
Radfords · Design system for fresh-produce software
Fewer scrolls on the packhouse floor 🍏
Reworking the Radfords design system for people who work between fruit bins and tablets.
Radfords runs fruit packhouses from orchard to dispatch. Its screens were full of long forms, roomy grids and grids nested inside grids. I reworked the design system those screens are built from, so every screen built from it asks less of the people using it.
Role
UX designer
Employer
Arcadea Group, assigned to Radfords
Worked with
Radfords product and management team
Scope
Internal design system: components, templates, spacing and colour
13
form sections reorganised into one view that shows each section’s progress
+18%
more grid rows per screen (17 to 20)
54px
of space freed on one form screen by switching Montserrat to Inter
3
overall problems resolved
Three problems, one place to fix them
Radfords asked for help with three problems that kept appearing across its app. They showed up on screen after screen because they came from the internal design system every screen is assembled from. Fixing the system meant fixing them everywhere at once.
Problem 1
Long forms
Adding a person or an item meant one long, linear list of fields in a modal, with a lot of scrolling.
Problem 2
Wasted grid space
Tall rows, large type and generous padding meant few records per screen and early scrolling.
Problem 3
Nested-grid complexity
Grids inside grids, sometimes four levels deep, made it hard to tell which row belonged to which.



From the existing system: the single-column form, grids, and the nested grids views.
Learning the business before touching a screen
It was up to me to understand the company and its users before solving anything, so I started with research, not Figma.
Radfords is a post-harvest software suite. Its modules cover grading and packing, quality control, sales and grower payments, labour and productivity, and reporting. Every module creates records with many fields, and every module lists them in grids. That breadth is why the three problems kept reappearing.
The company describes its product through five principles: intuitive, scalable, data capture, value and integration. I used the first as my yardstick. Radfords wants clear “next steps”, so that receiving fruit, setting up pack runs and managing finished goods feel obvious.
What clients said
I read what growers, producers and distributors say about the product. Before Radfords, they described double-handling data, counting bins on paper in the orchard, and a system “understood by only a few people”. They asked for fruit terminology anyone could use without training. After switching, they praised offline scanning, tablets on the shop floor and instant reporting.
Two concerns mattered most to me as a designer. Long-term staff were reluctant to adopt new technology, and the workforce changes every season.
Many people using Radfords are not desk workers, and many were never trained on the software.
Who uses it
User group
Where they work and what they need
Owners
Mostly in an office. Want oversight: calendars, reports, compliance, and confidence that things are happening as they should.
Managers
Between the field and the office. Manage people and equipment, review assessments, and prefer mobile to desktop software.
Administrators
At a desk and always busy. Set up and maintain master data quickly, and want fewer calls for support.
Quality controllers
Walking between stations. Enter samples fast and want to avoid preventable mistakes. Some prefer paper to software.
Field workers
Outdoors all day. Need simple, quick tasks with little or no training.
What that meant for the design
Screens must explain themselves
Training is scarce, so labels, progress and next steps need to be visible.
Every scroll costs time
Speed and convenience ranked high for almost every user.
Density must adapt
People move between desktops and tablets, so one fixed density won’t suit everyone.
Show what’s missing early
Preventable mistakes are a known frustration, so gaps should be visible before saving.
The system had the right parts, but no shared rules
Radfords gave me access to its design system: a card dashboard, three data-capture layouts, four modal types and a colour guide. Each template solved layout its own way.
The single-column form stacked every field with no sections and wide empty margins.
Grids gave ten columns equal width and tall rows, whatever their content.
A grid modal opened over screens that already had a grid, so nesting was built in.
The selector modal showed about 80 identical chips, and a selected chip was barely darker than the rest.
Across the modals, buttons were centred, left-aligned or aligned to the fields, and every button said “Default”.
The colour guide was 16 unlabelled swatches with no roles, and several greys used in the templates were missing from it.
From the existing system: the single-column form, the side panel with a ten-column grid, and the selector modal’s wall of chips.
Problem 1 · Long forms
Turning a 13-section form into one clear workflow
The test case was “Create Person”, the form for adding someone to the system. The challenge was to show many fields while using the screen well and sparing people a long scroll.
Four layouts on placeholder content
I explored four templates: a master-detail stepper, a multi-column accordion, a vertical accordion, and segmented progress tabs. All four shared two ideas that carried through to the end.
Progress per section. Each section shows two bars, for mandatory and optional fields completed. People can see what’s missing instead of meeting an unexplained disabled button.
Expandable section tiles in grid view. Each section can be expanded when required and each tile shows progress bars.
Accordion pattern. Each section is expandable and one after the other in a columnar view. It still has the progress bar.
Mandatory and optional side by side. A two-column split puts the old form’s empty margins to work and makes required fields obvious.




Four first explorations: the same content, broken up four ways.
Real data changed the direction
Placeholder content only goes so far, so I needed real form details. Radfords shared their current form. It had 13 sections, from Personal Details and Contact Details through Roles and Permissions, Bank Details and Medical Information to Facility Access and Grow. It also needed repeatable contacts, role chips, yes/no toggles, dates and a signature.
With that data I explored three workflows from the two strongest templates. Two were master-detail flows, with the person’s category picked from a dropdown or from visible tiles. The third was a vertical accordion, which turned out to be as long as the original form when fully expanded.
The biggest idea came from the real data: pick the person’s role first, and show only the sections that apply.

Master-detail flows with real data: category dropdown against category tiles, with colour and button-placement variants.

