How to define an ICP for your agency's outbound
An ideal client profile for outbound is a filter a stranger could apply: a named audience or a named service, a visible team, a live signal that the problem is happening now, and a short list of disqualifiers. Build it from the last ten clients you would take again rather than from an industry and headcount template, then test it by splitting your sends into segments and reading the reply rate of each one.
Pull up your last twelve months of closed deals and sort them by how little trouble each client turned out to be. The top third came from two or three company types that look nothing like the service list on your website. That gap is the problem with the ideal client profile that goes into most outbound campaigns: it was written in a room with the people who already work there. RSW/US, which runs an annual survey of agency leaders on new business, reports that 85 percent of agencies built their positioning using only their own employees, and only about one in five tested it on real prospects. Then 2,000 emails go out against that untested definition, three people reply, and the verdict is that cold outbound does not work for agencies. The list was the experiment. Nobody designed it.
What is an ideal client profile for agency outbound?
An ideal client profile for outbound is a filter a stranger could apply to a company in thirty seconds and reach the same verdict you would. That is a higher bar than a description, and it is the bar that matters, because the thing consuming your profile is a scraper, a search filter and a junior researcher, not your own judgement.
Four parts make it operational. A structure test says what kind of business it is, in terms visible from the outside. A named buyer says whose inbox you are aiming at. A signal says why this month rather than any month. Disqualifiers say who to drop even when the first three pass. Everything else in the usual template is commentary.
The distinction worth holding onto is between the profile and the persona. The profile is company-level and it builds the list. The persona is person-level and it decides the job title you search for and the first line you write. Instantly's ICP template, updated in May 2026 and one of the pages answer engines reach for on this question, gets that order right and then fills the company half with twelve fields: industry, company size, annual revenue, location, funding stage, tech stack, buying triggers, decision makers, secondary contacts, influencers, disqualifiers and pain points. Fill all twelve honestly for an agency buyer and you still cannot build a list from it, because six of those fields are unknowable from outside a private company.
Why do firmographic templates underperform on cold outbound?
Industry, headcount and revenue describe a crowd, and a crowd is not a reason to write to anybody in it. Those three fields have one property that makes them attractive and one that makes them weak: every database offers them, which is exactly why every competitor is filtering on them too.
Look at what the standard agency example in that template actually says: companies with 10 to 100 employees, revenue of one to twenty million, average deal size above ten thousand. Three of those clauses are guesses. Private agency revenue is absent or wrong in most data providers, average deal size is never published, and the headcount band is wide enough to hold a two-person studio and a company with its own operations director. A stranger handed that definition cannot sort a list with it, so they fall back on the one field that is populated, and you end up mailing by industry code.
The deeper issue is that none of those fields say whether the problem you solve is happening right now. A 40-person recruitment agency with a full pipeline and a 40-person recruitment agency whose referrals dried up in September are the same row in Apollo. Only one of them will reply. The filters that separate them are structural and behavioral, and you have to go looking for them.
There is also a market reason to get this right in 2026. RSW/US surveyed senior executives at agencies across the US and Canada in August and September 2026 and found referrals fell from 76 percent to 44 percent as a top source of new business in a single year, while 75 percent said breaking through to prospects is harder than it was, up from 35 percent the year before. More agencies are doing outbound into the same inboxes with the same database filters. Precision is the only lever left that nobody can copy from your website.
How much does your ICP move the reply rate?
Changing who you write to moves reply rate by roughly 3x on company size and seniority and by about 10x across industries, which is more than any subject line will ever return. Those multiples come from the only kind of dataset that can show them: a provider's own sends, broken out by attribute.
Belkins analyzed 7,530,489 cold emails sent across its client campaigns between January and December 2025, with 34,393 replies, and published the breakdown in June 2026. Reply rate there means unique replies divided by emails sent. The overall average was 0.45 percent, and the spread underneath it is where the argument for a real ICP lives.
| Who you wrote to | Reply rate per email sent | What it tells you |
|---|---|---|
| Companies with 0 to 10 employees | 0.72% | The buyer reads their own inbox |
| Companies with 11 to 50 employees | 0.49% | The practical middle for most agency offers |
| Companies with 10,000 or more employees | 0.22% | A screening layer sits in front of the buyer |
| Founders and owners | 0.57% | The person who can say yes without a committee |
| C-level at larger organizations | 0.42% | Delegated, screened, slower |
| VPs | 0.32% | The hardest seniority group in the dataset |
Belkins describes the company size relationship as almost perfectly linear, declining from 0.72 percent at the smallest companies to 0.22 percent at enterprises. Founders and owners beat C-level executives by 35 percent and VPs by nearly 80 percent, which the report attributes to the absence of the administrative layers that insulate executives at larger firms. The industry spread is wider still: food and beverage replied at 3.47 percent, nearly eight times the overall average, while banking and insurance sat at the bottom. Treat gaps under a tenth of a point as noise, since the report puts its own margin of error at plus or minus 0.05 points.
