The ICP workshops

A definition of a good customer your team wrote, tested against your own customers.

Most businesses can say who their good customers are. Very few have a written, scored profile that a salesperson, a marketer and a piece of software would all apply the same way. The workshops produce one, from evidence, with your team's judgement at the centre and the time-consuming research done for you.

You could ask an AI assistant to write your ideal customer profile in an afternoon. It'd give you something plausible and generic, because it doesn't know what your team knows.

What comes out

One written, scored profile your whole team uses.

Profile

Industrial distributors, 100 to 500 staff

What they are

  • Distributors of maintenance, safety and trade supplies into mining, construction and manufacturing.
  • Sell to trade accounts, with online ordering for repeat lines.
  • Carry other manufacturers' brands alongside one or two of their own.
  • Buy centrally, with site managers specifying for their own site.

What they are not

  • Retailers or online-only sellers with no trade accounts.
  • Manufacturers distributing only their own range.
  • Groups whose purchasing sits outside the territory.
Scoring
CriterionWeightGoldSilverBronzeCoal
Industrial customer base40Mining, construction and manufacturing all servedTwo of the three sectorsOne sector, or only resaleConsumer or none
Distribution model30Trade distributor with maintenance suppliesDistributor without trade orderingMainly own-brand manufacturerRetail or online only
Territory20Sydney and the rest of NSWNSW metropolitan onlyInterstate with an NSW branchOutside the territory
Company size10100 to 500 staff50 to 100 or 500 to 80020 to 50Under 20, or a group over 800
A made-up workshop output: one profile and its scoring model. The weights add up to 100. Your criteria and bands will be your own.
Why not just ask AI for your ICPs

An assistant gives you everyone's profile.

It knows a great deal about markets in general and nothing about your customers in particular. The workshops split the job in two – the system does the research, and your team does the judgement, in the room.

More on why

An assistant knows a great deal about markets in general and nothing about your customers in particular. It doesn't know which accounts pay on time, which buy the full range, which looked perfect on paper and turned into the worst account you've got, or why you lost the three deals you most wanted last year. Ask it to write your ideal customer profile and you get an average of every profile ever written – mid-sized, growing, values quality. Everyone's profile, which is to say nobody's.

The knowledge that makes a profile useful is in your team's heads and in your sales history, and it's never been written down. The problem is that getting it out, and checking it against what your customers actually look like, is days of research and analysis that nobody has time for. So most businesses carry on with a definition that lives in three people's heads and disagrees.

The workshops split the job in two. The system does the research and the analysis, which is the time-consuming part. Your team does the judgement, in the room, which is the valuable part. Neither half works without the other, and the judgement half stays yours throughout.

Two colleagues ranking blank sticky notes in columns on a glass wall
How it runs

Six steps, two lanes.

The system's work on one side, your team's on the other.

Who does what across the six stepsTwo lanes across six steps. The top lane, your team: the survey, then reading the analysis, then the first workshop on the descriptions, then deciding on the blind-test changes, then the second workshop on scoring and personas. The bottom lane, the system: the research of every company, the analysis, the drafts, the blind test, and the live scoring and persona drafts. The judgement stays in the top lane throughout.Your teamThe systemThe surveyan hour at mostAsks thequestionsNothingto do yetResearches everycompany, goodand poor fitReads the analysis.Merges, dropsWhich attributesseparate yourcustomersWorkshop one:the descriptions,reworded until rightDrafts a name,"what they are","what they are not"Decides on thesuggested changesThe blind test,from scratchWorkshop two:weights, tiers,personasResearches, scoreslive, drafts thepersonas
Two lanes. The research and the analysis sit in the bottom lane, which is where the days go. Every decision about what a good customer is sits in the top lane.
  1. Step 1

    Tell us about your business

    A short survey: your main customers, a few poor fits, and what makes a company worth chasing.

    More on this step

    The system: Nothing yet – a short survey asks the questions.

    Your team: Paste in the customers that make up roughly 80% of your sales, then a few companies that look similar but are a poor fit, with a line on why. Say which channels you sell into and what you look at when deciding whether a company is worth chasing.

  2. Step 2

    Every company on the list gets researched

    What each company does, its channel, its industry, who it sells to and where it is. Nothing for your team to do.

    More on this step

    The system: For each customer and each poor fit: what the company does, the channel it sits in, its industry, who it sells to, who owns it and where it is. The sources get kept.

    Your team: Nothing. This is the part that would take a person days.

  3. Step 3

    The analysis

    Which attributes separate your good customers from the poor fits, and how to segment. Your team reads it and argues with it.

    More on this step

    The system: Works out how well each attribute separates your good customers from the poor fits, recommends how to segment (by channel, say) with the reasons, shows the other groupings and labels the weak ones, and keeps the poor fits in red so they're never counted as part of a profile.

    Your team: Read it. Argue with it. A group with one customer in it gets merged with its neighbour, and a grouping the team knows is wrong gets dropped.

  4. Step 4

    Describe each profile

    A drafted name, what they are and what they're not. Your team rewords it until it reads right – the first workshop.

