Skip to main content
See what is happening

The Overview, card by card

The Overview is the home page of AI Governance. It answers four questions for one organization: is it working, what is risky, what changed, and what should I do next. This page walks through it from…

Written By Christopher Scaminaci

Last updated 4 days ago

The Overview is the home page of AI Governance. It answers four questions for one organization: is it working, what is risky, what changed, and what should I do next. This page walks through it from the top: the health strip, the Needs you band, up to eight cards and one collapsed section. For each, it says what it shows, where it links and what it never shows.

At a glance

  • What it does: one page for the state of AI use in your organization, with a link from every number to the page behind it.
  • Plan: every plan. Some cards say what a higher plan adds.
  • Who: any governance role. You see a card only when your role covers its data. A person with no governance role goes to My AI activity instead.
  • Where: AI Governance → Overview.
  • Needs: at least one source that has sent data. Before that, you see the first-run screen.

How the page behaves

  • Seven days against seven days. Most numbers show a change "vs the 7 days before". A number that has no stored history shows no change. These are the policy and training percentages, the readiness bars and the plan meters. StackJack does not make a change up.
  • Every count has its noun. A number says what it counts, and a small (i) beside it says where the figure comes from and which similar number it is not. The same people are counted in four ways on this page, and each way has its own name: people who used AI (the Trend card, the last 30 days), billed staff (the Plan card, the open billing period), people in your directory (the Coverage card) and people who owe training (the Compliance card). They differ on purpose, and the (i) on each says why.
  • Every number is a link, when your role can open the page it leads to. A number you cannot follow is plain text.
  • No ranking of named people, anywhere. The Overview counts people. It never lists a person, scores a person or sorts by a person. The people count is not split by tool or by client.
  • One card at a time. Each card is read on its own. If one cannot load, it says so in its own frame and the others stay. A card that your role does not cover is left out. It is never shown as "all clear".
  • Times are in UTC.
  • Several organizations. If you govern clients, the page is for the organization chosen in the AI Governance menu. For all clients at once, see All clients.

Before there is data

An organization that has not finished Get started and has no data yet sees the first-run screen: a welcome card, a link to Get started and the lines about sources that need a look. A person who has data but has not finished Get started sees a small card, Finish setting up AI Governance, with a Get started link. An administrator who finished sees a What next card with three picks and a Dismiss button. See Get started with AI Governance.

Three setup figures, three names

Setup is counted three ways, and each page names the one it shows:

WhereWhat it saysWhat it counts
The ring on the first-run screen and on the small setup card"5 of 8 setup checks done"The eight setup checks, ticked from live data: jurisdiction, staff notice, endpoint settings, a source with data, vendor keys, Microsoft 365, rule packs and alerts.
Get started"Step 5 of 7"The seven steps of the guided setup. The (i) beside it says how many are done.
The line under the health strip, and the Setup milestones column of All clients"Setup: 4 of 5 milestones done"Five milestones for one organization: the jurisdiction confirmed, the staff notice recorded (where the jurisdiction needs one), a source switched on, data arriving, and Get started finished.

The line under the health strip shows for an administrator while a milestone is still open, and it is the same figure the All clients table shows for the same client. The ring and the milestones are not shown at the same time. The three figures differ on purpose; they are not one score.

1. The health strip

One line at the top. On the left is the overall state: Working, Partly working, Waiting for data, Needs attention or Not set up. It reads Partly working when a source already works and another part is still waiting. Next to it is one chip for each area:

ChipWhat it checks
SourcesHow many of your sources work: " of sources working". It counts each source that is switched on. A source that belongs only to ways you marked as not used is left out. A source that StackJack reads on a schedule counts as working when it reads without a problem, even before it has found AI use.
CaptureHow many of the ways you use work: " of capture methods working". A way you marked as not used is left out. It reads the same states as Data sources.
Staff notice"Staff notice recorded", "Staff notice not required", or a block: "Staff notice not recorded: collection is blocked" or "Waiting for the works council agreement".
Alerts"No alert channel", " alert channels working", or how many need a look.
Archive"No archive copy", "Archive copy working", or how many targets need a look.

A chip that is not working names its fix and links to it. A part that is switched off on purpose, such as an alert channel nobody set up, is drawn quiet and never lowers the overall state.

While you are in monitor mode, the strip adds a line: "Monitor mode since , days. The packs alert only after you make one Live." with a link, Open Rule packs. The line goes when a rule of an enabled pack is made Live, or when no rule is left in Test mode. Turning a pack on in Test mode again starts a new count.

