Start your free trial

AI Assistant Guide

The Sparko AI Assistant is your intelligent companion for all HR-related tasks. Available 24/7, it can answer questions, complete tasks, and provide personalized recommendations.

Accessing the AI Assistant

Methods to Open

  1. Omni-search (recommended): Press Cmd/Ctrl + K (or /), type your question, and press Enter or choose "Ask Sparko about…". The answer streams in right there, grounded in what you can access, with follow-up suggestions and inline charts.
  2. Full chat page: Press Open chat in search, open My Hub > Sparko AI, or type G then A for the full conversation view. It remembers earlier turns in the same chat on every plan. Keeping those chats after you close them is a Business feature.

Interface Overview

The chat page's message box is the same Ask Sparko box you see on Home: a purple Sparko badge over a rounded field with a Send button. While the conversation is empty, four suggested questions sit above the field (Check my PTO balance, Review my goals, Grow my skills, Set a career goal); pick one to send it. Once the conversation starts they go away, so the thread has the room. Press Enter to send and Shift + Enter for a new line. If a quick pulse question or a tour invitation is open at the bottom of the screen, the box moves up so it is never hidden behind it.

🤖

Performance, time off, goals and career questions, answered from the records you already have access to.

✦ Ask Sparko Check my PTO balance Review my goals Grow my skills Set a career goal 📎 Ask about performance, time off, goals or your career Send

If you run hires, the top of the page switches between Chat and Runs; Runs lists the hires in progress (see People, Hire with Sparko AI).

AI Personas

The AI adapts its personality based on your question. A persona sets the tone of answers only. Every persona can do the same things.

HR Business Partner

For policy questions, HR processes, and administrative tasks.

Example questions:

  • "What's the parental leave policy?"
  • "How do I report a workplace concern?"
  • "How many days of leave do I have left?"
  • "Where do I find my latest payslip?"

Career Advisor

For career development, skills, and growth conversations.

Example questions:

  • "What skills should I develop for promotion?"
  • "Help me set a career goal"
  • "What learning paths do you recommend?"
  • "Am I ready for a senior role?"

Compensation & Benefits Advisor

For salary, bonus, equity, and total rewards questions.

Example questions:

  • "Where does my salary sit in my band?"
  • "What is the pay range for my role?"
  • "When is my next salary review?"
  • "Explain my stock options"

Common Tasks

Hiring

Ask "help me hire Mei Ling Tan" and add anything you know in the same sentence. The answer is a plan card: Open plan shows the plan, and Accept plan in it starts the hire and opens it in Runs, beside Chat in Sparko AI. Back in Chat the card reads "Accepted. Track it in Runs." with Open run. Correct a fact by saying it ("no, she starts 1 Dec"); the plan updates and shows what changed. Ask "where is Mei Ling's hire?" for the current step. For a contractor or freelancer, the plan card shows the risk band; open the plan to keep a Medium or High band as a contractor or switch to employee. After an employee plan, Hire as a contractor instead asks for the same hire as a contractor.

Time & Leave

Check PTO balance:

You: "How much PTO do I have?"
AI: "You have 12 days of PTO remaining this year.
     - Total: 20 days
     - Used: 8 days
     - Pending: 0 days

     Would you like to request time off?"

Request time off:

You: "I need to take off December 23-27"
AI: "I'll help you request time off for December 23-27 (5 days).

     This will use 5 of your 12 remaining PTO days.
     Your manager James Wilson will be notified.

     [Request Time Off] [Modify Dates]"

Performance

Get OKR help:

You: "Help me write an OKR for improving code quality"
AI: "Here's a suggested OKR for code quality:

     Objective: Deliver high-quality, maintainable code

     Key Results:
     1. Reduce code review revision requests by 40%
     2. Achieve 85% unit test coverage on new code
     3. Run 2 code-quality workshops with the team

     [Create this OKR] [Modify] [See alternatives]"

Prepare for 1-on-1:

You: "Help me prepare for my 1-on-1 with James"
AI: "Here's your 1-on-1 prep with James Wilson:

     📅 Meeting: Tomorrow at 3pm

     Suggested topics:
     - Your OKR progress (currently at 72%)
     - Recent project completion
     - Career development discussion

     Since your last 1-on-1:
     - You completed 2 key results
     - Received positive peer feedback
     - Questions about promotion timeline

     [Generate agenda] [Add notes]"

Career Development

Get career advice:

