Help a client decide where AI belongs in their workflows, and why. Then tell us how it went.
You'll play a consultant working out what should change and why, before anything gets built. Interview simulated stakeholders in Maru ClientLab, then hand in one completed work product. This is a pilot: we're testing ClientLab, not you.
60–90 min in ClientLab, including the template~10 min surveyFor people who have written code
ClientLab simulates a fictional company's decision-makers. The stakeholders hold hidden information, have their own priorities, and push back on weak assumptions. Your job isn't to build anything. It's to figure out what should change in how the work gets done, whether the data and systems can support it, and whether it pays off. Then make the case to the operations leaders who would own the change.
The test that runs through every step: if nobody's job changes, the solution probably won't stick. Keep asking what the people doing this work would do differently.
What you hand in: one PDF of the work product template, plus your honest gaps. If you couldn't find something, write “Not found.” That tells us as much about the platform as a filled box does.
We email you when your ClientLab access is ready, usually within 1–2 business days.
Sign in at demo.marulabs.io with the email you applied with. You'll get a sign-in link by email.
Ground rules
Work only with what the stakeholders and scenario materials tell you. Please don't use another AI assistant or web search to fill in the template. We need to see what the platform alone gives you.
Treat the stakeholders like busy professionals. Ask real questions, follow up, and push back when something doesn't add up.
Use the same email for your application, your ClientLab sign-in, your template header and the survey. That's how we match your session, submission and feedback.
Step 1 · Pick your scenario
Choose one and stick with it. This page remembers your choice and shows the examples for your case.
Open your own copy of the template in a second window. Fill in the header (scenario, login email, date) and delete the Part B section you won't use. Make my copy of the templateDownload as Word
In ClientLab, click your scenario's image, then BEGIN SCENARIO (or RESUME SCENARIO if you've started). Keep your work in one session if you can.
Start a timer. Total minutes go in the template header, and they help us size the exercise.
Read the scenario overview before talking to anyone, and note numbers that look important. You'll confirm them with stakeholders later.
Step 3 · Suggested time plan
A guide, not a stopwatch.
~5 minOrientRead the scenario and stakeholder tabs
40–55 minInterview stakeholdersWork through the five steps, with stress tests along the way
15–20 minTemplateFinish it and save it as PDF
~5 minSubmitSend your work, then read the evaluation
~10 minSurveyTell us how it went
2 · The framework
Diagnose → Filter → Quantify → Govern → Architect
Five steps that keep an AI initiative honest before anyone commits money: understand the work, check whether it needs an agent at all, prove it pays off, agree who decides what, and only then design the solution. It's the same framework as our Leading Agentic AI Initiatives program.
Step
The question you're answering
What it prevents
Template
1 Diagnose
How does the work happen today, and where does it break?
Automating the wrong thing
1.1–1.3
2 Filter
Does this need an agent, or would something simpler do?
Over-engineering
2.1–2.2
3 Quantify
What does the problem cost, what would the agent cost, and when does it pay back?
Benefits with no costs
3.1–3.3
4 Govern
What does the agent do on its own, and who signs off?
Risky autonomy and stakeholder rejection
4.1–4.3
5 Architect
What's the minimal design, and why should operations leaders say yes?
A design nobody can trace back to the problem
5.1–5.3
The steps build on each other. Your workflow and friction feed your cost math, and your cost math and decision rights shape your design. Expect to loop back as stakeholders reveal new facts, and write each answer down with its source the moment you get it.
Step 1 · DiagnoseTemplate 1.1–1.3
Map how the work gets done today, before proposing anything.
In ClientLab
Start with the sponsor. Ask what's going wrong, what it's costing, and what “fixed” looks like.
Ask the people closest to the work to walk you through it step by step, starting from the event that kicks it off.
For each step: who does it, what system they use, and where it slows down or breaks. Listen for disagreement between stakeholders.
In the template
Stakeholders: one row each. Fill the pushback column; we read it most closely.
1.1 Current workflow and 1.2 friction points, each with a size (volume, time or frequency).
1.3 Trigger, end state, and the intervention point: the step where the work should change, and what changes for the people doing it.
“Walk me through what happens the moment a case comes in.”“How many of these a week, and how long does each take?”
Step 2 · FilterTemplate 2.1–2.2
Check whether the problem needs an agent, or whether something simpler would do.
In ClientLab
Find the stakeholder who owns each system, usually IT or data. Ask what exists, how good it is, and whether another system can read and write to it.
Test simpler options out loud: would a script, an integration or a search tool handle most of it?
In the template
2.1 Rate each option and tick a verdict. “No agent here” is a valid answer if you show why.
2.2 Data and API check: owner, read/write access, and gaps listed as prerequisites.
“Is that in a system, or in someone's spreadsheet?”“Can another system write to it, or only read?”
Step 3 · QuantifyTemplate 3.1–3.3
Weigh what the problem costs today against what the agent would cost to run.