Then there is the distribution. Smartlead's benchmark set, built from more than 850 million emails sent through its platform, measures replies per contact rather than per email and reports a median of 0.74 percent, one reply per 135 people contacted. The top quartile reaches 1.43 percent, the top decile 2.63 percent and the top 5 percent 3.85 percent. That is a 3.5x gap between the typical sender and the top decile, and a 16x gap from top to bottom. Almost none of it is explained by writing.
One caution before you compare your own number to anything published. Belkins points out that a 5 percent reply rate measured against openers and a 0.45 percent rate measured against total sends can describe the exact same campaign, and it switched denominators in this edition because open tracking was hurting deliverability. Woodpecker's statistics page quotes 5.8 percent for campaigns under 50 recipients against 2.1 percent for large lists, sourced from an earlier study with a different denominator again. All three can be true. None of them is your baseline.
How do you pull an ICP out of the clients you already have?
The raw material is the last ten clients you would happily take again, and four questions asked about each one. Everything else in the exercise is bookkeeping.
Write a row per client and fill four columns. What was broken in their business the month before they contacted you, in their words from the call, not yours. What event happened in that same month: a hire, a lost client, a funding round, a new product, a regulation, a quarter that missed. Who signed, and who else was in the room. What made them cheap to deliver for, which is usually a structural fact like an existing team, a single decision maker or data you did not have to clean.
Now look for the clause that repeats. Three of ten sharing a structural fact is a pattern worth testing; one of ten is a story. The useful finding is almost never the industry. It tends to be something like "they had already hired for the role and the hire was drowning" or "they sell one service to one audience and had outgrown referrals". That is a sentence you can go looking for in public data.
Agencies resist this because narrowing feels like turning away money, and the industry numbers argue the other way. Promethean Research surveyed 119 agency leaders in February 2026, averaging 31 employees, and found agencies that reduced their service mix grew 13 percent and posted 30 percent net margins against a 13 percent industry average net margin for 2025. Sales, profit margins and lead generation were the three biggest concerns heading into the year. A sharper definition of who you serve is the cheapest of those three to fix.
One honest limit on the exercise: if you have fewer than about five clients you would repeat, you do not have a pattern yet. Interview buyers instead and treat the first list as a hypothesis, which is a different and slower job than filtering.
How do you write a structure test a stranger can apply?
A structure test is one sentence a person with the company's website and LinkedIn page open can mark true or false without asking you. The test for your test: hand it to someone who does not work in your business, give them twenty companies, and see whether their verdicts match yours.
Three rules keep it applicable. Require either a named audience or a named service, because a company that sells three or more disciplines to everybody cannot be positioned against and cannot be qualified from outside. Require something visible that proves a team exists, like people in a delivery role on LinkedIn or named staff on a site, since that is the closest honest proxy for whether they can pay you. And write it as clauses that fail, not adjectives that flatter.
Here is the same profile before and after. Before: "B2B companies, 10 to 100 employees, one to twenty million in revenue, that value marketing and have budget." Every clause there is either unverifiable or a compliment. After: "Recruitment agencies in the US or UK that sell to one named industry, show at least three recruiters on LinkedIn, and have posted a job for a recruiter in the last 60 days." A stranger can run that. A scraper can run most of it.
Write the disqualifiers at the same sitting, in the same voice, because the two halves are one document. And keep the offer in view while you do it: a precise profile pointed at a promise a stranger cannot price will still fail, which is a separate problem covered in why your agency's offer does not convert on cold outreach.
Which signal tells you the pain is live this month?
A signal is the event that makes your offer urgent now rather than interesting in general, and it is the clause that separates a list from a database export. Structure tells you a company could buy. A signal tells you something changed.
