How to define an ideal customer profile an agent can use

An ideal customer profile an agent can use has five parts: one sentence that describes the buyer, fit rules that can be checked on public pages, the events that make the need urgent this month, the companies to exclude, and proof you can defend. Each part must be concrete enough for software to apply without guessing. A vague profile produces a long list; a precise one produces a short list worth writing to.

Most ideal customer profiles are written for a slide, not for a search. Mid-market companies that value innovation and want to scale. Every word is positive and none of them can be checked. Hand that sentence to a person and they will fill the gaps with judgment. Hand it to an agent and it will fill them with guesses.

This guide shows how to write a profile that works for both: specific enough that software can apply it consistently, and honest enough that the messages it leads to are relevant to the people who receive them.

Why most profiles fail an agent

A profile fails in one of three ways. It uses adjectives instead of rules, so nothing can be checked. It describes who could buy without saying when, so every company that fits looks equally urgent. Or it says nothing about who to avoid, so the list includes current customers, competitors and companies that asked to be left alone.

The right-hand column is less inspiring and far more useful. Every line can be confirmed or ruled out from a company's own pages, a job post or a licensed dataset. That is the property an agent needs, and it is also what makes a list defensible when you look back at why a company was on it.

The five parts of a usable profile

The rest of this guide takes each part in turn.

Part 1: the buyer in one sentence

Start with a single sentence in this shape: a role, at a type of company, in a situation. The role tells the agent whom to look for inside the company. The type of company narrows the search. The situation is the part most profiles leave out, and it is what turns a category into a customer.

A useful test: give the sentence to two people separately and ask each to name ten companies. If their lists mostly overlap, the sentence is specific enough. If they do not, the sentence is still doing the job of a slide.

The situation clause also does legal work. Data protection rules in several jurisdictions turn on whether a message is relevant to the recipient's job. France's CNIL allows B2B email prospecting without prior consent when the subject relates to the recipient's profession, and Canada's regulator says implied consent from a published address depends on the message being relevant to the person's role. A buyer sentence that names the role and the problem is a written record of why each message was relevant.

Part 2: fit rules a machine can check

Fit rules decide which companies could buy. The discipline is simple to state and hard to keep: every rule must be checkable from evidence you can actually get. If a rule depends on information nobody publishes, it is a hope, not a rule, and an agent will either ignore it or guess.

Keep the list short. Three to five fit rules is usually enough, and each additional rule removes companies for reasons that become harder to explain. In Snaut's scoring, company fit is one criterion out of six and worth 25 of 100 points: industry match 10, size in range 8, geography 5, and 2 for not being excluded. Fit matters, but it is deliberately not enough on its own, for reasons the next part explains.

Part 3: timing, because most of your market is not buying

Fit tells you who could buy. It says nothing about who will buy this month, and for most businesses that is the more important question. The marketing scientist John Dawes made the point memorably in The 95:5 Rule. Corporations change providers such as their main bank or law firm around once every five years on average, he writes, which means only about 20% of business buyers are in the market over a whole year, and something like 5% in a quarter.

Dawes is careful to call the 95% figure a heuristic, not a precise rule, and he gives a way to estimate your own version: if your category's average time between purchases is two years, about 50% of buyers are in the market over a year and around 13% in a quarter. Either way, the implication for prospecting is the same. Most of the companies that fit your profile are not looking right now, and a message that arrives at the wrong moment is ignored no matter how well it is written.

There is an important counterpoint in the same article, and it would be dishonest to leave it out. Dawes argues that buyers mostly rely on memory and prefer brands they already know, so a business that only targets in-market buyers never builds the familiarity it needs. Timing signals make outreach relevant. They do not replace being known in your market, and a profile is not a substitute for a reputation.

Write the timing events down

A timing event is a public change that creates the problem you solve, or makes it worse. Write down three to five that follow directly from your buyer sentence:

• Hiring for a role your service replaces or supports. A company pays for a role only when the work has outgrown the team.
• Opening a location, entering a market or launching a product, which creates work before the team exists to do it.
• Complaints in public reviews about the exact thing you fix.
• A public request for a recommendation in your category.
• A funding announcement, which creates budget that did not exist last quarter.

