Improvement·

Legatus 1.1: better facts on every playbook

An account playbook is only worth something if the data under it is right.

Legatus runs on the same models you open in ChatGPT or Claude. The difference is what it was built to do. ChatGPT is built to answer anything. Legatus is built for one job: research an account and turn it into a strategy. It knows what a company size is supposed to mean, which financial year counts, and what makes an angle worth opening with. Ask an AI to size a company and it hands you whatever it finds first, often one registered entity in one country, because nothing told it that was the wrong answer.

We tell it. Since launch we have run 100s of real account playbooks, used to open and win real accounts. At the start of September we read back through them and found every place the agent had read a fact loosely, reached for the wrong source, or built a case on a number that was no longer the newest one.

Today we ship Legatus 1.1. Twenty six of those corrections, on every playbook you build from here.

Here is what changed on the page you actually read.

Old financial numbers can no longer carry an argument. A playbook reaches for financial data to build the strategy. Maybe a company missed its own expectations. Maybe it is growing fast. Maybe the numbers make the case that now is the moment to reach out.

The research turns up several numbers, and sometimes the argument got built on the older one. It pulled from 2024 or 2023 when the 2025 annual report was sitting right there. So you send a case about a weak year to a company that has since published a strong one, and they stop reading.

The new rule: a number published more than a year ago can never carry an argument. It can only give a sense of direction. Everything you see carrying weight on a playbook is from the past year.

Company size now describes the company, not one piece of it. A playbook uses size to work out who you are approaching. A 300 person company and a 3,000 person group are different sales.

The research took size from whichever source it found first, often a national business registry. A registry only counts the entity registered in that country. So a global brand with 1,000s of employees came back as a few 100, and the strategy was written for a company that size.

The new rule: annual report first, the company's own site next, registry last.

Now the size on a playbook is the size of the business you are selling to.

Every source link opens the page the fact came from. A playbook makes claims about a company. When one matters, you click the source to see it for yourself.

About 1 link in 16 opened the company's front page instead. The research knew where the fact came from but never wrote the address down, so the playbook made one up. You clicked a claim about a hiring plan and landed on a homepage.

The new rule: the agent keeps every page it opens and cites the fact against that page. A link the research never carried does not get written.

Now every link goes where it says it goes.

The playbook thinks before it writes. Legatus works out plenty about an account before it builds the strategy. What to lead with. Who to open with. And what to stay away from, like a sore point or an angle the company is tired of hearing.

It only worked out what to avoid after the case was already written. So the avoid list came too late to shape anything.

The new rule: everything Legatus knows about the account goes in before the case gets built, not after.

Now the strategy is built on the full picture, and the case holds together with the advice around it.

You can see which sources Legatus could not read. Your Research Guide tells Legatus where to look. Industry sites, registries, wherever you know the good information sits on your accounts.

Some of those sites never open. A few block automated readers on purpose. Others sit behind a login, or a cookie wall, or answer so slowly that the request times out. You had no way of knowing, so a source you were counting on could be missing from every account and the research would look complete.

The new rule: if we try a site several times and never get an answer, we tell you it is not reaching your account research, so you can swap it for one that does. Leave it in if you want. We retry every blocked source once a month to see if it lets us in.

Now you know which of your sources made it in.

Also shipped in Legatus

This release brings Legatus to 1.1

  • More stable recognition of who still holds a role, so fewer people drop out of a playbook for no reason

  • People who have moved on stay out of the outreach and sit separately as background

  • Headcount reflects the people a company shows today, not an old filing

  • Fewer accounts come back with a gap in the research

Also shipped in the app
  • If any part of the research fails, the account keeps the research it had with a note asking you to run it again. No playbook gets built on an unfinished run

  • Far more cards show a headcount backed by a checked record

  • Revenue shows on a card only when it comes from a checked figure, and appears on more accounts as they run

  • Some accounts were processed as their truncated name, like Toogoodtogo. Now it’s verified on the account’s own website and reads right, like Too Good To Go.

  • Bug in the Intelligence tab fixed, meaning it shows on all workspaces running Danish.

  • Why now and Why us keep the short version on screen when you open the extended one on the brief page

  • An account waiting in queue says so instead of just saying “Researching”

  • When every contact on an account has left, the page says so plainly and gives you one click to report it

  • Faster research when running multiple accounts together.


← All changes