Research
/
Study
The research behind the moves
The audit, the user study with 18 people, and the before-and-after tests that shaped each move and measured it.
Feature growth · May to October 2026
Method
A four-phase loop, weighted toward analysis
Understand the operational user first, then frame, build, and validate against the original.
Show the 4 phases
01
Analysis
Understand
IA · Heuristics
Competitor review
SUS · NASA-TLX
02
Ideate
Frame
Interpret data
Opportunities
Problem statement
03
Design
Build
User flows · UI
Design system
Hi-fi prototypes
04
Validate
Prove
Effectiveness
Efficiency
NASA-TLX · SUS
Analysis
/
Inspection method
Seven of ten heuristics show friction. Two critical.
Walked against Nielsen's ten heuristics, from first landing to the working map, plus map- and data-specific principles.
01
Weak
02
Weak
03
Weak
04
Weak
05
Not yet
06
Fail
07
Not yet
08
Fail
09
Not yet
10
Missing
01
Visibility of system status
No loading states on layer changes; model update times not surfaced.
Weak
02
Match with the real world
Raw meteorological jargon, with no translation to user outcomes.
Weak
03
User control & freedom
Auth flow traps users; no back navigation; no exploration before commitment.
Weak
04
Consistency & standards
Mixed visual languages (photo thumbnails, flat UI, dense chrome) with no system.
Weak
06
Recognition over recall
Unreadable layer icons; users must remember what each model means.
Fail
08
Aesthetic & minimalist design
Overloaded map interface; critical information has no visual priority.
Fail
10
Help & documentation
No onboarding, tooltips or model explanations anywhere in the interface.
Missing
#5 Error prevention, #7 Flexibility and #9 Error recovery need deeper task flows; flagged for future round.
Analysis
/
Landscape
Where WeatherEx sits
Windy wins on maps, Tomorrow.io on alerts. Neither does both simply.
Show the 3 tools
Windy
Gets right
Visualisation and easy model switching.
Falls short
Little support for operational alerting.
Tomorrow.io
Gets right
Enterprise alerting and APIs.
Falls short
End-user UI is data-dense and configuration-heavy.
The Weather Company
Gets right
Consumer app and enterprise services from the same data.
Why it matters
The clearest reference for what general and operational audiences each expect.
Analysis / USER STUDY
User tasks
Task 01
Pick the best day
You and your friends are planning a trip to Matheran, Maharashtra. Check WeatherEx to find the best day next week.
Tests · comparing days
Task 02
Is tomorrow safe?
You are shifting house tomorrow. Check WeatherEx for any weather alerts and confirm everything is OK.
Tests · reading alerts and severity
Task 03
Go or wait, today
It's time to add fertiliser to your outdoor plants. Check the weather before you do it today.
Tests · reading today's forecast
Analysis
/
Enquiry method
Baseline measures
18 users: weather enthusiasts and general public. Each session ran as an interview, three observed tasks, then SUS and NASA-TLX. The numbers below are the original platform's starting line.
1
Interview
Who they are, what they decide
Role, the weather calls they make each day, and the tools they use now.
2
Observed tasks
Three tasks on the live product
I watched where they went first and where they stalled.
3
Questionnaires
SUS and NASA-TLX
Perceived usability and workload, scored per participant.
SUS · original
52.5
/ 100
68
Under the 68 baseline: short of comfortably usable.
Effectiveness
60
%
Share of tasks completed.
Efficiency
8.0
mins / task
Speed to task goal.
NASA-TLX · original
71.0
/ 100
Perceived workload. Lower is better.
What I watched users do
The scores say how bad it felt and the sessions showed where it broke.
Search
Panned and zoomed to find the location. Search went unused because nobody saw it as search.
Layers
Picked layers by trial and error. Users had to memorise the icons and buttons.
Timeline
Lost their place after every interaction. With no "now" marker, they dragged the scrubber back to re-orient.
Models
Could not say why one model beat another in a particular situation, so the model picker added doubt.
Auditing the original interface
Findings across the map interface and auth flow, ordered by severity.
FIG. 01 — ORIGINAL · DESKTOP MAP WORKSPACE · WEATHEREX.AI
01
Everything shouts at once
Login, About, menu and wind toggle sit at the same weight as the map. No clear home state.
02
Layers in an unknown language
Photo thumbnails don't convey much in the UI. Eight parameters in one flat, ungrouped stack.
03
Search hides in a corner
A 34 px icon, no field. Users panned and zoomed to find their own site instead.
04
Models without meaning
GFS 22 km, ECMWF 9 km, ICON 13 km. No default, no source, no timestamp, no reason to pick one.
05
No "now" on the timeline
The scrubber has no present-time marker, so users lose their place after every drag.
FIG. 02 — ORIGINAL · MOBILE WEB · SAME PROBLEMS, LESS ROOM
01
Account buttons own the screen
Login and Home are the two largest elements. The forecast, the thing users came for, has no surface at all.