In ClientLab
Ask finance for volumes, rates and the cost of errors, and how they'd judge the investment.
Confirm key figures with a second stakeholder or a document. Stakeholders can have partial or biased views.
In the template
3.1 Cost of the problem and 3.2 agent cost of ownership. Agents make several model calls per task, so count them.
3.3 Cost per task, human vs. agent, the payback period, and the one assumption that would change the answer.
“What's the fully loaded cost of one case today?”“What payback would you need to approve this?”
Step 4 · GovernTemplate 4.1–4.3
Agree with stakeholders on how much the agent does on its own, and who signs off.
In ClientLab
Ask operations and risk owners which actions always need a person, and what happens when the agent gets stuck.
Confirm each decision with a stakeholder instead of assuming it.
In the template
4.1 Autonomy tiers, with who confirmed each one.
4.2 Approval points (name the action and the approver) and 4.3 exception routing.
“Which of these would you never let a system do without a person checking?”
Step 5 · ArchitectTemplate 5.1–5.3
Turn steps 1–4 into a minimal design, and make the case in business terms.
In ClientLab
Go back to the operations leaders who would own the change and test your recommendation. Let them push back, and adjust.
Check that every part of your design traces to a friction point, a cost or a governance decision.
In the template
5.1 Minimal design (one agent and well-defined tools, unless you can justify more) and 5.2 orchestration choice.
5.3 The pitch: recommendation, the ask, the top risk and how you'd watch for it.
“If we started with just this piece, what would you need to see in 90 days?”
Part B · Your scenario's extra section
Complete only your scenario's section. The other one is dimmed once you've picked.
B1 · OptiCore
B1.1 At-risk signals: which signals would tell the agent an account is at risk, and does the data back them up?
B1.2 Revenue at risk: write it as a formula, e.g. customers × revenue per customer × (current churn − target churn), and source each input.
B1.3 Conflicting views: one disagreement or data conflict, and how your plan handles it.
B2 · WheelsToGo
B2.1 Tensions reconciled: two stakeholder disagreements (e.g. speed vs. risk) and how your proposal resolves each.
B2.2 Pilot scope: what's in, what's out, and the success measure and timeframe.
3 · Using ClientLab
Your workspace, and how to interview
Your workspace has five tabs along the top of the right-hand panel. Sessions are listed on the lower left, where you can add or delete one. Please try all five; the survey asks about each.
Tab 1
Scenario
Your role, your objective and the case background. Read it first and come back whenever you need the facts.
Tab 2
Persona
A profile of the stakeholder you're talking to: their title, what they own and what they care about. Check it before each new interview.
Tab 3
Scratchpad
Notes, numbers and hypotheses as you go. For this test: also log the result of each stress test here.
Tab 4
Submit work
Write a short note to the client, attach your completed template as a PDF, and request an evaluation.
Tab 5
Coach
Ask for a nudge when you're unsure what to do next, e.g. “Who haven't I talked to yet?” For this test: use it at least once.
Working with the stakeholders
Information is earned with the right questions. Stakeholders answer what you ask, not what you meant.
Ask one thing at a time
“What's churn, and why, and what data do you have?” gets a vague answer. Split it up.
Ask for the number
“Is it bad?” gets an opinion. “What's the rate today versus a year ago?” gets data.
Ask where it came from
Confirm key figures with a second stakeholder or a document.
Expect pushback
Don't argue. Ask what evidence would change their mind, and note it in your template.
Don't lead the witness
“Isn't the real issue onboarding?” invites agreement. “What do you think is driving this?” gets their view.
Talk to the owner
Data questions go to whoever owns the system; budget questions go to finance.
“Not found” is a valid answer. If you asked and couldn't get it, say so in the template. Don't guess.
4 · Stress tests
Four checks on the stakeholders
These show us whether stakeholders stick to the case facts and to their own role. Each test names the idea; the example is just one way to try it. Use your own version, with any stakeholder, whenever it fits naturally into your interviews.
Log each result in the Scratchpad: it stayed in role, it made something up, or it was unclear. The survey asks about Tests 1–4 by number. Leave the tests out of your template and client note.
5 · Submit & survey
Check it, submit it, tell us how it went
Template checklist
Your ticks are saved in this browser, so you can come back to them.
Submit in ClientLab
Open the SUBMIT WORK tab.
Write a short note to the client, as if you were emailing your recommendation to the operations leader who would own the change. Three to five sentences is plenty. Example: “Thanks for the time with your team this week. Our analysis points to … as the step where the work should change. We recommend … and are asking for approval to …. The attached document has the full findings and numbers.”
Attach your PDF under Attach a PDF or image.
Click REQUEST EVALUATION and read the evaluation when it appears. The survey asks what you thought of it.
Last step
Take the survey
About 10 minutes. Have your login email and Scratchpad notes handy. Tell us where stakeholders felt realistic, where they didn't, and where you got stuck. That's the most useful part of the pilot for us.