You: "What should I focus on to get promoted?"
AI: "Based on your profile as Software Engineer II,
     here's what you need for Senior Engineer:

     Skills to develop:
     🔴 System Design (Critical gap)
     🟡 Technical Leadership (Moderate gap)
     🟢 Code Quality (Strong)

     Recommended actions:
     1. Lead the design of a cross-team system
     2. Lead a cross-team project
     3. Mentor a junior engineer

     Readiness score: 68%
     Estimated timeline: 9-12 months

     [Create development plan] [Find mentor]"

HR Policies

Ask about policies:

You: "What's the remote work policy?"
AI: "Here's our remote work policy summary:

     ✅ Eligibility: All employees after 90 days
     📍 Options: Full remote, hybrid, or in-office
     📝 Requirements: Manager approval, home office setup

     Key points:
     - Core hours: 10am-3pm your timezone
     - Equipment: Company provides laptop, monitor
     - Internet: $50/month stipend available

     [Read full policy] [Request remote work]"

Advanced Features

Context Awareness

The AI understands your context:

  • Current page you're viewing
  • Recent actions you've taken
  • Your role and permissions
  • Time of year (reviews, enrollment periods)

Example: When you're on the Performance page:

You: "Help me with this"
AI: "I see you're looking at your Q4 OKRs.
     Would you like me to:
     - Add a check-in for your current OKR
     - Suggest updates based on progress
     - Compare to team averages"

Multi-Turn Conversations

The AI remembers the chat you are in, on every plan:

You: "Show me the engineering team"
AI: [Shows engineering team list]

You: "Who has the most tenure?"
AI: "In the Engineering team, James Wilson has the
     most tenure at 8 years, 3 months."

You: "What's his email?"
AI: "James Wilson's email is [email protected]"

Attachments

You can share files with the AI:

  • Screenshots for help with UI issues
  • Documents for policy questions

Action Buttons

AI responses often include action buttons:

  • Primary action: Main suggested action
  • Secondary actions: Alternative options
  • Learn more: Additional information

How answers are laid out

When the assistant looks something up, the result comes back as cards rather than a paragraph listing everything again. What you see depends on the answer:

  • Record lists. People, jobs, goals, reviews and plans come back as rows. Select one and its detail opens in a panel beside the conversation. Open full record is the only thing that leaves the chat.
  • Figures sit on one compact line, so a couple of numbers do not turn into a wall of tiles.
  • Tables and charts get the wider layout, with your reply above them. Everything else is grouped into a single card with the answer's own heading and the scope it covers.
  • Sources. Answers that stand on documents carry their own Grounded in N sources line, folded away until you open it. It travels with the answer, so scrolling back to an earlier turn still shows what that answer was based on.
  • Follow-ups. One-press prompts to keep going sit under the answer.

When a list is longer than the card can hold, the card says how many there are in total and how many it is showing, so nothing goes missing quietly.

Asking what it can do

Ask "what can you do?" and the answer is built from your own permissions, not written by the model. It comes back as prompts you can send with one press, split into things it can look up and things it can set up with your confirmation, plus a few examples suited to the page you are on. Because it is computed from your access, it can never offer you something you would be denied.

Whose data an answer covers

Questions like "what needs my attention?" and "how is my team doing?" cover you and your direct reports by default. Ask about your whole org, your indirect reports or the entire company and the assistant widens to your full reporting line. Either way the answer states the scope it used, so a team roll-up is never mistaken for the whole company. It can only ever summarise your own reporting line, never someone else's team.

Agentic AI: Action Execution

The AI can execute actions on your behalf, always with your explicit approval:

What changes by plan

Asking questions works on every plan. The assistant reads only what you already have permission to see, and answers in the chat.

On Core, each AI answer uses one of your company's monthly AI actions. The "AI actions" chip at the bottom of the sidebar shows how many are left. On Growth, Business and Custom, answers are not counted.

Trials include the Business feature set, with 250 AI actions and 75 assistant actions for the whole trial, shared across the workspace. These allowances do not reset monthly or grow with headcount. Trial candidate screening runs overnight. See Creating your workspace.

Answers cite your company's documents on every plan. Ask about a policy and you get the handbook or announcement it came from, whether you are on Core or Custom.

Proposed actions work on every plan. What differs is the count: on Core and Growth, an action the assistant carries out costs one of your monthly assistant actions, a separate allowance from AI actions, and on Business and Custom nothing is counted.