2. Needs you

The most urgent lines, as cards with one action each. See all () opens the rest. Each line has a sentence that says why it matters. The lines are:

  • collection blocked until the staff notice, or the works council agreement, is recorded,
  • open findings by severity,
  • prompts or tool calls blocked, and warned, in the last 7 days,
  • AI tools that nobody has decided on. This line counts tools and servers. The AI tools card counts models too, so its number can be higher,
  • clients that drift from a standard, and changes that wait for your decision (only when Deploy and manage is on for your account and you may open a client with it),
  • name reveal requests waiting for your decision,
  • staff data requests waiting for an answer,
  • source problems: a source that is off, sent nothing for 24 hours, refused records before the notice, has gaps, fails to pull, or stopped pulling.

Open findings count the last 90 days. The other lines count the last 7 days. A line names no person and shows no text. If nothing is wrong, it says "Nothing needs your attention."

3. Trend

One chart with a switch for the metric: AI calls, Tokens, Estimated cost, Findings, Errors, Blocked prompts and Warned prompts. Choose 7, 30 or 90 days. Above the chart, two sentences give the total and the change, for example "57,798 AI calls in the last 30 days. +30,284 in the last 7 days, compared with the 7 days before." Cost is an estimate from telemetry, never a vendor figure. A day with no data is left off, not drawn as zero. The card also shows how many people used AI in the last 30 days, as one count, for example "21 people used AI in the last 30 days". Each person counts once from the daily usage totals; applications and automations are not people. This is not the number you are billed for and not the number of people in your directory.

Blocked and warned counts come from the most recent 20,000 findings. When the history is longer, the card says the oldest days can read low.

4. Risk

NumberWhat it isOpens
Open findingsFindings still open, with the count of the highest severity. Findings caught in test mode are not part of it.Activity → Findings, filtered to open
Blocked, 7 daysDistinct prompts or tool calls StackJack stopped on the paths it sits in. A note shows how many were warned.Activity → Findings, filtered to blocked
Unapproved tool useUse of a tool nobody approved, in 7 days.A plain number

Details shows open findings by severity and the three rules that fired most. Each rule links to that rule. If no rule fired, the card says "No rule fired in the last 7 days." See Findings and triage.

Caught in test mode. When the rules wrote findings in test mode over the last 90 days, a second line says so beside the open count, for example "0 open. 412 caught in test mode in 90 days." A rule in test mode records what it would catch and does nothing: nothing is blocked, warned about or sent as an alert. These findings are never counted as open, and the two figures are never added. The number links to those findings, and Make a pack Live opens Rule packs, where you turn a pack from test mode to Live. A new organization has every pack in test mode, so the open count is zero until a pack goes Live; this line is the proof that the rules are working. No line shows when nothing was caught in test mode.

5. Deploy and manage

Shown only when Deploy and manage is on for your account and there is something to say. When it is off or below your plan, the card is not drawn. When you may open no client with it and nothing waits for you, the card is not drawn either. It is never shown as a row of zeros. The figures are the ones the Clients and standards page shows, for the clients you may open. On your own organization it covers every client; on one client it covers that client. The changes waiting are counted the way the Changes tab of the menu and the Clients and standards tile count them, for the same clients.

NumberWhat it isOpens
Clients matching"9 of 11": clients that hold a standard and match it in every setting StackJack read, over the clients that hold a standard. The note says "11 of 12 clients have a standard". Drawn only when a client holds a standard.Clients and standards
Clients driftingClients where at least one setting differs from its standard and nobody accepted the difference. Drawn only when a client holds a standard.Clients and standards
Changes waitingPlans with a change that waits for a person to approve it. Drawn whenever it is above zero, even with no standard in use: a purchase or a seat cleanup waits for a decision without one. It is the number the Changes tab of the menu shows.Changes

A line under the numbers says how many seats nobody uses, for example "14 of 120 seats are unused". An unused seat is a seat holder whose use was read and who has had no use inside their seat group's cleanup window; a seat group with no reading counts nothing. The line is left out when no seat group is set up. See Seats and purchases.

When no standard is in use at all, the card says "No standard is in use yet. Publish a standard and add it to a client, and this card shows where each client stands." with a link to Standards. When only your own organization holds a standard, it says "No client holds a standard yet." and, below, where your own organization stands, for example "Your own organization drifts: 3 settings differ from its standard." The client numbers never count your own organization, and the card never reads "0 of 0". It does not show a tick. The same figures give the Needs you lines about drift, about your own organization drifting, and about waiting changes, so the card and the lines cannot disagree.

The card counts at most 500 clients. If you may open more than 500, the card, the Needs you lines and the tool count the first 500 and say so: "Counted over the first 500 clients. The rest are not counted." A client after the first 500 that drifts is in no figure, so a card in this state never says that every client matches, and it adds a Needs you line that says the lines count the first 500 clients only. The Clients and standards page uses the same limit and shows "the first of more" beside its count.

