Reading your company results

The one results page: what a company admin sees, what a manager sees, per-person levels, and the five-person floor on individual answers.

The AI literacy check’s results open when an admin sends them. While a round of it is waiting to be sent and at least one person has finished, its pages show no figures: not the company level, not the team lines, not the charts, not the per-person levels. You still see who was sent it, who has finished and who has not.

A survey is not held. It shows its results as people finish, because Send results belongs to the round card the AI literacy check runs on and nothing else has one.

A total moves by one each time somebody finishes, and Levels by person says who that somebody was, so a total you can watch while people finish is one person’s own result read off the difference. Sending the results opens all of it at the one press, at the same moment the people who took it can open their own reports. See sending an assessment.

Both a company admin and a manager can read results. They do not see the same thing. A company admin sees the whole company. A manager sees only the people who report to them, directly or further down, aggregate figures plus their levels.

Your results start on Home, the first page on the rail. Home reads the newest round that went out. Once that round has results, Home shows it in short: a row of figures, the first three Next moves with an owner (on a Know Your AI round from October 2026 on, What to change next instead), and the Lowest safety checks. For an admin, the top bar names the round and its state, and links to Rounds, where you send, remind and close. Before anything has been sent, Home is one card with one action instead.

The row of figures always has people asked out of your roster, company level, and how many people are at L3 or above. It adds safety checks under 80% once any safety check has enough answers, policy rules with a figure when the round asked about your policy, and, for an admin, AI tools with a use policy once you have AI tools listed. A manager reads your team’s level in place of the company level. The level needs five graded answers. Until then, and while the results are held, Home shows who has answered instead. A Know Your AI round sent from October 2026 on has no safety checks. It asks about your rules as situations, and each rule gets its own line under the policy rules. On those rounds the row shows the share of people at L3 or above in place of the company level and the count at L3 or above, and no average level appears anywhere.

Open Round N results, at the top of Home, opens that round’s own results page. Report on a closed or past round on Rounds opens the same page, and so does the link in the email you get when a round closes. Its top is a summary you can read in one screen: a one-line headline, the same row of figures, the parts of the check, who was asked in each team, and every next move with an owner. Under it, What each line means opens each line into the detail behind it: who was asked, the safety checks with the answers people gave, levels, the policy rules the round asked about, and, for an admin, your AI tools against your policy. On Know Your AI each policy rule’s figure is the share of people who picked what the rule says in its situation, and “I’m not sure” counts against it. A rule the round did not ask as a situation says “Not asked as a situation this round, so there is no figure.” Your two ground rules, on personal data and on where a wrong answer goes, are listed under the policy rules as rows of their own. A round sent when your company had no AI rules asked about skills only. It shows one line, Your AI rules, in place of the safety checks and the policy rules, and its Article 4 record has no policy part. Under Levels and your Builders, See levels by team and since the last round opens the company report, which has each team’s level and who moved since the last round.

The AI literacy check has an extra page of its own, Levels by person, reached from See each person’s level under Levels and your Builders on a round’s results page. Every company has that too.

The top of a Know Your AI round from October 2026 on

The round’s results page opens on these blocks instead of the summary above.