Acting on flight-risk flags and pay anomalies is on Business and Custom. Core and Growth don't see the Act button. It works even when your workspace has AI turned off, because it calls no AI model. See "Acting on flags and pay anomalies" below.

When the assistant looks something up, it reads what it found before it replies, so you get an answer in plain language with the actual figures in it, not just the raw result. The lookup card, chart and sources stay with the answer.

Keeping your chats is a Business feature. On Core and Growth a conversation lasts as long as you have it open. On Business and Custom it is saved to your history, and only you can see it. A saved conversation keeps what each answer came with: the sources, the charts, the lookup results and the steps the assistant took, so reopening it shows the same answer you got the first time.

In an active workspace, a conversation you have not touched for 90 days is archived. It is not deleted. It drops out of your History list. To see conversations you archived, open History and pick the Archived tab. Send another message in an archived conversation and it returns to the Active tab.

The assistant proposes an action, shows you exactly what it would do, and carries it out once you confirm. It can only ever propose an action you have permission to perform yourself.

What is Agentic AI?

Instead of just answering questions, the AI can:

  • ✅ Create career goals and development plans
  • ✅ Create OKRs with key results
  • ✅ Request feedback from colleagues, or start a full 360 feedback workflow
  • ✅ Schedule 1:1 meetings, and put a real invite on Google Calendar when both people have connected theirs
  • ✅ Submit time off requests
  • ✅ Update your profile
  • ✅ Request a mentorship match
  • ✅ Draft performance reviews in STAR format (managers can approve and submit)
  • ✅ Create a job posting, then publish it or send it for approval
  • ✅ Add a hiring location, so a posting is not blocked on a city you have not set up
  • ✅ Draft an onboarding plan for a new hire, for you to review and activate
  • ✅ Look up a pay range from your salary bands before you post a role

How It Works

  1. You ask in natural language: "I want to become a Senior Engineer in 12 months"
  2. AI proposes actions: Shows you what it will do with all details. An action about one person names them on the card ("For Priya Shah", linked to their profile), and a person in an answer shows their name, title and department, as long as you can see their profile
  3. You confirm: Review and approve (or cancel)
  4. AI executes: Action is completed automatically

A long action, such as writing a performance review, a development plan, an onboarding plan or a 360 request, shows Checking on its card while it runs. The result appears in the chat when it finishes, and if you reload the page the chat picks the action up again.

Example: Creating a Career Development Plan

You: "I want to become a Senior Engineer"

AI: "I'll help you create a career goal and development plan.
     Let me set this up for you..."

[Action Card Appears]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 Generate Development Plan
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Target Role: Senior Software Engineer
Timeline: 12 months

I'll create a 3-phase development plan:
• Phase 1: Foundation (Months 1-4)
• Phase 2: Building (Months 5-8)
• Phase 3: Mastery (Months 9-12)

[✓ Confirm Action] [✕ Cancel]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

[After confirmation]
AI: "✅ Development plan created.
     - 12 milestones across 3 phases
     - Skills analysis complete

     You can view your plan in Career Development."

Example: Creating an OKR

You: "Create an OKR for Q1: Launch new mobile app"

AI: "I'll create this OKR for you with key results."

[Action Card Appears]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🎯 Create OKR
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Objective: Launch new mobile app
Key Results: 3 results
Quarter: Q1 2025

Key Results will include:
1. Release MVP by end of January
2. Achieve 10,000 downloads
3. Maintain 4.5+ star rating

[✓ Confirm Action] [✕ Cancel]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

AI: "✅ OKR created. Track progress in Performance."

Example: Requesting Feedback

You: "Request feedback from Sarah about my presentation"

AI: "I'll send a feedback request to Sarah Johnson."

[Action Card Appears]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💬 Request Feedback
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
From: Sarah Johnson
Context: My presentation

[✓ Confirm Action] [✕ Cancel]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

AI: "✅ Feedback request sent to Sarah Johnson."

Example: Scheduling a 1:1

You: "Schedule a 1:1 with my manager next Friday"

AI: "I'll schedule a 1:1 meeting with James Wilson."

[Action Card Appears]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📅 Schedule 1:1 Meeting
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
With: James Wilson (Your Manager)
Date: Next Friday (Dec 8)
Duration: 30 minutes
Agenda: General check-in

[✓ Confirm Action] [✕ Cancel]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

AI: "✅ Meeting scheduled for December 8 at 2pm."