Each event should be specific enough to quote. "Growing" is not an event. "Posted two dispatcher roles this week" is. We go deeper on which events carry real information in buying signals that actually mean something.

Timing events also need an expiry. In Snaut, a signal counts at full weight for its first seven days and then decays until it expires, when it stops counting at all. Evidence from careers pages and job boards expires after 14 days, and news, reviews and ordinary web pages after 30. A hiring post from last quarter has been filled or abandoned, and treating it as news is how outreach starts to sound automated.

Part 4: who must never be contacted

Exclusions are the cheapest part of a profile to write and the most expensive to forget. Contacting an existing client with a cold pitch, or a competitor with your playbook, is the kind of mistake people remember. Write the list before the first search:

• Current clients and anyone with an open deal or active conversation.
• Competitors, including companies that resell a competing product.
• Partners and suppliers, who should hear from you through the relationship you already have.
• Anyone who has opted out or asked not to be contacted, permanently.
• Kinds of business you cannot serve well, however good the fit looks on paper.
• Places where you have not confirmed the rules. Snaut flags recipients in Germany and Canada for review, because both are consent-first for email marketing.

In Snaut, excluded accounts score zero on the exclusion line of company fit, and the send gate refuses any message to an account marked as a client, competitor or partner. Suppressed people are blocked at the send gate too, and suppressions are kept forever. The legal background is in our guide to whether cold email is legal.

Part 5: proof you can defend

The last part answers the reader's real question: why should I believe you? Proof is a specific result with a source you could show if asked: a case study page, a public review, a testimonial you have permission to use. It is what an offer stands on.

Snaut treats proof strictly because it is where outreach most often goes wrong. An OFFER sentence in a draft may include a number only if that number appears in a proof item you entered with a source address. An unsourced "we cut costs by 30%" is removed before you see the draft. We explain the mechanism in our article on AI sales emails that make things up.

When you do not have results yet

Many businesses using a tool like Snaut are early and have no case studies. The answer is not to borrow someone else's numbers or to promise results. It is to describe what you do, concretely, and leave results out until you have them. "We run overnight dispatch for regional carriers" is an honest offer with no claimed outcome. It is also more persuasive than an invented statistic, because the reader cannot catch it being false.

No proof is a weaker position than real proof. Invented proof is weaker than none.

Profile, persona, segment: three different documents

The words get used interchangeably, and the confusion produces profiles that cannot be acted on. They answer three different questions, and only one of them tells an agent what to do on a Tuesday morning.

The persona is not wasted. Once the profile has produced a company and a person, the persona is what keeps the message in that person's language, which is the difference between a relevant email and a relevant email that reads like it was written by someone who has never done the job. The mistake is starting there. A beautifully researched persona attached to a vague segment produces very well-written messages to the wrong people, and the writing makes the irrelevance harder to spot.

The same logic applies to a segment. It is a useful boundary for a product and a marketing budget, and it is a poor instruction for outreach, because it has no timing and no role. If a segment is all you have, the fastest improvement is to add two lines: who inside those companies feels the problem, and what public change makes it urgent this month.

How the parts become a score

Once the profile is written, an agent can score each company against it. Snaut scores every account out of 100 across six criteria, each computed from stated rules rather than a model's impression, so the score can be explained line by line:

The default threshold to pass is 70. Look at what that implies. A company that fits perfectly, with the ideal buyer, a highly relevant offer and excellent evidence, scores 25 + 15 + 10 + 10 = 60 without any observable problem or signal. It cannot pass. The rubric is built so that fit alone never wins: a company needs evidence that it has the problem, or a reason to act now, before the agent spends your attention on it.

Separate criteria also make a profile easier to fix. If the list is full of companies with high fit and low signal scores, your timing events are too rare or too vague. If signals are strong but buyer scores are low, the agent is finding the right companies and the wrong people, and the buyer sentence needs a clearer role.

A worked example

Here is a complete profile for an invented business, a bookkeeping firm that serves online stores, written in the form an agent can use:

Buyer:      Founders of online stores, 2 to 25 staff, who do their own books
            and have just started selling in a second country.