02
Labels disappear entirely
Seven unlabelled photo thumbnails. A half-open second column floats over the map with no way to tell what it belongs to.
03
Search is one icon among eight
Same size and style as the layer thumbnails beneath it, so it reads as another layer.
04
Model names and grid sizes, nothing else
"ECMWF 9km" means nothing to a port operator. "2 more…" hides the rest.
05
A timeline with no present
The scrubber shows a date tooltip but never marks "now", and the fifth day is cut off at the edge.
Ideate
/
Opportunities
Nine opportunities, each traceable to a problem above.
The problem
The move
06
Value invisible at entry
→
Lead with what the product does for you
One line that answers "what does this do for me", then the answers users can get.
03
Search unnoticed
→
Search as the primary entry
"Search your port, site or region." Prominent, focused, auto-suggesting.
01
Density without hierarchy
02
Incoherent layer selection
→
Reduce map chrome
Layers collapse into role-grouped toggles with readable icons and explicit selection states.
01
Density without hierarchy
→
Show less by default
A minimal default view. Detail surfaces only at the point of need.
04
Models confuse
→
Show why a model can be trusted
Validation, sources and update times made visible.
04
Models confuse
→
Say what the weather means for the job
"Rain expected 1400 hrs, crane ops at risk."
07
No way back
08
Missing affordances
→
Let users in before asking for details
Google / LinkedIn sign-in; work-email and industry fields wait until after first value. Add the password toggle.
H10
Help & docs missing
→
Onboarding on a real task
The first session walks one real task, search to insight in three steps.
09
No differentiation
02
Incoherent layer selection
→
A visual identity of its own
Synoptic briefing-sheet aesthetic with strict brand–data colour separation.
Weather intelligence that answers questions, not one that just shows data.
Any user should reach their core operational question without training, and the product should meet them there.
What the product shows
Gust
35 kt
Rain · 3 h
12 mm
Wave Hs
1.8 m
Models
5 runs
↓
What the user is asking
Can I
X
run cranes
at
Y
JNPT Berth 4
on
T
Sat 06:00
?
Answered in seconds, without training.
A measured redesign.
A growing product,
sharper decisions.
Continue reading
Part 3 of 3
Research
/
Study
The research behind the moves
The audit, the user study with 18 people, and the before-and-after tests that shaped each move and measured it.
Feature growth · May to October 2026
Method
A four-phase loop, weighted toward analysis
Understand the operational user first, then frame, build, and validate against the original.
01
Analysis
Understand
IA · Heuristics
Competitor review
SUS · NASA-TLX
02
Ideate
Frame
Interpret data
Opportunities
Problem statement
03
Design
Build
User flows · UI
Design system
Hi-fi prototypes
04
Validate
Prove
Effectiveness
Efficiency
NASA-TLX · SUS
Analysis
/
Inspection method
Seven of ten heuristics show friction. Two critical.
Walked against Nielsen's ten heuristics, from first landing to the working map, plus map- and data-specific principles.
01
Visibility of system status
Weak
No loading states on layer changes; model update times not surfaced.
02
Match with the real world
Weak
Raw meteorological jargon, with no translation to user outcomes.
03
User control & freedom
Weak
Auth flow traps users; no back navigation; no exploration before commitment.
04
Consistency & standards
Weak
Mixed visual languages (photo thumbnails, flat UI, dense chrome) with no system.
06
Recognition over recall
Fail
Unreadable layer icons; users must remember what each model means.
08
Aesthetic & minimalist design
Fail
Overloaded map interface; critical information has no visual priority.
10
Help & documentation
Missing
No onboarding, tooltips or model explanations anywhere in the interface.
#5 Error prevention, #7 Flexibility and #9 Error recovery need deeper task flows; flagged for future round.
Analysis
/
Landscape
Where WeatherEx sits
Windy wins on maps, Tomorrow.io on alerts. Neither does both simply.
Windy
Gets right
Visualisation and easy model switching.
Falls short
Little support for operational alerting.
Tomorrow.io
Gets right
Enterprise alerting and APIs.
Falls short
End-user UI is data-dense and configuration-heavy.
The Weather Company
Gets right
Consumer app and enterprise services from the same data.
Why it matters
The clearest reference for what general and operational audiences each expect.
Analysis / USER STUDY
User tasks
Task 01
Pick the best day
You and your friends are planning a trip to Matheran, Maharashtra. Check WeatherEx to find the best day next week.
Tests · comparing days
Task 02
Is tomorrow safe?
You are shifting house tomorrow. Check WeatherEx for any weather alerts and confirm everything is OK.
Tests · reading alerts and severity
Task 03
Go or wait, today
It's time to add fertiliser to your outdoor plants. Check the weather before you do it today.