Example: Generating a Performance Review

You: "Write a performance review for Connie Hennelly for Q4 2025"

AI: "I'll generate a comprehensive performance review for Connie Hennelly
     using data from OKRs, 1-on-1 notes, and 360 feedback."

[Action Card Appears]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✍️ Write Performance Review
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Employee: Connie Hennelly
Review Period: Q4 2025
Format: STAR (Situation, Task, Action, Result)

Data Sources:
• 3 OKRs (avg 82% completion)
• 8 1-on-1 meeting notes
• 5 peer feedback items

The review will include:
• Accomplishments (3-4 STAR examples)
• Areas for improvement (1-2 STAR examples)
• Goals for next period
• Overall rating recommendation

Quality checks:
✓ Plagiarism detection enabled
✓ Uniqueness scoring (70%+ required)
✓ Template pattern detection

[✓ Confirm Action] [✕ Cancel]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

[After confirmation]
AI: "✅ Performance review generated.

     📊 Quality Metrics:
     • Uniqueness: 87% (Excellent)
     • STAR compliance: 100%
     • Data sources: 3/3 integrated
     • Template patterns: None detected

     ✨ Accomplishments (3 examples)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
     Situation: Connie was assigned to lead the platform
     launch with a team of 4 developers...

     Task: Her goal was to deliver 3 major features by Q4
     end while maintaining code quality...

     Action: She implemented daily standups, code review
     automation, and mentored 2 junior devs...

     Result: All 3 features shipped on time, customer
     satisfaction increased 23%, team velocity improved 40%.

     [View full review] [Edit] [Send for approval]

     The review has been saved and is ready for your review.
     You can approve it directly or make edits first."

Example: Requesting 360 Feedback

You: "Start a 360 feedback process for my Q4 review"

AI: "I'll initiate a 360 feedback workflow for you."

[Action Card Appears]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💬 Request 360 Feedback Workflow
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Review Period: Q4 2025
Deadline: 14 days from now

Recommended reviewers (AI-selected):
• 3 peers you collaborated with most
• 2 cross-functional partners
• Your manager (auto-included)

Feedback areas:
• Technical skills
• Communication
• Collaboration
• Leadership

[✓ Confirm Action] [Customize reviewers] [✕ Cancel]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

AI: "✅ 360 feedback requests sent to 6 reviewers.

     You'll receive notifications as feedback comes in.
     Deadline: December 19, 2025"

Available Actions

These work on every plan. Each one still needs the same permission you would need to do it yourself, so your list is shorter than this if your role is narrower. A few draw on data that is a paid add-on, such as market pay, and stay unavailable until your plan includes it.

Career Development

  • Create Career Goal: Set target roles with timelines
  • Generate Development Plan: AI creates phased plans with milestones
  • Find a Mentor: Ask for a mentor in a skill area and Sparko ranks the people who could help, using the same matching as your profile's Career tab: shared skills where they are further along, what they mentor on against your goals, seniority, and how close they sit to you. Each one comes back as a card you can open
  • Request Mentorship: Connect with specific mentors

Performance Management

  • Create OKR: Set objectives with key results
  • Find a colleague by name: The assistant finds anyone in the Employee Directory by name, the same people you can find with search, managers included. Finding someone does not open their record: their performance, pay or reviews stay limited to what your role allows
  • Request Feedback: Ask a colleague for feedback on you, by name (e.g. "ask Priya for feedback on my demo; how clear was the pricing part?"). Your note and questions go with the request. It lands in their Feedback > To complete tab, which shows them what you'd like to know, and their answer arrives in your feedback inbox
  • Schedule 1:1: Book time with your manager, or with someone who reports to you (name them, e.g. "schedule a 1:1 with Maya next Tuesday")
  • Write Performance Review: AI generates complete reviews with STAR format. Name the person and the period (e.g. "write a performance review for Maya for Q3 2026"); Sparko finds them among the people you can see, and asks which one only when more than one person matches
  • Approve Performance Review: Validate and approve AI-generated reviews
  • Request 360 Feedback: Initiate 360 feedback workflow with reviewer selection

Connected Tools

These only appear when the connector behind them is actually connected. If it is not, the assistant does not offer them at all.