The signals worth building against share one property: they imply committed budget or a fresh wound. An open job posting is the strongest of them, because a company advertising for a role has already approved the money and the pain is on somebody's desk this week. Leadership changes work the same way, since a new head of growth is hired to change something in their first ninety days. Funding, a new office, a product launch, a public hiring freeze and visible engagement with a topic on LinkedIn all qualify. Pick two that map to your offer rather than ten that map to nothing, and go deeper on sourcing them in how to build a lead list from buying signals.
Write to the implication, never to the signal itself. "I saw you are hiring a recruiter" is a parlor trick and every founder has now received forty of them. "You are investing in outbound headcount, which usually means the pipeline is the constraint" is the same observation turned into a reason the message exists. RSW/US reaches a similar conclusion from the other end: facing 46 percent of agency leaders reporting prospects go dark after a first or second meeting, up from 16 percent a year earlier, its advice is smaller researched lists with a specific trigger per prospect rather than more volume.
The signal clause is also the part of the profile that ages fastest. Job boards change, funding cycles cool, a competitor floods the same trigger. Expect to replace it once or twice a year while the structure test holds for longer.
What belongs in the do-not-contact list?
Disqualifiers remove more waste than any positive filter adds, so write them before the list and apply them before you pay for a single email address. The useful ones are verifiable conditions that make a sale impossible or a delivery miserable, not a hunch about culture.
Six that earn their place for an agency selling services:
- Companies that sell the same service you do. Competitors read your copy and take your angle.
- Companies the founder no longer runs. A hired CEO or managing director day to day means the person who feels the pain cannot sign.
- Rosters made entirely of enterprise clients. The sales cycle outlasts your patience and your cash.
- Anyone who signed a twelve-month contract with a provider like you last quarter. The existence of a provider is not a disqualifier, a contract with ten months left is.
- Companies below the economic floor your price needs. Write the floor as a visible proxy, since the revenue figure you would prefer is not public.
- Whatever your own churn data says. The client type you fired twice is a disqualifier with evidence behind it.
Compare that against the generic set most templates hand out: too small to afford it, compliance you cannot meet, procurement too slow, no clear decision maker. Those are categories, not tests. Instantly's agency example does better by naming specifics, including deal sizes under five thousand and sales cycles past eighteen months, and it adds one that deserves an argument: it disqualifies companies with an in-house sales development team. For a tool vendor that makes sense. For an agency selling an outcome, an existing in-house team is often the buyer, because somebody there is already accountable for a number they are missing.
How many companies does an ICP need to support outbound?
Roughly 1,000 to 20,000 reachable companies is the band where systematic outbound works, and the arithmetic that gets you there runs backwards from meetings. Below the floor you recycle the same names every quarter. Above the ceiling you are describing a category and your copy goes general to cover it.
Start from the only number you control: contacts in the pipe per month. Here is the same 1,500 contacts at four different quality levels, using Smartlead's percentile bands for replies per contact, a positive share of 30 to 45 percent of replies, and two thirds of those positives turning into a meeting that happens.
| Monthly contacts | Replies per contact | Replies | Positive replies | Meetings held |
|---|---|---|---|---|
| 1,500 | 0.74% (median sender) | 11 | 3 to 5 | 2 to 3 |
| 1,500 | 1.43% (top quartile) | 21 | 6 to 9 | 4 to 6 |
| 1,500 | 2.63% (top decile) | 39 | 12 to 18 | 8 to 12 |
| 300 (profile too tight) | 2.63% | 8 | 2 to 4 | 1 to 2 |
Read the last row carefully, because it is the failure nobody warns about. A beautifully precise profile with 300 reachable contacts a month produces one or two meetings even when the targeting is top-decile. Precision has a floor, and under it the answer is not better copy but a second signal, an adjacent audience or a different motion entirely.
Now work it against a goal. Say you are at $120,000 a month and you want two new clients a month at a $4,000 average. If you close one in four meetings that happen, that is eight held meetings, which the table puts at top-decile reply rates on 1,500 contacts or median rates on something closer to 4,500. The discovery at this point is usually that the meeting target implies either a much sharper profile or three times the list that was planned, and that choice is cheaper to make now than in month three. The sending capacity that supports those volumes is its own calculation, covered in how many domains and inboxes you need for cold email, and what the resulting meetings cost sits in what a booked meeting should cost.
How do you turn the ICP into a verified list?