Tests · reading today's forecast
Analysis
/
Enquiry method
Baseline measures
18 users: weather enthusiasts and general public. Each session ran as an interview, three observed tasks, then SUS and NASA-TLX. The numbers below are the original platform's starting line.
1
Interview
Who they are, what they decide
Role, the weather calls they make each day, and the tools they use now.
2
Observed tasks
Three tasks on the live product
I watched where they went first and where they stalled.
3
Questionnaires
SUS and NASA-TLX
Perceived usability and workload, scored per participant.
SUS · original
52.5
/ 100
68
Under the 68 baseline: short of comfortably usable.
Effectiveness
60
%
Share of tasks completed.
Efficiency
8.0
mins / task
Speed to task goal.
NASA-TLX · original
71.0
/ 100
Perceived workload. Lower is better.
What I watched users do
The scores say how bad it felt and the sessions showed where it broke.
Search
Panned and zoomed to find the location. Search went unused because nobody saw it as search.
Layers
Picked layers by trial and error. Users had to memorise the icons and buttons.
Timeline
Lost their place after every interaction. With no "now" marker, they dragged the scrubber back to re-orient.
Models
Could not say why one model beat another in a particular situation, so the model picker added doubt.
Auditing the original interface
Findings across the map interface and auth flow, ordered by severity.
FIG. 01 — ORIGINAL · DESKTOP MAP WORKSPACE · WEATHEREX.AI
03
02
01
04
05
FIG. 02 — ORIGINAL · MOBILE WEB · SAME PROBLEMS, LESS ROOM
01
02
03
04
05
01
Everything shouts at once
Login, About, menu and wind toggle sit at the same weight as the map. No clear home state.
ON PHONE
Account buttons own the screen
Login and Home are the two largest elements. The forecast, the thing users came for, has no surface at all.
02
Layers in an unknown language
Photo thumbnails don't convey much in the UI. Eight parameters in one flat, ungrouped stack.
ON PHONE
Labels disappear entirely
Seven unlabelled photo thumbnails. A half-open second column floats over the map with no way to tell what it belongs to.
03
Search hides in a corner
A 34 px icon, no field. Users panned and zoomed to find their own site instead.
ON PHONE
Search is one icon among eight
Same size and style as the layer thumbnails beneath it, so it reads as another layer.
04
Models without meaning
GFS 22 km, ECMWF 9 km, ICON 13 km. No default, no source, no timestamp, no reason to pick one.
ON PHONE
Model names and grid sizes, nothing else
"ECMWF 9km" means nothing to a port operator. "2 more…" hides the rest.
05
No "now" on the timeline
The scrubber has no present-time marker, so users lose their place after every drag.
ON PHONE
A timeline with no present
The scrubber shows a date tooltip but never marks "now", and the fifth day is cut off at the edge.
Ideate
/
Opportunities
Weather intelligence that answers questions, not one that just shows data.
Any user should reach their core operational question without training, and the product should meet them there.
What the product shows
Gust
35 kt
Rain · 3 h
12 mm
Wave Hs
1.8 m
Models
5 runs
→
What the user is asking
Can I
X
run cranes
at
Y
JNPT Berth 4
on
T
Sat 06:00
?
Answered in seconds, without training.
Ideate
/
Opportunities
Weather intelligence that answers questions, not one that just shows data.
Any user should reach their core operational question without training, and the product should meet them there.
What the product shows
Gust
35 kt
Rain · 3 h
12 mm
Wave Hs
1.8 m
Models
5 runs
→
What the user is asking
Can I
X
run cranes
at
Y
JNPT Berth 4
on
T
Sat 06:00
?
Answered in seconds, without training.
Ideate
/
Opportunities
Nine opportunities, each traceable to a problem above.
06
Value invisible at entry
↓
Lead with what the product does for you
One line that answers "what does this do for me", then the answers users can get.
03
Search unnoticed
↓
Search as the primary entry
"Search your port, site or region." Prominent, focused, auto-suggesting.
01
Density without hierarchy
02
Incoherent layer selection
↓
Reduce map chrome
Layers collapse into role-grouped toggles with readable icons and explicit selection states.
01
Density without hierarchy
↓
Show less by default
A minimal default view. Detail surfaces only at the point of need.
04
Models confuse
↓
Show why a model can be trusted
Validation, sources and update times made visible.
04
Models confuse
↓
Say what the weather means for the job
"Rain expected 1400 hrs, crane ops at risk."
07
No way back
08
Missing affordances
↓
Let users in before asking for details
Google / LinkedIn sign-in; work-email and industry fields wait until after first value. Add the password toggle.
H10
Help & docs missing
↓
Onboarding on a real task
The first session walks one real task, search to insight in three steps.
09
No differentiation
02
Incoherent layer selection
↓
A visual identity of its own
Synoptic briefing-sheet aesthetic with strict brand–data colour separation.
A measured redesign.
A growing product,
sharper decisions.