    More on this step

    The system: Drafts a name and two descriptions for each group from the research – what they are (what they have in common, what they do, how they use the product, how they buy) and what they're not (the companies that look similar but should be kept out).

    Your team: The first workshop. Go through each description and change the wording until it reads right. You know your customers better than the research does, and this is where that shows.

  5. Step 5

    The blind test

    Each company is researched again and placed using only your descriptions. Where the test disagrees, the definition gets fixed.

    More on this step

    The system: Researches each company again from scratch and places it in a profile using only your descriptions. The poor fits get tested too – a poor fit passes when it lands in none of them. Where the test disagrees, it suggests what to change in the definition.

    Your team: Decide on the suggested changes. In the recording, 16 of 18 matched first time and two poor fits had slipped into one group. One sentence in that definition fixed it and all 18 matched.

  6. Step 6

    Scoring and personas

    Weights, tier thresholds and personas, set by your team with your own customers scored live – the second workshop.

    More on this step

    The system: Researches the three to five attributes that separate a top customer from a marginal one, for every company in each group. Scores your own customers live as the weights and tier rules change. Pulls the profiles of the real people you name and drafts what each persona cares about and what triggers a decision.

    Your team: The second workshop. Set the weights (they add up to 100) and the tier thresholds, and try things – move a weight, drag a threshold, type in a company you know is a bad fit and check it lands at the bottom. Name the personas and tick which apply to which profile.

The blind-test figures are from the recording of a real run, with the company anonymised. Your numbers will differ – the method won't.

In and out

What your team puts in, and what it gets back.

In · from the survey

The customer list

The customers that make up roughly 80% of your sales, and a handful that looked right and weren't, with a line on why. The poor fits matter as much as the good ones, because the definitions have to keep them out.

In · two sessions

Two working sessions

One on the descriptions, one on the scoring and the personas. The people who know your customers, which means sales as well as the owner. Everything between the sessions gets done for you.

Out · written down, at last

The profiles, in plain language

Each one with what they are and what they're not, in your team's words, tested against your own customers until the test agrees.

Out · runs against any company

A scoring model

Three to five attributes per profile, weights that add up to 100, and the tier rules for gold, silver and so on, set by your team with your customers scored live. How we score fit shows what it looks like on a record.

Out · who to talk to, and why

The personas

The types of people you want to reach inside each profile, drafted from real examples you named, with what each cares about and what triggers a decision, ticked against the profiles they belong to.

Out · software and people, same list

One definition everyone uses

The same profiles drive the research, the scoring and the outreach that run for you, and give marketing, business development and sales one definition to line up behind.

Where the time goes

Days of research, none of them your team's.

Researching one company properly takes a person 20 or 30 minutes. A customer list of 60 to 100, plus the poor fits, is days of work before anyone has made a single decision. In the workshops that research runs between the sessions, without your team's time.

More on the split

Researching one company properly takes a person 20 or 30 minutes. A customer list of 60 to 100, plus the poor fits, is days of work before anyone has made a single decision, which is why it's never been done. In the workshops that research runs between the survey and the first session, and again before the blind test, without your team's time. What your team spends is the survey and the two sessions, and that time goes entirely on judgement – is this description right, does this attribute matter more than that one, is this person really the buyer.

That split is the point. The tools are good at reading 200 websites and bad at knowing what you meant by "the right kind of distributor". Your team is the other way round. The workshops put each on the part it's good at.

A working session around a table, with notes and a laptop
What happens with it next

Everything else runs on the profiles.

Your CRM gets scored

The companies already in your CRM get scored and tiered against them, so the team can see where to start.

New companies get checked

New companies get found and checked against them before anyone spends time on them.

People get checked too

The people found at each company get checked against the personas.

Outreach follows

Where outreach is switched on, it goes only to companies and people that passed.

More on what happens next

Everything else we do runs on the profiles. The companies already in your CRM get scored and tiered against them, so the team can see where to start. New companies get found and checked against them before anyone spends time on them. The people found at each company get checked against the personas. And where outreach is switched on, it goes only to companies and people that passed. How it works sets out the sequence, and the Claude or ChatGPT page explains why automation pointed at an undefined profile multiplies the wrong records.

The profiles aren't finished for ever. Customers change, markets move, and a definition that was right in March is slightly wrong by the following March. We'd re-run the blind test once a year, and whenever the business adds a product line or a market.

Five traps

Where this goes wrong.

Asking an assistant to write it

Writing the profile from nothing, and getting everyone's profile.

Skipping the poor fits

The definitions exist to keep them out, and without them nothing gets tested.

The owner doing it alone

The people who sell know things the owner doesn't, and the definition has to be one they'll use.

Twelve profiles

Most businesses have two to four, and a group with one customer in it isn't a profile.

Never re-testing

A year later the definition is quietly wrong and still trusted.

Three people talking across a table in a glass meeting room
The first step

It starts with a discovery meeting.

A conversation about your market, what's in your CRM today and what a good customer looks like for you. If there's a fit, you get a written proposal.