What leadership estimated. Admins only, and only if the round has estimates from Rounds. A dark band puts each estimate beside what your people answered:
- How many know where the AI rules are. If the round went out when you had no AI rules, it is how many answered that you have none, and a second line gives the share who think you have some.
- How many tools that five or more people picked are not on your list, as on Where AI is involved.
- The share at L3 or above.
- The share of last week’s tasks people did with AI. Each person’s task counts once.
An estimate gets no card while fewer than five people are behind its answer. With no estimates, there is no band. If the quick test on the Velory homepage filled in your first round’s estimates and nobody saved the estimates form before the send, those cards say “From your website check”.
Your rules, in practice. For each rule, the share of people who picked what the rule says in its situation and, where a team can be named, the lowest team. “I’m not sure” counts as not knowing the rule. Your two ground rules and Knows where the rules are are rows here too. A rule from the Flow template is tagged Flow template. A bar in amber is under 80%. Under the rules, one line gives the share of people who say their team puts work information into AI tools that aren’t approved.
What your people can do. The share at L3 or above, for the company and for each team. Hover or tap a share to see the count behind it. A team under five answers is folded into the team above it or into Other teams.
Where AI is involved. Admins only. The AI tools people picked in the first question, how many picked each, and whether each is on your list. When someone types a tool under “Something else”, it shows by name only when it matches a tool Flow knows. Anything else is counted under Named, not on the list, and nobody sees what was typed. A tool picked by fewer than five people is not named: its people are counted together under Other tools.
How work gets done. The tasks people picked as part of their job, and how they did them last week. The company line splits them into By hand, With AI and Automated (AI did it and the person checked it). Each team’s line has two parts, by hand and with AI, where with AI includes automated: automated work is rare, and a team line showing none of it would say that about every person in the team. A manager’s line for their own branch has the same two parts, and when every task in the branch was done one way it shows no figure. It shows an estimate (“about 6 in 10”), never hours or an exact percentage. A team needs five people who did a task last week. When every task in a team of five or more was done by hand, or every one with AI, no team is shown, only the company.
What to change next. Up to three things to act on, each with the numbers that put it there: a rule under 80%, with its lowest team; a tool five or more people named that isn’t on your list (admins only); and a task some people did by hand and spent more than 4 hours on last week (both in the same answer), that people picked as the one they most want help with. They are ranked by how many people each reaches: who didn’t choose the rule, who use the tool, who want help with the task. The ranking is plain arithmetic. A task shows by its words on the role’s task list, never by words someone typed, and the task lines are about the whole company (or, for a manager, their own branch), never one team. For a manager, a line whose count would be every person in their branch, or none of them, is left out.

If the round went out when your company had no AI rules, the page leads with that: “You have no AI policy”, and the share of your people who think you do. The rules block is left out.

On Rounds, each past round of this kind shows its share at L3 or above where an earlier round shows its level.

How teams stay anonymous here. Every team figure needs five answers. No team is shown at 0% or 100%, because that figure would be each person’s own answer. When a team of five or more answered all the same way, no team is shown for that figure at all, only the company figure. That includes What your people can do. The page also never shows a set of teams that, taken away from the company figure, would leave fewer than five people or a group that all answered alike. When Other teams is the lowest group on a rule, the rule says “Lowest: Other teams”.

A company report also carries the paper trail, which links the rounds a run follows from and feeds into.

Almost every block is a total across a group. The exception is the AI literacy check, which also shows each person’s level by name. No block anywhere shows what one person typed.

In this guide:

The top of a Know Your AI round from October 2026 on
Levels by name
Last round, and who moved
The five-person floor
What a company admin sees
What a manager sees
When a team appears

Levels by name

The AI literacy check gives each person a level and a read on each part it asks about. Those sit next to the person’s name on the Levels by person page. Open a round’s results, then open Levels and your Builders and follow See each person’s level. A company admin sees everyone in the company. A manager sees the people who report to them, directly or further down, and nobody else.

Each row shows the person, their team, and their level. Open a row for the read on each part the assessment asked them about. Where a safety answer held the level back, the row says which one. On a Know Your AI round from October 2026 on, a missed rule holds the level at L2, and the row says only that. Which rule, and how the person answered it, is on their own write-up and nowhere else.

The five-person floor does not apply on that page. It protects a total from being traced back to one person, and a row with somebody’s name on it has nothing left to protect. Every total on the results page keeps the floor.

Nothing else about a named person is readable here. Not their answers, not their free text, and not their write-up. There is no setting that opens those.

The level comes from a published rubric rather than a model’s judgement, so a person can be told why they landed where they did. Flow discloses this to each person before they answer the assessment’s first question. Treat a level as a starting point for a conversation with someone, not a rank.

Last round, and who moved

Once you run Know Your AI a second time, the second round’s results carry a Last round panel under Where people are. See the kinds of assessment you can run for how to start a round.

