Complex BI reports to decision ready presentations in 10 mins
How I designed Google’s first AI assisted workflow connecting Data Studio (formerly Looker Studio) with Google Slides - making automated storytelling reliable while preserving user control and visual fidelity.
Role:
Complexity:
Sole Product Designer in a team with PM, SW Eng, AI Eng & UXR
Two storytelling paths - AI generated and user curated - across 40+ chart types, within Google Slides' technical constraints.
Outcome:
Impact:
Data Analyst’s effort has been cut down from 3 hours to 10 mins
Demoed at Google Cloud NeXT for consecutive years. Preview released to selective enterprise customers.
Why BI reports don't become decision-ready stories on their own?
Business Intelligence (BI) means converting the raw business data into useful insights for decision making.
BI reports bring together multiple pages of charts, visualizations and active filters. They organize the evidence, but rarely turn it into a clear, decision ready narrative.
Examples of BI dashboards in the industry:
Publicly available illustrative examples, rights belong to their respective owners.
The persisting problem
The insights are buried within these visualizations. Analysts still have to interpret what matters and turn into a narrative the leadership can act on.
Hence data analysts have to spend hours converting BI reports into Google Slides with insights & summaries for executive leadership.
This conversion effort is tedious, repetitive and undesired because there is no interoperability between Looker Studio and Google Slides.
The pain of manual narrative creation
Two powerful Google applications are in complete isolation, forcing users to use the clipboard as a crude data pipeline. This has been leading to a lot of manual effort in the entire workflow of converting the BI report content into executive slide decks.
Taking screenshots from BI reports
Choosing right size & ratio to fit slide layout
Stretched, pixelated visuals with broken links
Manually writing insights for each chart
Charts & insights mismatch during update
Repetition & huge loss of productive time
UXR Feedback on the existing method
Creating decks and manually updating them is a big pain and frustrating.
- An enterprise customer
Inter-operability is the need between reports & slides.
- Product forum
We already seamlessly work with PowerPoint.
- Power BI & Tableau
What began as a UX problem in time-saving revealed a strategic gap in Google’s data analytics ecosystem.
This was identified as an opportunity towards building the reliable decision intelligence platform while building the integrations with Google Workspace and that space was explored as a parallel effort.
A hard problem to solve
Technical constraints
Looker Studio architecture
Fragmented ‘AI for BI’ efforts
No partnership with Slides engg
+
AI challenges
AI given text was very verbose
Generated content had speculations
Evolving abilities made it difficult to plan
My Role
Sole designer for a highly ambiguous pioneering initiative with AI.
Responsibilities
End to end workflows
AI behaviour for UX
XFN alignment
Cloud NeXT demos
Collaborated with
PM
1
SW & AI Engg
4
UX-R & Writer
2
BI AI UX Teams
3
Principles for AI Behaviour
I consolidated the emerging product & design principles that made the AI to work within real constraints.
These principles emerged incrementally as reliability, user control, scalability and constraints were addressed. Each of these principles is not generic AI terms but have deeper meaning & purpose contextual to the BI storytelling project.
Below 9 seconds video reveals the meaning behind each Principle for this project with Reliability as core.
How evolving AI principles shaped a reliable generation pipeline
This retrospective synthesis shows how the end to end workflow is implemented from the BI report to the final Slide deck. AI behaviour is controlled by this to generate reliable output while retaining the ability of user judgement and refinement.
Design Decision #1:
A dedicated Gemini panel for centralized AI discovery
Initial feature specific question
Where should ‘Generate Slides’ live?
Natural choice for export functions
Under Share / More → Generate Slides
Now the question became platform level
Where should AI live in Looker Studio?
The decision
One dedicated Gemini panel
Shared alignment questions
Where should AI live?
How to align on terminology?
What should AI generate?
Centralized Discovery
One visible home for all AI workflows
Gemini Consistency
Aligned with AI patterns across Google
Non-intrusive Access
AI panel doesn’t cover report canvas
Cross Product Foundation
Supports workflows beyond Studio
Design Decision #2:
Two paths to generate Slides for better control.
UXR reports indicated that users approach BI stories with fundamentally different intents. Some wanted a quick presentation with charts and insights; others needed to curate with precise visuals, auto-insights and control the narrative.
I therefore designed two paths from the same Gemini entry point rather than forcing one fully automated flow.
Intent determines control
Shared enabler :
Add-on at Google Slides
Generate deck
Review in Slides
Update with Add-on
To review and update the content for both paths.
Able to scale multi chart, multi page and multi report content to a business presentation and vice versa.
Selection enabler :
Transient mode at Studio
Initiate selection path in report
Select in Transient mode
Generate deck
A new temporary interaction layer, a foundation for narrative building as per user discretion.
No architectural changes or new learning required at Studio.
Feedback
“All visualizations” was the most-used flow and was described as a breeze.
Response
Preserved it as the fastest route from a complete BI report to a presentation.
Feedback
Users wanted
Multi-chart selection
Partial selection of a chart
More charts per slide
A more intuitive flow
Response
Added
Drag to multi-select
Snip to partial select
Three visualization limit per slide
Improved chart to slide mapping
Drag to multi-select
One drag can select multiple charts and titles
Snip to select part of a chart
Snip selects a chart area and dims unselected
Template with 3 viz max per slide limit
Preserved legibility while supporting varied chart ratios.
Slide mapping with viz placeholders
Intuitive ‘Chart to slide’ mapping by thumbnail placeholders
Feedback
Users prioritised accuracy of data
Key insights with less verbosity
Ability to disable insights
Response
Prioritised grounded, concise insights
Reduced verbosity & highlighted key data
Selective path retained insights on/off
➤
70 seconds duration
With voice over
Implementation evidence
These publicly available early GIF demos verify that both workflows were functional. They reflect an earlier design stage, not the version released for preview.
The selected interface screens above all have been recreated for clarity.
‘All visualizations’ demo
‘Let me choose’ demo
Impact
Reduction in manual deck creation effort
From 3 hours → 10 minutes
Product demos and exposure
Cloud NeXT 2023 - 25
Generated in Preview
Thousands of presentations
Average size of automated presentation
12+ slides
Individually recognized by the leadership & team through 3 spot bonuses and 2 peer bonuses.
What Remained
The automated path reached preview and was positively received. The selective path established an important direction for narrative control, but still required cross functional refinement. Context awareness and continued storytelling in Slides remained beyond the preview scope.
The Reflection
This effort reinforced that enterprise AI needs progressive human control to delegate faster while retaining the ability to define scope and refine result.