Pattern · Product Design Internship
Custom Reports in Predict
YEAR · BY MONTH | AMAZON US | ALL | $USD
12
Create a New Report
More Templates- Blank Report
- Monthly Business Review
- Advertising Deep Dive
- Organic vs. Paid
- Buy Box & Pricing
- Ad Efficiency
- Traffic to Orders
- Peak Season Readiness
- Shared with group
Q1 Business Review · Aurora Wellness
- Shared with meAmazon US · Advertising Deep Dive
- PrivateBuy Box & Pricing Health
- Shared with groupOrganic vs. Paid · Weekly Tracker
- PrivatePeak Season Forecast Readiness
- Shared with meMarch Town Hall Report
How might we let brand managers build and share a client-ready report without falling back to Excel?
Validated through usability testing and shipped by the team after my internship. This page only claims what I observed before handoff.
- Discover
- Scope
- Design
- Test
- Handoff
Discover
Act 01: A one-line ticket exposed a much larger problem
My first assignment was a small ClickUp ticket: let users duplicate a widget so they would not have to rebuild the same configuration. It was estimated at one or two days.
As I worked through the flow, duplication started to look like a symptom. Users were cloning widgets because creating and editing reports was slow, rigid, and repetitive.
The ticket where it started
A one-to-two-day duplication ticket revealed problems across the entire reporting workflow.
Six reports, four of them copies the symptom, in the product
The Custom Reports list sat behind a BETA sub-tab on the Reports page, with no previews and no way to tell a report from its copies. The only way to reuse a configuration was to duplicate the whole report and edit the copy.
The old Custom Reports home: a bare list with no previews, where a report and its copies look identical. Building was harder than copying, so people copied.
This sparked a key question:
Why were people cloning widgets instead of building the report they needed?
The evidence the problem was not demand
I ran weekly discovery interviews throughout my Pattern internship. Reporting workarounds repeatedly appeared in conversations about other parts of Predict. I also held a focused session with four UK brand managers who had used Custom Reports.
Available product data showed that roughly 60% of eligible users created at least one report during the feature’s first year, but only about 15% returned. People understood the potential value but did not find the existing workflow useful enough to repeat.
Six recurring problems from the research
Two metrics per chart
Comparison stopped at two lines. Anything more meant Excel.
Filters locked at creation
Filters were unclear and could not be edited once a report existed.
Two confusing modes
Separate view and edit modes hid basic actions behind a switch nobody expected.
Look-alike controls
Buttons looked the same as each other, and some things that looked like buttons did nothing.
No words on a widget
A chart could not carry a note, so the explanation lived in the call.
No export
Sharing a report with a client meant a screen share, screenshots, or Excel.
The old tool viewing and editing, kept apart
- 1The ask, in the product
A test widget titled “Multiple metrics on a line chart?” The question was already on screen. - 2Two lines, then stop
Ad Clicks and ACOS. The third metric had nowhere to go. - 3Looks clickable
The cloud in every header was not a button. Neither were the metric dots. - 4Locked filters
Set once at creation, then out of reach. - 5Exit to edit
A pencil that leaves the report. Editing was a different place.
Two modes a switch nobody expected
Presenting: the chart, but no way to change a filter or a metric without leaving.
The two-metric limit was not a Custom Reports decision. It came from Predict’s charts everywhere: every page in the platform drew two metrics on two axes, and Custom Reports inherited it.
Predict’s own Traffic page. Two metrics, two axes, the same ceiling the report widget inherited.
Success meant a useful first report without Excel, with more control revealed only when someone needed it.
Who it was for two groups who reported every week
The brand manager
Builds the client report, and often edits a chart while presenting it on the call.
The advertising strategist
Compares more than two metrics at a time and needs the comparison to travel with the report.
Scope
Act 02: The most flexible concept was the one we could not build
I brought the evidence to the Predict product and design team. With my design manager and Director of Product, I scoped a nine-week project around the highest-value workflow: creating reports, configuring widgets and filters, and editing report content.
The largest constraint was engineering capacity. I could not introduce new data sources, replace Predict’s design system, or assume the team could rebuild the report builder from scratch.
Advanced users wanted a free canvas to drag charts and layer dimensions
A structured report with list-based reordering, and the canvas on the roadmap
I explored a Slides-style canvas. It matched the mental model of advanced users, but it exceeded the engineering budget. I presented the concept, user evidence, and tradeoffs. Leadership made the scope decision, and I focused on preserving the most important value inside a structured experience the team could ship. Users still gained control over report structure without requiring the team to build an entirely new editing system.
The canvas concept matched how advanced users thought about building a report, but it needed an editing system engineering could not build in nine weeks.
Before: a text list behind a BETA sub-tab. No previews, and no way to tell a report from its copies.
Design
Act 03: The redesign guided the first report, then got out of the way
Because many users did not return after their first attempt, the redesign focused first on making report creation easier to understand. The goal was not to expose every reporting option immediately. It was to help users create a useful first report and reveal more control when they needed it.
A first report could be silently limited by scope and filters nobody explained
A guided creation flow that explains scope and filters before they bite
The new home replaced a text list with templates and visual report previews. A guided creation flow collected the report name, scope, filters, and first widget before placing the user inside a working report.
Report scope and filters are explained in plain words before they can quietly limit the report.