Through your AI tools, the figures leave out every client whose AI tools are switched off, because those clients' governance does not reach AI tools. The Portal shows them all. If StackJack cannot tell which clients have AI tools off, the tool leaves the Deploy figures out and says so. See Deploy and manage.

6. AI tools

NumberOpens
Waiting for a decision, with the note "48 waiting in all. 40 new in 7 days. 8 are models." and how many staff requestedAI inventory → Inventory, filtered to new tools
ApprovedThe inventory, filtered
DeniedThe inventory, filtered

Waiting counts every tool, server and model nobody has approved or denied. "New in 7 days" counts the waiting ones first seen in the last 7 days. The Needs you line about new AI tools counts tools and servers only, because a model never raises that alert, so its number can be lower than the card's. The card also shows the newest tool nobody has decided on, as a link to it, and By risk class (Not screened, Prohibited, High risk, Limited risk, Minimal risk). The header link is See AI tools. See The AI tool inventory.

7. Coverage

What can see AI use: sources that deliver data, devices with the endpoint agent and the browser extension. People are counted, never listed. The blind-spot lines name the group they count: "27 of 29 people in your directory have no browser extension". Your directory is every person record that is not erased or merged into another; applications and automations are not people and are not in it.

NumberOpens
Sources you use, " of " with "working" under itData sources → Status
Browser installs, with how many people they coverStaff computers → Browser extension
Devices with the agentStaff computers → Endpoint agent

Sources you use counts the sources you use, and how many of them work. A source you marked as not used is left out. It is the first headline number of the Data sources page. While you use no source, it shows 0 and "No source is connected yet".

Under the numbers, one line gives the second headline number: how many of the places where AI use can happen are covered, for example "1 of 5 places covered." Next to it is the link Open Data sources.

A number that could not be read shows a dash and "Could not be read right now". The line then reads "The places covered could not be read right now." The card never shows a zero that nobody counted.

By capture method lists the state of each way you use. The header link is Open Data sources. The card does not split the people count by tool.

8. Compliance

NumberWhat it isOpens
Policy acceptedThe share of people in scope who accepted the current AI use policy, " of people in scope accepted". "No policy is published" when there is none.Compliance → AI use policy
Training doneThe share of the people who owe training who finished it, for example "0 of 26 people who owe training have finished". "No training is required" when there is none.Compliance → AI training
Open incidentsIncidents still open, how many opened in 7 days, and the nearest notification deadline.Alerts → Incidents

Below them, one bar for each framework shows how many of its controls your evidence covers. Each bar links to the report pack that holds the evidence. "Readiness appears here once policies and evidence are set up." The header link is Open readiness. See Compliance readiness and the frameworks.

9. Spend

Three separate values, never added together:

  • Estimated from telemetry, 30 days, with a note if some model calls had no price,
  • Reported by AI vendors, listed on the Spend page,
  • Budget used, with the budget name, the amount and "hard stop" if it has one. "No organization budget is set" otherwise.

Models and vendor lines expands to the detail. The header link is See spend by month. See Spend and budgets.

10. Plan

Your AI Governance plan and how much of it this billing period has used:

  • Billed staff, " of ", or the count with "Unlimited" on a plan with no limit. These are the people StackJack saw using AI in this billing period, plus any applications past the number your plan includes. It is the number your plan bills on, and it can differ from the people in your directory,
  • Stored text, with what the plan says about it ("Stored with StackJack, charged per GB" on Pro and Business, "None on this plan" on Free, or "No fixed allowance" where that applies),
  • AI judgments, used against the monthly allowance, or Paused if the limit was reached.

The header link is Open billing. See Plans and what each includes.

11. Where the usage goes

A collapsed section at the bottom with the usage breakdowns: by tool, vendor, model, client and team. A team is shown only under the rule that keeps one person out of two rows. Nothing here lists a person.

What the Overview never shows

  • A named person, a ranking of people, or a score for a person.
  • The text of a prompt or an answer, or a matched value. A finding shows its rule and a masked token.
  • A cost that adds estimated spend to vendor-reported spend.

Troubleshooting

What you seeWhat to check
The first-run screenNo source has sent data yet. Open Get started or Data sources → Status.
A card says it could not loadReload the page. The other cards are not affected.
A card is missingYour role does not cover its data. See Roles and who can do what.
A number has no change beside itIt has no stored history. This is expected for percentages and plan meters.
The open count is zero but you expect findingsNew organizations start with every pack in test mode. Look for "caught in test mode" on the Risk card and use Make a pack Live.
There is no Deploy and manage cardDeploy and manage is off for your account, your plan is below Pro, or you may open no client with it.
Waiting for a decision is higher than the Needs you lineThe card counts models too. The line counts tools and servers only.
The Trend chart reads low for old daysThe blocked and warned counts use the most recent 20,000 findings.
You land on My AI activityYou hold no governance role. See What staff can see about themselves.

Next