  • Check Calendar Availability: Once you have connected your Google Calendar in Account Settings, the assistant can check when you and the other person are both free before it proposes a 1:1 time. If it cannot see a calendar, because that person has not connected theirs or their free/busy is not shared with you, it says so rather than guessing they are free
  • Add the Calendar Invite: When a 1:1 is scheduled and both of you have connected your calendars, Sparko creates a real Google Calendar event and invites you both. If that fails, the 1:1 is still booked in Sparko and the assistant tells you it could not add the invite
  • Check Recent Work Activity: If your company has connected Linear, managers can ask what one of their reports has recently been working on there, as objective context for a performance conversation. It needs the same permission as seeing that person's performance data
  • Search Company Policy: If your company has connected Notion, HR admins can have the assistant answer policy questions from the company's own written pages, so the answer reflects what your company actually wrote

Analytics

  • Generate a Chart: Describe a chart in plain language and Sparko builds a permission-scoped visualization from the data you can access (e.g. "headcount by department", "how is attrition trending")

HR Operations

  • Draft an Onboarding Plan: Generate a phased plan for a new hire. It is saved as a draft for you to review and activate, so the new hire sees nothing until you release it
  • Request Time Off: Submit PTO requests (managers can request on behalf of a report)
  • Update Profile: Modify your employee information
  • Check Leave Balance: View remaining PTO
  • File an Expense Claim: Describe a receipt, a mileage trip or days away, or attach a receipt photo. Sparko proposes one claim for you to confirm
  • Check My Expense Claims: See where your own claims are (a lookup, so it uses no assistant action)

Job Structure

Available on Growth and above, the same plans that open the career framework editor.

  • Propose Job Architecture Changes: Ask Sparko to look at your job families, career ladders and levels, how job titles map onto families, and which rungs carry a pay band, and it comes back with a list of changes that would tidy it up. You can ask for the whole company or narrow it to one department. Nothing is written by the proposal itself. It reads a company-wide roll-up of job titles and headcount, so it needs the same permission as the job titles page in Compensation, not the one everyone has for browsing the framework
  • Apply Structure Changes: Tick the changes you want and Sparko applies only those. Needs the same permission as managing the career framework, and each change is separately re-checked against the permission its own kind needs, so a change your role cannot make is refused by name even if you tick it

How the proposal reads. Every proposed change says what it is, what it would touch (how many job titles, how many people, how many bands) and, where it cannot run, why not. A change that stands on another one, such as mapping a title into a family the same set creates, is greyed until you tick the one it depends on. Untick that family and the title drops out of the set with it. If an apply is refused, the list stays on screen with the refused change marked, so you can untick it and apply the rest.

What Sparko is given. Counts and structure only: family names, ladders, level names, how many rungs are priced, your locations with headcount, and the job titles your people hold with their departments and countries. No salary figures and no personal details go into the proposal.

Undo. Every change that lands is recorded with the exact values it replaced, so it can be reversed rather than guessed back. Reversing one is a decision of its own, so Sparko will not undo a structure change from the chat even though it can make one. The place to undo one is Admin mode, under Career > Structure, where Recent changes lists what has been applied and offers Undo on each change nothing later depends on.

Reviewing the proposal. Sparko shows its proposal as a checklist card under its reply. Untick anything you do not want, then apply. Rows that cannot be applied yet say why. If you ask again within a day, Sparko shows the same proposal instead of making a new one.

After you apply. The result says how many changes were applied and why any were left out, with one chip per applied change. See it in the structure settings opens the structure page, where an admin can undo a change.

Recruiting

  • Create a Job Posting: create a job posting, then publish it or send it for approval from the result card. Give it a career path and level and you can leave the description and requirements out, and the assistant writes them from the competencies that level asks for
  • Look Up a Pay Range: The assistant reads the salary band for that career path and level, plus any market benchmarks stored on the band, and explains the range in plain language. When no band covers the role it says so instead of estimating a number
  • Add a Hiring Location: Add a city, country and pay currency as a company location. Needs the same permission as adding one on the Locations page, and an existing location is left alone rather than duplicated

Naming things in plain language. You do not need ids. Say "the Manila office" or "Engineering, Senior" and the assistant looks the name up in your company's active locations and career framework before it acts. If nothing matches, it says so rather than inventing a location or a level.

When something is missing, it offers the step that fixes it. Ask for a job posting in a city your company has not set up yet and the answer opens with what is standing in the way, each blocker carrying the one action that clears it. Confirm "Create Dublin location" and the posting picks the new location up straight away, so you do not have to say it again. Blockers you only need to read carry an Acknowledge and continue instead.