Up to six metrics on one chart two axes, many kinds of number
Predict’s charts had two y-axes and stopped at two metrics. Users wanted to compare five or six, and to mix lines with bars. The hard part was not drawing more lines. It was that the metrics come in different units: dollars, counts, percentages, ratios, and a chart still only has two sides.
The rules I landed on: every metric has a unit. The first unit family on the chart takes the left axis, the second takes the right, and a third family is blocked with a reason rather than squeezed onto a scale that would lie. Each metric chooses line or bar. A switch lets same-unit metrics share one scale, or split across both axes when one would flatten the other.
Sales
$40.18M$2.84M7.62%▲
Ad Sales
$18.03M$1.39M8.38%▲
Pattern Page Views
3.08M95,5043.20%▲
True ROAS
2.420.3011.05%▽
Page Views
7.08M396,2415.30%▽
Incremental Ad Sales
$17.41M$1.26M7.80%▲
ROAS
2.510.2910.47%▽
- LSales
- LAd Sales
- RPattern Page Views
- RPage Views
Live. Change metrics, marks, or the axis switch and the chart, ticks, badges, and header follow the same rules.
The same widget carries a caption that travels with the chart, so a report can explain itself when it is shared instead of relying on whoever is presenting it. And the numbers on it agree with each other: ROAS on a card is the ad sales over ad spend that the chart is drawing.
Test
Act 04: Testing removed the boundary between viewing and editing
I tested the redesigned flows with 20 users: ten brand managers and ten advertising strategists. I expected them to prefer a clean separation between presenting a report and editing it. Testing showed the opposite. Brand managers often edited charts while presenting them to a client. Switching modes added friction at the moment they needed the tool to feel most responsive.
One combined mode the most important change testing produced
- 1Edit in context
The action sits on the widget. No mode switch. - 2Filters stay editable
Scope can change after the report exists. - 3More than two
The multi-metric chart, presented and edited in one place.
I removed the separate edit mode and combined viewing and editing into one experience. Actions appeared in context, filters remained editable after report creation, and the controls that looked like buttons now were buttons. It simplified the interface while supporting the way reports were actually used.
The starting templates came from a report-building competition with brand managers and advertising strategists, not from my assumptions about a good client report.
Participants used the new system to create the reports they found most useful in their actual work. The strongest entries became the templates in the product, and the competition gave me another round of feedback from the people most familiar with the workflow.
Create a New Report
More Templates- Blank Report
- Monthly Business Review
- Advertising Deep Dive
- Organic vs. Paid
- Buy Box & Pricing
- Ad Efficiency
- Traffic to Orders
- Peak Season Readiness
- Shared with groupQ1 Business Review · Aurora Wellness
- Shared with meAmazon US · Advertising Deep Dive
- PrivateBuy Box & Pricing Health
- Shared with groupOrganic vs. Paid · Weekly Tracker
- PrivatePeak Season Forecast Readiness
- Shared with meMarch Town Hall Report
Sales
$40.18M$2.84M7.62%▲
Ad Sales
$18.03M$1.39M8.38%▲
Pattern Page Views
3.08M95,5043.20%▲
True ROAS
2.420.3011.05%▽
Page Views
7.08M396,2415.30%▽
Incremental Ad Sales
$17.41M$1.26M7.80%▲
ROAS
2.510.2910.47%▽
- LSales
- LAd Sales
- RPattern Page Views
- RPage Views
Ad Clicks
2.26M129,9346.10%▲
ACOS
39.99%4.18 pts11.68%▲
Ad Sales
$18.03M$1.39M8.38%▲
Ad Spend
$7.19M$1.24M20.90%▲
Ad Orders
79,0007,50710.50%▲
CPC
$3.19$0.3913.82%▲
Ad Conversion Rate
3.50%0.12 pts3.60%▲
- LAd Clicks
- RACOS
Handoff
Act 05: Validated in testing and shipped after my internship
The redesign was validated through 20 usability sessions and shipped by the Pattern team after my internship. I delivered annotated designs, the prototype, supporting documentation, a prioritized backlog, and a follow-up testing plan.
- Reports with up to six metrics per chart, lines and bars together, on two axes that follow the unit rules
- One combined mode in place of separate viewing and editing
- Filters that can be changed after report creation
- A caption on every widget that travels with the chart
- Reports that can be reordered, duplicated, shared, and exported
- A template library seeded by user-built reports
- Annotated designs, prototype, documentation, backlog, and a testing plan
My first instinct was to include everything users requested. Trying to build all of it would have produced another complicated report builder.
The harder and more valuable work was identifying the changes that improved the core workflow for everyone, then fitting them inside what engineering could support.
The redesign still did not solve missing forecast, wholesale, or logistics data. Those were platform limitations, not interface problems, and they are what stands between Predict and being someone’s complete reporting tool.
What would come next?
- Send a report to the client from Predict, through Gmail, so the old routine of a screen share or screenshots ends. That flow is in design now.
- Measure whether users return to build a second report
- Investigate the forecast, wholesale, and logistics gaps that still send people to Excel
Read next
Find or summarize a document...
What are you looking for today?
Find or summarize a document...
Find docs
Locate files by person, timeframe, or canvas text
Summarize
Get the gist of a doc or a group of them
Build a diagram
Create flowcharts and visual layouts from a text prompt.
Catch up
Summarize your recent workspace changes
