What trustworthy account intelligence looks like
A rep opens a call by congratulating a director on a role the director left four months ago. Another names a tool the company retired last year. A third quotes a funding round as news when it closed before the new year. None of them did zero research. They did research, trusted it, and it was wrong.
One wrong detail costs more than no research. A rep who knew nothing can ask questions. A rep who is confidently wrong has to be corrected by the buyer, and the buyer remembers it. This article gives you a five-point checklist for judging any account research, whether it comes from a tool, a teammate or your own notes. Each point has a test you can run in about a minute.
The five-point checklist
- Every fact has a source and a date.
- Guesses are labeled as guesses.
- The research is about the right company.
- Each person named still works there.
- Gaps are shown, not hidden.
1. Every fact has a source and a date
A fact with no source is a rumor, and a fact with no date might be last year's news. You need both, so you can judge for yourself how much weight to give it.
One-minute test: pick three statements at random from the research. For each one, can you open the source in one click, and does the source actually say it? Does it show a date? If any of the three fails, treat the whole document with more caution.
2. Guesses are labeled as guesses
Research mixes what a source states with what the author concluded from it. Both are useful. Only one should look like a quote. "They use Kubernetes" is a fact if the job post says so. "They are probably migrating off their old tool" is a conclusion, and the page should say so.
One-minute test: find a sentence that sounds certain and compare it to its source. If the source only hints at it, the sentence should have been marked as inferred.
3. It is about the right company
Many companies share a name, and a domain can lead to a different business than the one you meant. Research on the wrong company is the most expensive mistake on this list, because everything looks fine and every detail is wrong.
One-minute test: check the name and website at the top of the research against the account you meant, and confirm the product description matches what you know. Do this first.
4. Each person still works there
People change jobs all the time. A contact record is a snapshot from the day someone last collected it. Contact databases go stale because nobody is paid to notice when one person moves, and the record doesn't know it is out of date.
One-minute test: for each person you plan to contact, find one public item from the last few months that places them at the company, like a post, a talk or the company's team page. Check whether the record shows a "last confirmed" date, and do not trust it if it doesn't.
5. Gaps are shown, not hidden
Some things can't be found. Research that hides the gaps fills them with plausible text. Research that shows them tells you what to ask.
One-minute test: look for a place where the research admits it found nothing. A document about 15 accounts that never says "I couldn't find this" has probably smoothed over something.
Why contact data goes stale, and what to check
A contact database is a copy. Someone collected each record once, and the copy ages from that day. Titles change, people leave, teams are reorganized, and the record stays as it was until the next refresh. Two things tell you how much to trust one:
- The last-confirmed date. Some tools show when a record was last checked. A record with no date gives you no way to judge its age.
- A current public trace. A recent post, a conference session, a commit or a line on the company's team page shows the person at the company on a known date. That is evidence you can cite, which a database row is not.
The same applies to a title. A person promoted last month may still show the old title in your list. Read it as "was a director at some point" until you see it somewhere current.
Why general chat assistants need your context
A general chat assistant writes well and can search the web. That makes it a useful research partner, but it has limits. It knows only what you paste in and what its search happens to find. If you ask about an account without giving it your sales profile or a list of sources to use, it may answer with something plausible that isn't tied to evidence.
You can get far better results by giving it three things: a description of who you sell to, the account and its domain, and an instruction to list a source and date for every claim and to mark anything it inferred. Then run the checklist above on what it returns. The same rules apply to every tool.
How Repclose handles each point
Here is what Repclose does today for each point, and where it stops.
| Point | What Repclose does | What it does not do |
|---|---|---|
| Source and date | Each point links to the sources behind it, with the publication date where one is available. A source with no date is marked "undated". | It can't date a page that doesn't state a date. Open the link and read the source. |
| Guesses | Points the model concluded from a source, rather than quoted from it, are marked as inferred, with a dashed source link or an "Inferred" tag. Your own notes show as "Your note", and what a customer told you shows as "Customer said". | The label tells you to check. It doesn't confirm the point. |
| Right company | It researches the domain you give it. | It does not ask you to confirm the company. You check the name and site at the top of the plan. |
| People still there | For your top-ranked accounts it lists about 10 people each and runs a profile check on the top 5. | A profile check is not a promise. Confirm before you write. Anyone can ask not to be suggested. |
| Gaps | An empty section says "Not found yet" and suggests what to ask. | It can't know about a source it never reached. |
Each researched point also carries a confidence word, high, medium or low. It comes from how many dated sources back the point, how recent they are, and whether the point is stated or inferred. It does not measure whether the point is correct. A recent source can still be wrong.
Some limits are plain. Repclose results can be wrong. Repclose has no security certifications or audits yet, and the security page lists what it does not have. For people data, Repclose keeps work information, such as name, title and employer. Research results are deleted 365 days after they are made. The privacy page has the details, including how people opt out.
Audit your own brief in one minute
Take the last account brief you wrote, or the next one you read, and do this:
- Circle the three claims you would be most embarrassed to get wrong on a call.
- For each, write the source and its date next to it. If you can't, cross it out.
- Mark every remaining sentence that is a guess with the word "Guess".
- Check that the company and the website match the account you meant.
- Check one current item that places each person at the company.
- Add one line at the bottom: "Not found yet:" followed by what you still don't know.
What's left is a smaller brief, and one you can defend. If you want to see what a finished page looks like, open the sample brief. To run your own accounts, start here. For the method behind the brief, read how to research a technical account.