Hiring in one sentence. Ask for a hire in one sentence, for example "Hire a senior backend engineer in Singapore, 120k to 150k". Sparko shows the posting as one card, with each missing piece as a compact row inside it: the career path, the location, the pay band from the range you gave, and the headcount note. Confirm each one. When your company needs no approval, the card's button reads Create and publish; otherwise it reads Create and send for approval. The pay range is optional: without one, it fills from the band.

  • Save a Pay Band (needs Create Compensation Bands, on the Business plan): saves the range you gave as the level's band.
  • Publish a Job Posting (needs Edit Jobs): puts a job on your careers page, or sends it for approval.

Safety & Control

You're Always in Control:

  • AI never executes without your confirmation
  • You see all details before approval
  • You can cancel any action
  • Every action the assistant carries out is recorded in your company's audit log, visible to admins alongside every other security event

What AI Shows You:

  • Clear action description
  • All parameters and details
  • Impact of the action
  • Option to cancel or modify

Tips for Better Results

Be Specific

❌ "Help with time off"
✅ "I need to take December 23-27 off for vacation"

Provide Context

❌ "What should I focus on?"
✅ "What skills should I develop to become a Tech Lead?"

Use Natural Language

❌ "query: pto balance status"
✅ "How much vacation time do I have left?"

Ask Follow-ups

"Tell me more about the first option"
"Can you explain that differently?"
"What if I did it this way instead?"

Privacy & Security

What AI Can Access

  • Your profile and employment data
  • Data you're authorized to view
  • Company policies and handbooks
  • Your conversation history (this session)

What AI Cannot Access

  • Other employees' private data
  • Data outside your permissions
  • The open web
  • External tools, unless your company or you have connected them (Google Calendar, Linear, Notion)
  • Previous session conversations (unless saved)

Data Handling

  • Conversations are encrypted
  • Not used to train external models
  • Retained per company data policy
  • You can delete conversation history

Opening Sparko AI

Open search and press Open chat, open My Hub > Sparko AI, or type G then A on any page. What you were reading in search carries over into the chat. On plans without the full chat page, the button in search reads Ask Sparko and answers right there.

Setting up entities with Sparko

Ask "Set up our legal entities" or "Assign people to entities". Sparko answers with a checklist card: the entities it found, with a name you can edit on each row, who goes where and why, grouped by entity with a select per person, and anyone it could not place. Untick what you do not want and apply. Ask "Suggest who runs HR in each entity" for role suggestions. Roles for one entity are on Growth and above; on Core, Sparko says so in one line.

Client work and payroll with Sparko

  • Describe a deployment in one sentence, or upload a PO, statement of work or email, and Sparko proposes the client, project and assignments for you to check. The card opens as a short summary, and its Review button (for example Review 12 changes) shows every row.
  • Upload a timesheet as a table, a spreadsheet or a photo, and Sparko turns it into a draft sheet to review, with the photo beside the rows. Hours are never approved on their own.
  • Ask for an advance in one sentence, and Sparko proposes it with the instalments.
  • For a country with no payroll rules, Sparko drafts them for you to check in Payroll rules.

Sparko never shows or repeats a pay rate it has stored. It only uses figures you type in the conversation.

Expenses with Sparko

You need to be able to submit expense claims. Ask "Claim 42.50 SGD for a taxi to a client yesterday", "Claim 120 km of mileage to the Penang site on Monday" or "Where are my expense claims?".

To file from a receipt, click the paperclip beside the message box (Attach a receipt) and pick a photo or PDF. Sparko uploads it, reads it, and then proposes the claim. If it cannot read the receipt, say so in your message or type the amount, currency, date and what it was for.

  1. Sparko shows one card, Expense from your receipt (or Expense claim when you described it). It lists the merchant, date, amount, category, the tax, any policy flags and the Approver the claim would go to, named.
  2. Check it. If something is wrong, choose Not now, nothing is filed, and you can describe the expense again.
  3. Click Confirm. Sparko files the claim and submits it to the approver on the card, and the card shows Open claim. A small claim under your company's automatic limit is approved at once. A claim that breaks a blocking rule, such as a category limit set to block, cannot be confirmed, and the card says why.

Sparko only files for you and only from your own receipts. It cannot file for someone else or use a receipt that is not yours. Whatever it reads from a receipt is treated as data, so text printed on a receipt cannot give it instructions.

