B2B account research: what to look for, and in what order

Most account research produces a document nobody uses.

Headcount, funding history, the leadership team, recent news, a list of technologies. All accurate. None of it changes what you say. The rep reads it, learns nothing they can act on, and writes the same email they would have written anyway.

The problem is not effort. It is that research is usually organised around what is easy to find rather than around what changes the argument.

The one thing you are looking for

Good research answers a single question:

> What does this company believe about its own future, and what is currently in the way?

Everything else is context for that. Headcount, funding, tech stack and news are inputs. That sentence is the output, and if your research does not produce it, it has not finished.

A rep who can answer it writes a different email than one who has a fact sheet. The first opens on something the buyer recognises as true. The second opens on a merge field.

The seven things, in order

Order matters. Each layer makes the next one readable.

1. How they describe themselves

Read their homepage, their product pages and their careers page. Not for facts, for vocabulary.

Every company has internal language for what it does. If they say "the connected back office", that is the phrase. If they say "activation" where the industry says "onboarding", use theirs.

This is the cheapest and most under-used research step in B2B. Using a buyer's own category language is close to the strongest signal of attention you can send, and it costs ten minutes.

2. What changed recently

Product launches, repositioning, a new market, an executive appointment in a relevant function, a pricing change.

Change creates the reason to talk now. A company that has not changed anything has no urgency, and urgency is what separates a reply from an archive.

Look for changes in how they *describe* themselves, not only in what they *announce*. A quiet homepage rewrite often reveals more strategic movement than a press release.

3. What they are hiring for

The most under-used public signal in B2B, and the one that most reliably produces a specific opening.

You are not looking for whether they are hiring. You are reading job descriptions for the shape of the problem. Three roles doing manual work your product automates tells you their current approach is failing, publicly, with a date on it.

Read the responsibilities section, not the title. Titles are generic; responsibilities describe the actual gap.

4. Their tech stack

Better than industry code as a predictor of fit, and it tells you two things at once: whether adjacent budget already exists, and how sophisticated they are.

Removals matter more than additions. A tool that disappeared means a contract ended or a decision reversed, and there is a gap right now. This is the signal-reading discipline covered in B2B buying signals.

5. Who is actually involved

Not the org chart. The functions that will have to agree.

You want function and veto power, not seniority. The person who blocks the deal is usually mid-level and holds a policy, not a title. The full mapping is in the B2B buying committee.

6. What they have said publicly

Conference talks, podcast appearances, engineering blog posts, executive commentary.

This is where you find what a company genuinely believes rather than what its marketing says. A CTO's conference talk about a rebuild is far more useful than any press release, because it describes a real problem in real language.

7. The moat

The thing that makes them different from their competitors. Their wedge.

This is the highest-value item and the one most research skips. If you understand a company's moat, you can frame your argument around extending it rather than around fixing a weakness. Buyers respond very differently to those two framings.

It also protects against the most damaging research error: underselling their scope. If they are an end-to-end platform and you address them as a point tool, you have announced that you did not do the homework, and no amount of accuracy elsewhere recovers it.

The 15-minute version

Full research is not always warranted. For lower-tier accounts, four steps:

1. Homepage. How do they describe themselves? Note the exact phrases.

2. Careers page. Any roles describing your problem?

3. One recent change. Anything in the last quarter.

4. One sentence. What do they believe about their future, and what is in the way?

That is enough for a specific, non-generic first touch. It is not enough for a bespoke asset, and the difference is exactly why tiering exists. The economics are in 1:1 ABM personalization.

What machines do well and badly here

The cost of this work collapsed, and it is worth being precise about which parts.

Machines do well: gathering. Firmographics, open roles, tech stack, recent public activity, vocabulary extraction, summarising public commentary. Everything in steps 1 through 6 is now minutes rather than hours, and more thorough than most humans manage under time pressure. The structured half of that is a waterfall enrichment problem.

Machines do badly: step 7. Identifying the moat and deciding which single observation to lead with is judgment. A machine will produce three plausible observations. Choosing which one lands, and which would read as obvious or presumptuous, is still a person's job.

The failure mode: treating the gathered research as the output. A ten-page account brief is not research, it is raw material. The output is one sentence and one observation worth opening on. Skipping the compression step is why so much automated research produces outreach that is comprehensive and unpersuasive.

Turning research into an opening

The rule: the research chooses the timing and the topic. Never the opening line.

Do not report what you found. "I saw you're hiring three ops analysts" is accurate and reads as surveillance. Open on the problem it implies, in their language, and let them make the connection.

• Bad: "I noticed you posted three reconciliation roles."
• Good: "Most teams at your stage hit a point where reconciliation headcount grows faster than transaction volume."

Same source, same insight. One reads as informed, the other as watched. What gets built on top of that opening is covered in the B2B pitch deck that gets read.

Frequently asked questions

What should you research before contacting a B2B account?

In order: how they describe themselves and the vocabulary they use, what changed recently, what they are hiring for, their tech stack, which functions will be involved, what they have said publicly, and their competitive moat. The moat is the highest-value item and the one most commonly skipped.

How long should account research take?

For a top-tier account, as long as it takes to answer what the company believes about its future and what is in the way. For lower tiers, fifteen minutes covering homepage vocabulary, careers page, one recent change and one summarising sentence is enough for a specific first touch.

What is the most useful thing to find in account research?

The company's moat, meaning what makes it different from its competitors. Understanding it lets you frame your argument around extending a strength rather than fixing a weakness, and it prevents the damaging error of addressing an end-to-end platform as though it were a point tool.

Are job postings useful for account research?

Very. Read the responsibilities rather than the titles. A role describing manual work your product automates is public evidence that their current approach is failing, with a date attached. It is the most under-used public signal in B2B.

Can AI do account research?

It handles gathering well: firmographics, open roles, tech stack, public commentary and vocabulary extraction now take minutes. It handles judgment poorly, specifically identifying the moat and choosing which single observation to lead with. Treating the gathered brief as the finished output is the common failure.

---

*NomiOS is RZLT's GTM and ABM engine. Point it at a target and get back finished, branded work built on a real read of that company.*

[See how NomiOS works →](https://nomios.rzlt.io)