Fit rules:  - Sells physical products online (own store or marketplace)
            - 2 to 25 employees
            - Based in the US, UK or Ireland
            - No in-house finance role on the careers page

Timing:     - Announces shipping to a new country (site or press page)
            - Posts a first finance or operations hire
            - Public reviews mention late refunds or invoicing problems

Exclude:    - Current clients and anyone in an open proposal
            - Accounting firms and bookkeeping software companies
            - Stores we referred to a partner

Offer:      Monthly bookkeeping with multi-currency and sales tax handled.
Proof:      Only results with a public source, such as a client's published
            review; otherwise describe the service and claim no outcome.

Notice what is not there: no adjectives about the companies' values, no claim about market size, no promise of results. Every line is either checkable or a decision you made. An agent can apply it consistently, and you can audit any company on the resulting list against it.

Test the profile in a week

1. Before running any search, write down ten companies you would love as customers. These are your control group.
2. Run the profile and look at the first twenty companies it produces.
3. Mark each one yes or no: would you contact it? Note why for every no.
4. Check how many of your ten control companies appeared. If few did, a fit rule is excluding the customers you want.
5. Tighten the rule behind the most common no, and run it again.
6. Stop when most of the twenty are yes and a few are companies you had never heard of with a real reason to talk.

That last condition is the point of the exercise. A profile that only finds companies you already knew is a filter. A profile that finds good companies you did not know about is doing real work.

The 20-minute exercise that writes the buyer sentence

Most people write a buyer sentence from memory and end up with a category. A better starting point is the customers you already have, because they carry evidence of why they bought. Set aside twenty minutes and do this with a spreadsheet, not a brainstorm.

1. List your ten best customers of the last two years, best meaning easiest to serve and most profitable, not biggest.
2. For each, write the role of the person who first replied, their company's industry and size.
3. Write one line answering this: what had just happened at that company in the weeks before they replied? A hire, a launch, a complaint, a funding round, a departure.
4. Group the lines. Most lists collapse into two or three situations, and one of them is usually the situation you are best at.
5. Write the buyer sentence around the strongest group: that role, at that kind of company, in that situation.

The exercise produces something a category cannot: the situation that actually preceded a purchase. It also exposes a business that sells to two different buyers with two different problems. That is not a failure of the exercise; it is the reason plans include more than one profile, and the reason a single message cannot serve both.

Two practical checks on the result. If the sentence describes fewer than a few hundred companies in your market, it is a target account list rather than a profile, and the agent will run dry within weeks. If it describes tens of thousands, the situation clause is doing no work and you are back to a category. Aim for a group you could never contact all of, and never would.

The last check is honesty. Write down what you would have to believe for the situation to be true, and make sure a public page could show it. If the only way to know that a company is short of dispatchers is to call them, then the situation is not observable and the agent cannot use it. That is not a limitation of software; it is the difference between a signal and a guess.

Common mistakes

• Describing the market instead of the buyer. "Small businesses" is a market. A role, at a type of company, in a situation, is a buyer.
• Fit rules nobody can check. If no public page or dataset could confirm it, remove it.
• No timing events. Without them, every company that fits looks equally urgent, and none of them is.
• Too many profiles at once. Each needs its own offer and proof; start with one.
• Forgetting exclusions until the first awkward reply.
• Borrowed or invented proof. It is the fastest way to lose a reader's trust, and in some places it creates legal exposure.
• Treating the profile as finished. Revisit it after every few weeks of results.

Relevance is also how you stay in the inbox. Google asks senders to keep their spam rate below 0.10%, and people mark messages as spam when they cannot see why they received them. A narrow profile is the first line of deliverability, which we cover in our guide to cold email deliverability.

How Snaut uses the profile

In Snaut, the profile is built in conversation rather than in a form. The agent reads your website, proposes a buyer sentence, an offer and the proof it found, and you correct it. From then on, the profile drives everything: which signals the agent watches, which companies it scores, which person it looks for and what the first message may claim. GDPR's Recital 47 frames legitimate interest partly in terms of what people could reasonably expect, and a profile built around the recipient's role is the practical way to stay within that expectation.

Plans include a set number of profiles, from two on Starter to ten on Scale, as listed on the pricing page. For most businesses, one sharp profile beats several vague ones. The full loop, from profile to approved draft, is described in how an AI agent finds customers.

The scoring weights, threshold and expiry periods are Snaut's production defaults as of October 2026. The bookkeeping firm and the freight examples are invented for illustration.