The profile becomes a list when every clause maps to a filter you can run or a page you can scrape, and the clause that fits neither is the one worth building around. The setup that holds up at volume is custom scrapers plus intent signals, with waterfall enrichment and verification behind them, because the signal clause is almost never a field in a database.
| Job | Tool | List price | Pick it when |
|---|---|---|---|
| Signal-based lists | Custom scrapers on Apify plus job boards and saved searches | Starter $19 a month, Scale $199, with the plan price doubling as usage credit | Your profile has a clause no database field covers |
| Search and firmographics | LinkedIn Sales Navigator | Core $119.99 a month, Advanced $159.99, per seat | You want the freshest company and people data and will work by hand |
| A closed database | Apollo | Basic $49, Professional $79, Organization $119 per user a month billed annually | You need volume this week and can accept contacts others are mailing too |
| Enrichment and orchestration | Clay | Launch from $167 a month, Growth from $446, data credits from $0.05 each | You want waterfall enrichment and the whole flow in one place |
| Verification | A standalone verifier, before any send | Cents per address | Always. Verified lists reply at roughly twice the rate of unverified ones |
Three shapes, depending on where you are. Lean: Sales Navigator plus manual exports plus a verifier, under about $200 a month, which is the right answer while you are still testing whether the profile is real. Standard: one scraper budget on Apify for the signal, a waterfall of two or three email finders, a verifier, and a sender, which lands near $300 to $400 a month at agency volumes and is itemized in how to build an AI outbound system for a B2B agency. Multichannel: add Sales Navigator seats and a LinkedIn sender once email alone stops carrying the number.
The order inside the pipeline matters more than the tools. Qualify the company against the structure test and the disqualifiers first, then find the decision maker, then pay for the email, then verify it. Reversing those steps is how a $300 enrichment bill turns into 400 addresses at companies you would have rejected on sight.
How do you tell a broken ICP from broken copy?
Split the list into segments before you send, because one blended reply rate cannot tell you which of your two guesses was wrong. Three segments, the same sequence and the same copy in each, one variable changed: who is in it.
Give each segment about 1,000 contacts before you read anything. At a 0.74 percent reply rate, 200 contacts produce one or two replies, and two replies cannot distinguish a good audience from luck. Belkins applies a minimum of 20,000 emails before it will publish an industry figure and flags its own Sunday number as statistically noisy at 21,000 sends. Your patience has to scale down from there, not up.
Three readings cover almost every case. A reply rate in the normal range with positive replies under about 15 percent of the total means people read it and shrugged, which points at the offer or the audience rather than the subject line. Near-zero replies across every segment points at deliverability or dead addresses, and no amount of profile work fixes it. One segment replying at two or three times the others is the finding you were looking for: cut the losers, keep that clause, and change exactly one thing next.
Compare your segments to each other and never to a published average. Instantly's template tells you to review the profile quarterly against closed-won and closed-lost deals and to sanity check it by list size, which is sound and slow. The faster signal is already arriving in your inbox every week, segmented, if you set the campaign up so it can be read.
Where a tight profile is the wrong move
A tightly filtered profile is the wrong tool when the reachable set is under a few hundred companies or when one deal is large enough to justify research on every account. In those cases the honest answer is account-based selling: name 150 companies, learn something real about each, and reach them through referrals, events and warm paths instead of a sequence.
The same applies before you have clients. With fewer than five engagements you would repeat, you are not filtering a market, you are guessing at one, and interviews will teach you more in a week than 2,000 sends will in a month. Outbound rewards precision applied to a population. It punishes precision applied to a hunch.
What to do next
- List the last ten clients you would take again and fill the four columns: what was broken, what event happened that month, who signed, what made them cheap to deliver for.
- Write the structure test as one sentence of pass-or-fail clauses, then hand it to someone outside your business with twenty companies and compare verdicts.
- Write six disqualifiers, including at least one drawn from a client type you have already fired.
- Pick two signals that imply committed budget, and check they are scrapable before you commit to them.
- Count the reachable companies. Under 1,000 and you need a second signal or an adjacent audience. Over 20,000 and you need another clause.
- Build three segments of about 1,000 contacts, send the same sequence to each, and read reply rate and positive share per segment at the end of the month.
If the segment test says the targeting is right and the list is the constraint, Ourix installs the outbound system inside your business and runs it, with the tools included (on Ourix's accounts by default, moved to your own at no charge whenever you want), verified lists of the buyers you want, outreach on email and LinkedIn, every reply in one inbox and a pipeline view of what got booked. Campaigns go live in about 30 days and all of it is yours to take over at any time. Ourix does not publish prices: the exact number comes on a free growth audit with Jeff, who says so plainly if it's not a fit, and there is no lock-in.
