Customer discovery interviews, and how many are enough
B2B customer discovery interviews: the sample size the research converges on, why your segments change that number, and the questions that get honest answers.
Read
Not every account is worth the same effort, so not every account should get the same effort. Tier the list, and let the published platform limits decide how much volume the bottom tier can carry.
Personalising cold outreach at scale is usually sold as a tooling problem, and it is almost always a sorting problem. The tools will happily insert a company name into two thousand emails. What they cannot do is decide which forty of those two thousand deserve ten minutes of a human being's attention, and that decision is the whole of the result.
The useful way to think about it is tiers. Not every account gets the same effort, because not every account is worth the same, and pretending otherwise is how teams end up with uniformly mediocre outreach that nobody answers.
Split the list before you write anything. The split is by what the account is worth to you if it closes, not by how easy it is to find an email address.
Reserved for the accounts you would reorganise a quarter around. Perhaps twenty to forty of them. These get a genuine read: the last thing they announced, who recently joined in the relevant function, what their careers page implies about where they are investing, and what a competitor of theirs just did.
The message that comes out of this does not have a template underneath it. If it does, the research was decoration.
The bulk of a real target list. Here you are not writing from scratch, you are choosing. One observation specific to the account, dropped into a structure you have already proven works. The observation should be something only someone who looked would know, and it should connect to the reason you are writing.
This tier is the one people get wrong. The answer for the long tail is not thin personalisation, it is segment-level relevance with no personalisation pretended. A message that says nothing about the reader but is precisely right about their situation outperforms one that gets their company name right and nothing else.
"I noticed you are the Head of Operations at [Company]" is not personalisation. It is proof you have a database.
Inside tier one, the research itself splits into things that change what you say and things that only prove you looked.
Worth it: a hiring pattern that implies a project, a product launch that creates the problem you solve, a regulatory change that applies to their sector, a named person moving into the role that owns your category, a public commitment with a date attached.
Theatre: their podcast appearance, their post from four months ago, the city they are in, the university they attended, congratulations on a funding round that every other vendor also congratulated them on. These produce an opening line and nothing after it. The reader can tell, because the second paragraph reverts to the pitch.
The test is simple. If the research changed the offer, it was research. If it only changed the first sentence, it was theatre, and it cost you ten minutes to produce a sentence that reads like software.
The word "scale" is where this stops being a craft question and becomes a compliance one. Two sets of published rules decide what is available to you.
LinkedIn's help centre has a page on prohibited software and extensions which states plainly that "any third party software, including 'crawlers', bots, browser plug-ins, or browser extensions" that scrape, alter the appearance of, or automate activity on the platform violates the User Agreement. It cites Section 8.2, which prohibits developing, supporting or using "software, devices, scripts, robots or any other means or processes" to do so, and using "bots or other unauthorized automated methods to access the Services".
The stated consequence is that members using these tools risk "having their accounts restricted or shut down", and that prohibited tools may stop working "without notice". So an outbound motion built on a connection-request automation tool is a motion with a single point of failure that is not under your control, and the account at risk is usually a founder's.
Gmail's email sender guidelines set out requirements that took effect on 1 February 2024 for bulk senders, defined as those who "send more than 5,000 messages per day to Gmail accounts". The documented requirements include SPF, DKIM and DMARC authentication, with DKIM keys "1024 bits or longer", and one-click unsubscribe via the List-Unsubscribe-Post: List-Unsubscribe=One-Click and List-Unsubscribe headers for marketing and subscribed messages.
The number that governs behaviour is the spam rate. Google says to keep spam rates reported in Postmaster Tools below 0.30%, and the guidance recommends staying below 0.10% and avoiding 0.30% or higher.
Read that as a budget rather than a threshold. At 0.10%, one complaint in a thousand is your working allowance. That is the arithmetic case for tiering: volume without relevance spends a budget you cannot see until you have overspent it, and recovering a domain's reputation takes considerably longer than the campaign took to send.
Tiering collapses under pressure unless the system enforces it. Four things make it hold.
If you have not yet settled which accounts belong in the top tier at all, that decision comes first and is covered in account based selling for small teams.
Teams put all the personalisation into message one and then send four generic follow-ups behind it. That sequence teaches the reader that the first message was the exception.
Invert it. A shorter, plainer first message, with the specific observation held for the second or third touch, tends to survive the sequence better. The reader who ignored a good opener will sometimes answer a follow-up that adds something new, and almost never one that says "just bumping this".
How many touches is a separate question with its own answer, in how many touches to book a B2B meeting. The channel question is in cold email versus LinkedIn outreach, and the structure of the message itself in how to write a cold email that gets replies.
One more thing decides more outcomes than any of this: what happens in the minutes after someone replies. KF Agency looked at response speed against close rates, in Arabic, in their piece on lead response speed and its effect on closing. The context there is inbound rather than outbound, but the finding transfers: a carefully personalised message that earns a reply, answered two days later, has wasted the personalisation entirely.
We are not going to give you a reply rate for personalised versus unpersonalised outreach. The figures in circulation come from vendors selling sequencing tools, measured on their own customers' campaigns, with no disclosed sample and no definition of what counted as personalised. We have seen the same claim quoted at three different numbers in one week.
What we will say is testable in your own account: split one week's tier two list, send half with a specific observation and half without, and read the reply rate after two weeks. That number is small, local and yours, and it beats a benchmark.
The other limit is more uncomfortable. Personalisation cannot rescue a bad fit. If the account does not have the problem you solve, ten minutes of research produces a well-informed message to someone who still has no reason to answer. Most outreach that fails does not fail on craft, it fails on list, and the research time would have been better spent cutting the list in half.
Send three things: how many accounts are on the current list, how many replies the last hundred messages produced, and what observation the top twenty messages opened with. We will come back with which tier the list belongs in, which of those observations were doing work, and whether the fix is better writing or a shorter list.
Book a Free Consultation