrote the follow-ups, pre-wrote replies. The mechanics worked exactly as designed.
The problem sat one layer up. Beginner freelancers rarely have a tested offer, a repeatable pitch, or a predictable pipeline of their own to compare against. When a qualified call landed on their calendar, most of them did not know what to say once they got there.
A great agent handed them a great opportunity, and an unproven sales process wasted it. Our own call-to-close conversion on that segment stabilized around 3%.
We break down exactly how to spot this failure mode before it costs you a year in how to define your ICP before turning on an AI prospecting agent.
What changed when we pivoted to established SMBs?
We rewrote our targeting toward established SMBs and small firms: consulting practices, agencies, businesses with a few years of revenue behind them and a sales process that already existed, even a manual one.
Nothing else moved. Same agent, same €97/month starting price, same 2-week trial, same LinkedIn and email mechanics.
The difference showed up the moment the call started. These buyers already knew their numbers, their pain, and their budget. They did not need to be convinced that prospecting mattered; they needed to be convinced our agent could do it better than the 2 hours a day they were already spending on it manually.
That is the exact profile behind the anonymized case we reference across the site: SMBs in consulting and solar panel installation, each spending roughly 2 hours a day on manual prospecting, going from around €400k to €600k in monthly revenue after switching to Formula. Full detail in our case study.
How much did the ICP pivot actually move our numbers?
One number changed everything, and it had nothing to do with the product.
| Signal | Beginner freelancers (before) | Established SMBs (after) |
|---|---|---|
| Revenue at signup | Unstable, often pre-first-clients | Existing, recurring revenue |
| Manual prospecting history | Rarely established | Already prospecting manually, often daily |
| Offer clarity | Still being defined | Already validated with paying clients |
| Fit for a €97/month tool | A hesitant, discretionary spend | A straightforward line item |
| Call-to-close conversion | 3% | 18% (6x) |
Every row above the last one explains the last one. The agent stayed constant. The buyer changed. The conversion rate followed the buyer, not the tool.
Why does ICP matter more than the AI agent itself?
Salesforce research found that sales reps spend less than 30% of their time actually selling — the rest disappears into admin and searching for the right prospects.
An AI prospecting agent is built to reclaim that time: it prospects on its own, choosing targets, writing every outbound message and every follow-up, on LinkedIn and email, with no human validation needed on outbound.
The only moment you step in is when a prospect replies: a swipe from your phone, a few seconds, to validate the agent's proposed reply before it sends.
Formula sends 28 LinkedIn invitations a day — the platform's own limit — combined with email on the same prospects, reaching around 840 people a month across both channels.
Reply rates on those campaigns run 30 to 60%, depending on sector and offer, against a 7 to 11% market average for classic outreach. Replying in under 5 minutes multiplies conversion by roughly 100 compared to a next-day reply.
None of that changes who you are reaching. An agent this fast just runs the experiment on your ICP much faster than you could by hand, for better or worse.
“The agent didn't get smarter when our numbers went from 3% to 18%. Our prospect list did. We spent a year optimizing message copy when the real lever was sitting one step earlier, in who we were even talking to.”
— Laura Terriou, co-founder, Formula.
How do you know if your current ICP is wrong?
Four questions we now ask before anyone activates an agent, learned the hard way on our own funnel:
- Do they already have revenue? Not projected revenue, existing recurring revenue that makes €97/month a rounding error, not a leap of faith.
- Have they been selling manually for 3 to 6 months? A repeatable pitch and a real sense of their own pipeline matter more than any tool.
- Is their offer already validated? Paying clients, not just interest. An agent scales what converts; it cannot invent conversion from nothing.
- Would they treat the spend as a line item, or a gamble? Hesitant buyers churn fast, whatever the reply rate looks like on paper.
If you answer no to more than one of these, the fix is not a better agent. It is a narrower, more established ICP, applied before you automate anything.
We walk through this filter in more depth in which B2B businesses actually benefit from an AI prospecting agent and what a realistic B2B pipeline looks like for a solo consultant in 2026.
Frequently asked questions about ICP and AI prospecting agents
What is an ICP and why does it matter before turning on an AI prospecting agent?
An ICP (ideal customer profile) is the specific type of company and buyer your offer converts best with: their size, their budget, their maturity, their existing buying behavior.
An AI prospecting agent amplifies whatever list you point it at. If the ICP is wrong, the agent just personalizes and automates rejection faster. Fixing the ICP is the first lever, before any tool.
Why did Formula's call-to-close rate go from 3% to 18%?
We originally targeted freelancers who were just starting out: unstable revenue, unproven offers, tight budgets.
We pivoted to established SMBs and small firms with existing revenue and an already-validated offer. Same product, same pricing, same agent. Call-to-close conversion went from 3% to 18%, a 6x improvement, because the buyers themselves were a better fit.
Can beginner freelancers ever use an AI prospecting agent?
Not as a starting point. An agent scales an offer and a sales motion that already work manually. Someone just starting out usually has neither yet.
The right order is to validate the offer manually first, build a repeatable sales conversation, then bring in an agent to scale what is already converting.
How is Formula's AI agent different from just running outbound automation?
Formula's agent prospects on its own: it picks targets, writes each outbound message and follow-up, on LinkedIn and email, with no human validation on outbound.
You only step in when a prospect replies, with a swipe from your phone to validate the agent's proposed reply before it sends. That is the only human moment in the loop.
What does a good ICP look like for an AI prospecting agent?
A good ICP already has 3 to 6 months of regular manual sales, a persona that converts predictably, and enough revenue to treat an outbound tool as a normal line item, not a gamble.
If manual prospecting has never produced a client, an agent will not fix that: it will only run the experiment faster.
Established revenue, a validated offer, and a manual prospecting habit already in place?
Start the 2-week trialSee pricing if you want the numbers before the trial.