Key takeaways
- Who you write to moves reply rate by about 3x on company size and seniority and by roughly 10x across industries, which is more than any subject line will do.
- Industry, headcount and revenue are the filters every database already offers, so an ICP built only from them points you at the same contacts your competitors are mailing.
- A usable ICP has four parts: a structure test, a named buyer, a signal that the pain is live this month, and disqualifiers written before the list.
- If your ICP yields fewer than about 1,000 reachable companies it is an account-based problem, and if it yields hundreds of thousands it is a category rather than a profile.
- Validate by splitting sends into segments of about 1,000 contacts and comparing their reply rates to each other, never to a published average.
- Published reply rates disagree by tenfold because some divide replies by opens and others by emails sent, and 5 percent against openers can be the same campaign as 0.45 percent against sends.
Questions people also ask
What is the difference between an ICP and a buyer persona?
The ICP describes the company and the persona describes the person inside it. For outbound you need both, in that order, because the filters that build a list are company-level: industry, structure, team size, a signal. The persona decides which job title you look for once the company passes, and what the first line of the message talks about. A persona with no company filter in front of it produces a list nobody can qualify.
How many companies should be in my ICP?
Between roughly 1,000 and 20,000 reachable companies is the range where systematic outbound works. Below about 1,000 you cannot keep a list refreshed without recycling the same names every quarter, so name the accounts and research each one instead. Above 20,000 you are describing a category and your messaging will be too general to earn a reply. Count the companies you can actually find and contact, not the size of the market.
How do I know if my ICP is too broad?
Two tells. The first is that your list filters are all available in any database, which means every competitor is mailing the same companies. The second shows up in the data: a reply rate in the normal range with positive replies under about 15 percent of the total means the message reaches people who read it and shrug. Split the list into segments and send the same copy to each. If one segment replies at three times the rate of another, the broad version was hiding it.
Should revenue be part of my agency's ICP?
Use it as an inference, not a filter. Private company revenue is unavailable or wrong in most databases, and a figure you cannot verify will quietly throw away good companies. Headcount, the number of people in a visible delivery role, client logos, pricing pages and open job postings all tell you more about whether a company can pay you. Keep a price floor in your head and test it on calls.
How often should I rebuild the ICP?
Review it every quarter against closed-won and closed-lost deals, and rebuild it when a segment test shows one group outperforming the rest by more than double. The clause that ages fastest is the signal, because job boards, funding patterns and hiring cycles move. The structure test and the disqualifiers are stable for a year or more. Change one clause at a time so you can tell which change did the work.
Can I run cold outbound without an ICP?
You can send, and the numbers show what it costs. The median cold sender gets one reply per 135 contacts while the top 10 percent get one every 38, and the main difference between them is list selection rather than copy. Sending to a loose list also burns domain reputation and spam complaints you then have to recover from, so the untargeted version is slower and more expensive than the hour it saves.
Does Ourix define the ICP for you?
Ourix builds the outbound system inside the client's business and verified lists of the buyers they want are part of it, so the buyer definition is agreed with the founder before anything sends. The lists, the domains, the accounts, the inbox and the pipeline view all belong to the client. Ourix does not publish prices: you get the exact number on a short call with Jeff, who says so plainly if it's not a fit, and there is no lock-in.
Sources
- What are B2B Cold Email Response Rates? (2026 Study), Belkins, 2026-06-26
- Cold email statistics and benchmarks for 2026 (The State of Cold Email), Smartlead
- What is a good cold email response rate?, Smartlead
- Cold Email Statistics Based on Sending Over 20M Cold Emails, Woodpecker, 2026-06-23
- The RSW/US 2026 Agency New Business Report, RSW/US, 2026-10-15
- Silence Isn't a No: Why Prospects Go Dark, and How to Respond, RSW/US, 2026-11-05
- 2026 State of Digital Services: Digital Agency Industry Research, Promethean Research, 2026-03-01
- Ideal Customer Profile Template for Outbound Sales Teams, Instantly, 2026-05-08
- Compare Sales Navigator plans, LinkedIn Sales Solutions, 2026-09-27
- What Does Apollo's Pricing Actually Include?, Apollo
- Clay pricing, Clay
- Apify pricing, Apify
