It sounds like one request. It is four, and the easy one is the least predictive.
Every business has a favourite customer. Pleasant to deal with, pays on time, uses the thing properly, occasionally says something nice in public. The obvious instinct is to want ten more of them, and the obvious next step is to ask for a list of companies that look like them.
Then the list arrives and it is disappointing in a specific way: every company on it is plausible and none of them is interesting. That is not a failure of the tool. It is a consequence of which resemblance got measured.
The four resemblances, from least to most useful
They look alike on paper. Same industry, similar headcount, same city. This is the one that gets searched for, because the data is public and the filter is easy to write. It is also the one every competitor can run, so the result is a list several other vendors are already working, and nothing in it suggests anybody has a reason to buy anything this quarter.
They buy for the same reason. Your best customer came to you because something specific had happened: they had grown past a tool, lost someone who held everything in their head, been asked for reporting they could not produce. That reason is the actual pattern, and it lives in your own records rather than on anybody's website.
They decide the same way. Same number of people in the room, same length of cycle, same person signing. This decides whether your sales process fits, which is why two firmographically identical companies can be an easy close and a six-month exercise in futility.
They produced the same outcome. Not similar to the customer, similar to the result. This is the one almost nobody asks for, and it is often the right question, because what you want more of is the engagement that went well, not the logo that resembles it.
First, fix the definition of best
Ask three people in a business to name the best customer and you get three answers: the biggest, the nicest, and the one somebody personally won.
None of those is a definition you can search against. A usable one names what you want more of, and in most businesses that is some combination of profitable at the price you charge, cheap to serve after the first ninety days, unlikely to leave, and willing to be a reference.
Run that definition over your existing customers and the result is frequently uncomfortable. The biggest account is often the least profitable, because it was won on price and consumes support out of proportion to what it pays. The best customer by this measure is usually a mid-sized one nobody talks about.
That reordering is the most valuable thing in this whole exercise, and it happens before any search runs.
Write the description you would disagree with
The test of an ideal customer profile is whether it excludes somebody.
"Growing Australian businesses that want to work more efficiently" excludes nobody and therefore describes nobody. A profile that earns its keep says something that would cause an argument internally: a size range with a bottom to it, an industry list that leaves out a sector someone is fond of, a trigger that rules out companies who are merely curious.
If a sales team reads the profile and nobody objects, it has not been written yet.
What the technology can and cannot do
Here the honest answer is more useful than the enthusiastic one.
What works well: searching against a written description. Describe who you sell to and why they buy, in plain words, and have something go and find matching companies continuously. The reason this works is not clever matching. It is that writing the description is the step most businesses have never completed, and the search forces it.
What generally does not exist yet: true similarity search over your own records, of the kind people imagine when they picture pointing at a customer and saying "more of those". Most CRM search is filtering, not similarity. HubSpot, to their credit, document this plainly for their own MCP server: search results come from the CRM search API, which does not include vector search, and the query tool supports no joins and no association traversal. So "find me companies like this one" asked of a CRM assistant is answered by a filter you could have written yourself, built from fields somebody remembered to fill in.
Which is worth knowing before you plan a process around it, and is covered in more detail in what HubSpot's MCP can and cannot do.
Why it has to be standing rather than one-off
The list you want does not exist on any single day.
A company becomes a good prospect when the trigger fires: they hire the role, open the site, lose the person, win the contract that makes the current approach untenable. Ask for ten companies today and you get the ones who matched today, which is a fraction of the ones who will match this quarter.
A standing search solves this by changing what you are asking for. Instead of a list, you are asking for a notification. The companies arrive over time, a few at a time, each one new since yesterday, which is both a better input to a sales week and considerably easier to act on than a spreadsheet of two hundred.
This is what Daily Spark does in Signal: describe your ideal customer once, and the matching companies arrive each morning with the decision-makers inside them, in the same system that holds the deal. Nothing enters the pipeline until somebody accepts it. The on-demand version of the same search is Prospect Spark, for when you want to go looking rather than wait.
The part that matters for this post is not the email. It is that a company arriving because it matched a description you wrote comes with the reason attached, and the reason is the first line of the first email. Which is also the difference between a personalised email and a decorated one.
What to do this week
- Define best properly. Profitable, cheap to serve, likely to stay, willing to refer. Rank your customers on it and see who moves.
- Read the top five and find the trigger. Not what industry they are in. What had just happened to them.
- Write the profile until somebody objects. The objection is the signal that it says something.
- Make the search standing rather than one-off. Then measure which trigger produces the conversations, and narrow it next quarter.
The request "find me ten prospects like my best customer" is a good instinct pointed at the wrong variable. The resemblance that matters is almost never the one you can see from the outside.
