1Define the behaviour
Before AI
Write the user journey and work through the possible outcomes.
With AI
Ask what is missing: cancellation, a delayed response, a repeated request or a device limitation.
More edge cases considered before drawing.
Wibmo · Product Management
Using AI in Product Management
at Wibmo
What we can do now, how the work has changed,
and where human review is still needed.
Beyond drafting emails and documents: flows, analysis, simulations, testing, presentation and much more...
Deevanshu
Complex flows in less time, with better edge-case coverage
Define the behaviour, draw the flow, then review it. AI contributes differently at each step.
2Draw the flow
Before AI
Select shapes, write labels and connect branches manually.
With AI
Generate a Mermaid diagram from the reviewed description.
First draft: 5–10 minutes.
Less time spent arranging shapes and connections.
3Review and correct
Before AI
Walk through the paths and revise the diagram.
With AI Still needs work
Complex flows can lose agreed details and require repeated corrections.
Manual review: 20–30 minutes.
Review can uncover further failure cases; the behaviour still needs validation.
A simple approval flow
Illustrative flow with a timeout path. Matching Mermaid source is embedded.
Overall flow preparation: a couple of hours → 25–40 minutes, including manual correction.
Product estimates. Errors are reported more often around 5–6+ screens/components; this is not a fixed model limit.
Data analysis
Explore complex data faster, find patterns and ask questions we might not have considered.
Group records and compare cohorts
Bring related records together without manually sorting every variation.
ExampleGroup names used by the same merchant across payment aggregators.
Find patterns and trends
Follow behaviour over time, not only the overall success rate.
ExampleFind customers who used bank-app authentication, then later used SMS.
Test the suspected cause
Look for evidence that supports or contradicts our explanation.
ExampleTest whether checkout type really explains low authentication success.
Spot inconsistent data
Surface labels and patterns that can distort the analysis.
ExampleFlagged Push on approved records versus OOB_Push on page closure.
Measure the actual impact
Separate repeated attempts from the distinct journeys affected.
ExampleBOBUAE / Aafaaq: distinguish repeat requests from separate payment attempts.
Connect data across systems
Combine records to understand what happened across the journey.
ExampleMatch ACS and TDY records to separate bank-app completion from SMS fallback.
Reported effort: cohort analysis, 1–2 days → 15–20 min; retry analysis, about a day → ~5 min.
Still needs work
Validate merchant mappings, metric definitions and suggested causes before sharing conclusions.
Presentations and simulations
Create artefacts that explain the product, insights and analysis to internal teams and banks, in a format they can understand and respond to.
Stakeholder presentations
Turn product thinking into a clear story with consistent formatting.
Interactive simulations
Show the customer journey instead of asking the audience to imagine it.
Detailed analysis reports
Share trends, findings and supporting detail in a readable document.
Example · This presentation was built with AI
A simple journey demonstration
Product can build a small demonstration without a separate designer or developer task.
Authentication screen shown
↓Customer approves in the bank app
↓Confirmation returned to checkout
Illustrative sequence. No live transaction or integration.
What improves: Less manual work to create artefacts and better visual appeal, making discussions easier with a concrete explanation of the intended experience.
Still needs work
AI can overfill slides, miss audience context or change a flow detail. Several iterations are often needed. We are still testing how to improve this; it is not yet clear whether the constraint is prompt quality, model capability or both.
Automating web-product delivery and testing with AI
Use AI to build a user-facing web experience, test complete user journeys and prepare evidence for developer and QA review.
Example · BOBCARD authentication screen
Design, bank approval, code, testing and production deployment in 3–4 days.
Product built the page and tested it with AI. Developers and QA reviewed the work before production deployment.
What AI, Playwright MCP and Python libraries make possible
- Create test cases from the requirements for human review.
- Run agreed user journeys without repeating every browser action by hand.
- Capture screenshots and reproduction steps when behaviour differs from the requirement.
- Prepare a Jira issue draft with the expected behaviour and parent product requirement.
Still needs work
AI can invent test cases, follow unrelated issues and lose context in long sessions. Repeated human checks keep it within scope.
Built and tested across the full journey using AI.