What it costs: on Core, reading a receipt uses an AI action only when Sparko needs a model; reading it with rules uses none. The chat message uses its own AI action. Confirming uses one assistant action on Core and Growth. Paid Business and Custom plans have no action limit; a Business trial uses its whole-trial allowance. Confirming the same card twice files one claim and counts once. Asking about your claims uses no assistant action. If AI is off or your allowance is used up, claim in Payroll › Expenses by hand.

Acting on flags and pay anomalies

On Business and Custom, a flight-risk flag or a pay anomaly has an Act button: on the High flight risk card in Turnover & Retention, in the Compensation anomaly inbox, and in My Team › Compensation. You can also ask Sparko AI in chat, for example "What can I do about Daniel's flag?".

  1. Act opens a panel with why the person was flagged and the actions you may take.
  2. Pick one action. You can confirm one action per flag:
    • Propose a pay change. You type the increase as a percentage. Sparko never suggests a figure. The change goes into the normal approval chain.
    • Schedule a 1:1 with their manager at a time you pick. The manager is notified unless you are the manager.
  3. The panel shows what will happen, such as who approves first or when the meeting is.
  4. Click Confirm. The row shows what was done, and the score or signal is saved with your action.

If you manage the person, your own approval step is skipped and the pay change goes to the approver above you. If nobody is above you, it goes to an HR admin.

You can't act on your own flag, and you only see people you already may see. An acted pay anomaly leaves the inbox for 30 days.

What it costs: opening the panel uses nothing. Confirming is one assistant action, and Business and Custom have no limit on those. Act works even when AI is turned off for your workspace, because it calls no AI model. Confirming the same choice twice does it once and counts once.

Troubleshooting

AI Not Responding

  1. Check your internet connection
  2. Refresh the page
  3. Try logging out and back in
  4. Contact support if issue persists

"Sparko AI could not finish that request"

Something went wrong on Sparko's side. The turn is marked as failed and comes with a Retry button, so you can send the same question again without retyping it. A failed turn is not treated as an answer: it is left out of what the assistant remembers, and out of a saved conversation, so a follow-up is not built on it.

"Sparko AI is temporarily unavailable"

The model provider is busy or unreachable. This is not a plan limit and nothing is counted against your allowance. Use Retry in a moment.

Unexpected Answers

  1. Rephrase your question
  2. Provide more context
  3. Try being more specific
  4. Use "I meant..." to clarify

AI Made a Mistake

  1. The AI can make errors - always verify important info
  2. Click "Report issue" on any response
  3. Provide feedback to improve accuracy

"You have used this month's AI"

Your plan includes a set number of AI actions and assistant actions each month, shared across everyone in your company. On Core, asking Sparko a question that needs an AI answer uses one AI action, and having it carry out a task additionally uses one assistant action. On Growth only the tasks it carries out are counted, as assistant actions. When a pool is empty you get a short note with the date it comes back, plus what a higher plan includes.

Searching, browsing and dashboards never use an action. The chip at the bottom of the sidebar shows how much is left. Full detail is in the AI Governance & Allowance guide.

"AI is turned off for your workspace"

This is not a plan limit. An admin has switched AI off in Company Settings, either globally or for one area such as Recruiting. Ask an HR admin or Super Admin to turn it back on.

Answers From Your Company's Documents

When you ask about a policy or how something works, the assistant answers from real documents rather than guessing, and shows you where each answer came from.

This works on every plan. On Core it uses one of your monthly AI actions, the same as any other question you ask Sparko.

What it reads

  • Your company handbooks (including PDF handbooks)
  • Policy announcements
  • Sparko product help, for "how do I..." questions

What you will see

Ask "how much parental leave do I get?" and you get the answer plus a Sources line naming the document and section, for example Employee Handbook › Parental Leave. You can check the source yourself.

When the documents do not cover it

The assistant says it could not find it rather than inventing an answer. If that happens on something your handbook does cover, ask an HR admin to check the handbook is uploaded and current.

What you can see

Only what you can already open. Handbooks follow the same permissions and location rules as the Handbooks page, so you are answered from the handbook that applies to you. Announcements follow announcement permissions. Product help is available to everyone.

Keyboard Shortcuts

Shortcut Action
Cmd/Ctrl + K or / Open search, where you can ask Sparko
G then A Open Sparko AI
Esc Close search or a dialog
Enter Send message
Shift + Enter New line in message

Related Features

Still need help? Reach us at [email protected].

← Back to Help Center