The panel holds three things:

The level distribution of the round before this one, counted the same way.
How many of the people who answered both rounds are at a higher level than they were.
How many people answered for the first time this round. They are counted apart, because there is nothing to compare them with.

Each round is scored against the rubric that was fixed on it when it started. A later change to the rubric cannot move last round’s numbers. The scale is the same in every round. The situations the assessment asks people about are not, so read the two distributions as the same measure on a new set of questions.

Five things can leave the panel empty or short:

Either round is still waiting for its results to be sent. The panel needs both rounds open, because it prints last round’s spread beside this round’s movement.
This is your first round. There is no panel.
One of the two rounds has fewer than five graded answers. No figures open.
Fewer than five people answered both rounds. The distributions open, the movement count does not. With fewer than five people in it, that count would point at individuals.
Nobody who answered both rounds is higher. The panel says so.

No figure in this panel is attributed to a person. The distributions and the two counts are totals, and nothing in them names anybody.

Each round keeps its own Levels by person page, and a manager can open the one for last round as well as this one. So a manager can see the level a named person came out of each round with, exactly as that person’s own write-up shows it to them. What a manager cannot do is read the comparison off this panel: the movement count is a total over everyone who answered both rounds, and it says nothing about which of them moved.

The five-person floor

Anything built from individual answers is held to a five-person floor. That covers the survey questions people tap through and the AI literacy check block. It also covers how a team feels and any text people wrote in their own words. None of these is ever shown for a group smaller than five. So no single reply can be traced back to one person.

The number is five, and it is the same five for every company. There is no setting for it, and nobody can move it, including us. So you can say three things to your people and all three stay true: managers see each person’s level once the round’s results are sent. Nobody sees who gave which answer. A group under five stays folded into the company total.

What a company admin sees

A company admin sees everything. That includes the company-wide overview, every team’s picture, the people roster, every person’s level on the AI literacy check, and the free-text answers people wrote in their own words. The free text is unattributed, and it and every other individual-answer block still sit above the five-person floor.

What a manager sees

A manager sees the results for the people who report to them and nothing else. It is the same picture an admin reads, but only for their direct reports and everyone below them. A manager is not counted in their own view. The roster, the verbatim text, and the written summaries are stripped out. A manager sees totals, charts, and their own people’s levels on the AI literacy check, never a single person’s answers and never their report. Levels by person is the same page an admin opens, narrowed to their reports before it is drawn. Their rail has Home and Tools. Every other company page is closed to them.

A manager does not see a figure that all of their reports, or none of them, meet. That covers each rule and ground rule, Knows where the rules are, the question on whether you have AI rules, and the team question: each reads “Not shown for your team, so no one’s own answer can be read from it.” A survey question where one answer holds everybody, or a pick-many option everyone or nobody picked, gets the same line, and that month is left out of its over-time strip. How work gets done shows its empty note, and What to change next leaves out a rule or a task that all or none of them share. On a manager’s view a rule’s line gives the share who chose it and no “not sure” count. A manager whose reports all chose the rule, or none did, would otherwise be reading each person’s answer. The company-wide view still shows these figures.

Nobody grants this access. It follows the reporting lines on People, which come from your HR system, an imported file, or Edit on a person’s row. A manager whose reports cover the whole company sees the company-wide view.

When a team appears

A team gets a line of its own once at least five of its people answered. That holds for each card on the company report, and at the top of a round for What your people can do, How work gets done and a rule’s “Lowest” team. Below five, a team stays folded into the wider picture.

A team also gets no line that would give a smaller group away. If taking the shown teams away from the company figure would leave one to four people, the smallest team is folded in too. No team is shown at 0% or 100%. Under a literacy card, one line says how many teams count in the total but are not shown on their own. Survey answers and the task questions are never split by team.

Good to know

What feeds this page comes from your own setup: your teams, tools and policy. To shape it, see setting up your company.
Admins and managers read aggregates, with two named exceptions on the AI literacy check: each person’s level, and the completion evidence export. Neither carries an answer.