AI Governance & Your AI Allowance

This guide explains what kinds of AI your workspace can run, how AI usage is counted on plans that include an allowance, and which controls are available to workspace admins.

The five kinds of AI in Sparko

Not all AI costs the same or carries the same risk, so Sparko sorts every AI feature into a class. The class decides which plans include it.

Class	What it does	Included on	Counts as
Insight	Shows you something. Scores, flags and org-level signals, calculated from your own data with no AI model involved	Varies by feature and plan	Nothing
Generation	Produces something a person then applies. A job description, a review draft, an onboarding plan, a summary	Every plan	1 AI action; Core has a monthly pool
Scheduled insight	AI-generated insight narratives and digests for the whole workspace	Growth and above by default; contact Sparko support about an exception	Nothing for scheduled runs
Background	Compliance rule extraction and overnight screening enrichment	Every plan	Nothing
Assistant	Carries out a task for you after you approve it	Every plan	1 AI action for the conversation, plus 1 assistant action each time it actually does something

Insight calculations use no AI model and consume no AI actions. Eligibility still depends on the feature and plan: retention scoring needs the Business predictive features, while pay anomalies and band coverage need the Business compensation features. A class being included does not unlock every feature in that class or replace its permission checks.

AI-generated insight narratives are enabled for Growth and above by default. Core keeps standard reports, dashboards, and rule-based insights with their evidence and recommendations. Sparko manages any exception to the plan default; workspace admins cannot toggle AI insight eligibility. On-demand regeneration requires permission to generate insights (which can be granted through a custom role) and queues the existing job. When enabled on Core, regeneration uses an AI action from the workspace's monthly generation pool.

AI actions, assistant actions, and your monthly allowance

On plans with an allowance, Sparko counts two separate things: AI actions and assistant actions.

An AI action is one thing the AI drafted, summarised or answered. An assistant action is when your assistant actually does something for you, like filing a PTO request, creating a goal, or scheduling a 1-on-1, approval and all.

What uses an AI action

Drafting a job description or offer letter
Drafting a performance review, self review or 1-on-1 agenda
Suggesting OKRs, skills or an onboarding plan
Summarising an exit interview or a survey
Asking Sparko a question, including a message in the assistant chat
Reading a receipt for an expense claim (counted on Core only)
On-demand regeneration of an AI insight digest when AI insights are enabled for your workspace

What uses an assistant action

Every time you approve the assistant carrying out a task for you, such as filing a leave request, creating a career goal or filing an expense claim. Asking a question or having the assistant look something up does not use an assistant action, even in the same conversation.

What does not use anything

Opening dashboards, analytics or reports
Searching, filtering and browsing
Rule-based signals such as coverage warnings and retention scores
Compliance rule extraction and overnight screening enrichment (background tasks)
Scheduled AI insight digests when enabled for your workspace (see above)
Repeating a request Sparko already answered for your company

Both allowances are pooled across your whole company, not divided per person. They reset on the first of each month.

What each plan meters

Core includes 10 AI actions and 3 assistant actions per employee each month, as two separate pools. A five-person workspace gets 50 AI actions and 15 assistant actions a month that anyone can use.

Growth does not meter AI actions at all. It meters assistant actions only, 5 per employee per month, pooled the same way.

Paid Business and Custom meter nothing. Usage is still recorded so Sparko can spot an unusual pattern and talk to you about it, but nothing is ever cut off mid-task.

A workspace trial includes the Business feature set, with 250 AI actions and 75 assistant actions for the whole trial. These ceilings are shared by the workspace, do not scale with headcount and do not reset each month. Trial candidate screening runs overnight, at most 125 applications a night. See Creating your workspace.

Books suggestions

Books has its own allowance, separate from AI actions. It counts each suggestion Books makes with AI: a posting for a bank line, or a bill or receipt it read. Matches Books finds without AI never count.

On a Books Core plan you get 200 suggestions per entity each month, pooled across your entities. Growth is fair use. Business and Custom have no monthly suggestion cap. A Books trial includes the Business feature set with no monthly suggestion cap; the workspace's trial AI allowance still applies. When the monthly allowance runs out, Books stops suggesting with AI until the 1st of next month, and you match or post the rest by hand. See the Books guide.

Where you see it

A small chip sits at the bottom of the sidebar whenever your plan has an allowance. On Core it reads "AI actions" and its bar shows AI actions and assistant actions side by side. On Growth, where only assistant actions are metered, it reads "Assistant actions". During a trial it reads "Trial allowance" and shows what is left of the trial's actions; that allowance covers the whole trial and does not reset monthly. Either way it opens an explainer.

✨ AI actions 31 / 50

If your plan does not meter anything, the chip does not appear at all.

When the allowance runs out

Nothing breaks and nothing is hidden. The AI buttons stay where they are. Pressing one opens a short note saying what is spent, when it comes back, and what a higher plan includes. Running out of assistant actions does not stop AI actions, and running out of AI actions does not stop the assistant from looking things up for you (only from acting on them). The two pools run out independently. Every one of those screens still works the manual way, so you are never blocked from finishing the job.

Admins can also ask Sparko support to raise the allowance for your workspace. It takes effect on the next AI request, with no reinstall and no restart.

Bias and governance reporting

Business and Custom plans include an adverse-impact audit of the decisions Sparko helped with.

Where: Analytics, then the Governance tab.

What it shows: the four-fifths rule applied across three decisions.

Decision	Source
Hiring	AI screening outcomes
Promotion	Completed calibration sessions
Pay	Decided merit recommendations

Results are broken down by gender, race and ethnicity, veteran status and disability status, from the answers employees chose to give in their own account settings. Two views are available: by group, which shows anyone disadvantaged at more than one decision, and by decision.

How small groups are handled

Groups under 5 people are withheld entirely, so a ratio can never identify an individual
Groups under 30 are shown but marked provisional, because a ratio over a dozen people swings on a single decision
Withheld groups still appear as explicit rows in the CSV export, so nothing silently disappears from a compliance filing

Export: the CSV carries the audit window and a data-quality column, ready for a third-party audit under NYC Local Law 144, Illinois HB 3773 or Colorado SB 24-205.

Who can see it: only people with the demographic-detail permission (analytics.dei.view_pii), which is HR Admin and Super Admin by default. The Governance tab does not appear for anyone else.

Recruiting has its own narrower version of the same maths. Analytics > Recruiting, then Screening fairness shows adverse impact for AI screening alone, gated on recruiting.reports.bias_audit. The screening-oversight half of that page is on every plan; the adverse-impact half needs Growth or above.

Admin controls

Turning AI on and off

Company Settings, then the AI section. There is one master switch plus four per-area switches.

Switch	Covers
AI enabled	Everything below. Off means no AI anywhere in the workspace
Recruiting AI	Resume parsing, screening, job description drafts
Performance AI	Review drafts, calibration, peer feedback summaries
Compensation AI	Market band suggestions, equity analysis
People & Career AI	Career signals, skill detection, growth nudges

Turning a switch off is immediate and applies to everyone. People still see the same pages; the AI affordance simply explains that AI is off for the workspace rather than offering an upgrade.

Act on a flight-risk flag or a pay anomaly (Business and Custom) calls no AI model, so it keeps working with AI off. It still checks who you may act on, and each confirm still counts as one assistant action.

Seeing usage

The same AI section shows your monthly allowance when your plan has one: how many actions are used, how many are left, and when the pool resets. It is the same pooled number the sidebar chip shows.

Sparko never shows AI cost, tokens or model pricing to customers. Usage is always expressed in actions.

What admins cannot change

The size of the allowance is not editable in Company Settings. Contact Sparko support to have it raised for your workspace.

Trust and safety

AI only reads data the person asking is already allowed to see
Agentic actions always require an explicit approval before anything is written
AI-generated content is labelled as such, and drafts stay editable
Employee demographics used in the governance audit are encrypted, voluntary, and never visible to managers
No customer data is used to train external models