The vertical accordion. Expanded, it’s long again. Collapsed, all 13 sections fit as compact rows.
A component library for data input
To make the workflow repeatable across Radfords, I built the inputs as components, each defined in three states.
Active, not selected
The default state.
Selected or unsaved
Stronger borders and filled values, also used when a new detail is added but not yet saved.
Inactive
Greyed out, including its add and remove buttons.
The library covers text inputs, dropdowns, a signature pad, repeatable fields, removable role chips, toggles, date pickers and text areas, each with a label and helper text. A section-header component carries the section name and its progress, such as “Mandatory details: 2 / 2”.


The component library: every input in three states, plus the section header with progress.
The complete workflow
I designed every screen of “Create Person” using the master-detail flow with category tiles. The 13 sections run down the left with their progress, the selected section fills the right, and Cancel Form and Create stay in the same place on every screen.
Fields appear when they apply. In Facility Access, facility fields stay inactive until “Allow access to facilities?” is set to Yes.
Repeatable data stays in place. Extra contacts, roles and staff types are added without opening a new window.
The selector was rebuilt. “Add Roles and Permissions” swaps the wall of chips for module filters and roles grouped under each module.

Section by section, from contact details to facility access, including the grouped roles selector.
Testing the typeface
Radfords used Montserrat. I set the same screen in Inter and measured both.
Measurement
Inter
Montserrat
Title width
262 px
276 px
Section heading width
144 px
153 px
Helper text under a field
2 lines
3 lines
Free space below the last field
147 px
93 px
Inter saved about 5–7% of width per line and a line of helper text per field: roughly one more field before scrolling. I recommended Inter, with a condition. Montserrat still reads easily, so the switch needed testing with real users in their real work environments.

The same screen in Inter and Montserrat. Inter’s helper text takes two lines where Montserrat’s takes three.
Problem 2 · Grid density
Finding the one change that fits more rows
Radfords’ grids showed few records per screen. I started by recreating screenshots of the current grids in Figma, so every property could be changed and measured. The recreation kept what already worked: tabs, sortable columns, a filter button, green record links and edit actions on hover.
Heuristic evaluation
Working through the grid, I listed the issues and proposed six changes:
Responsive design: denser rows on desktop, larger ones on tablet and mobile.
Adjustable density: presets such as Default, Compact and Comfortable.
Smaller type: reduce 16 px grid text to a smaller size that stays legible, consistently.
Less padding around grid elements.
A collapsible filter bar for people who don’t need it.
Breakpoints: 1200 px and up for desktop, 768–1199 px for tablet, and up to 767 px for mobile.

The current grid (left) and a first denser version (right): thin dividers instead of zebra stripes, a header band and tighter rows.
Isolating the variable
Several changes were possible, so I tested three variables separately to see which one mattered: typeface, font size, and row height (the original 27 px against a height on an 8-point grid).
Typeface and size
Original 27 px rows
8-point grid rows
Montserrat, 16 px
17 rows
20 rows
Montserrat, 14 px
17 rows
20 rows
Inter, 16 px
17 rows
20 rows
Inter, 14 px
17 rows
20 rows
Only row height changed the count. Typeface and font size made no difference to how many rows fit.
That’s useful for the team. They can keep a comfortable font size and still gain about 18% more rows. A narrower typeface pays off across the width instead, where it reduces wrapping in a ten-column grid.
Choosing the final density wasn’t my call, and with so many devices and users it needed testing. I turned the findings into presentation-ready visuals and gave the Radfords team my recommendations to test and decide on.

The test matrix: in both typefaces and at both font sizes, 8-point rows fit 20 records against 17.
Problem 3 · Nested grids
Making it obvious which row belongs to which
Radfords gave me screenshots of real nested grids. They showed that nesting serves a purpose: reconciling a pack run means checking materials against scanned batches, and a spray diary row really does carry around 20 details. So the goal was clarity, not removing the nesting.

Four real screens: packrun reconciliation, assessment rules four levels deep, contacts with orders, and a spray diary with a form inside each row.
The complexity came from the same sources each time. Shallow indentation hid which row was the parent, stripes, borders and boxes competed, and every level had its own add, confirm and save controls. I wrote five principles to address them:
Simplify colour: one light background for child rows instead of alternating colours.
Indent child rows so the hierarchy reads at a glance.
Use borders and spacing to mark where each level starts.
Highlight on hover: hovering a parent highlights its children.
Collapse and expand individual parent rows to focus on one part of a large grid.
Packrun reconciliation
Grader rows get one light background and a filled dark dropdown icon; material rows get an outlined one. Because the icons already separate the levels, a slight indent is enough. Batch grids sit on their own light background, border lines replace stripes, and cancel is a red × beside a green tick.

Packrun reconciliation, before and after. The redesign adds a second grader to show the pattern holding up with more data.
Contacts and orders
Cell borders give way to row dividers and a grey header band. Numbers are right-aligned so prices and quantities can be compared, child grids are indented, and pagination moves from the grid component’s purple to neutral black.

Contacts and orders, before and after: fewer lines, aligned numbers, a clear hierarchy.
As with the grids, I set the packrun redesign in both typefaces and annotated every decision on two presentation slides for the Radfords team.

Add image: p3-annotated.webp
Recommendation slides: each annotation ties a visual change to the problem it solves.
Reusable parts, and evidence to decide with
The final design decisions belonged to Radfords, so I packaged the work for the product and management team to test and decide on. The response was positive.
Delivered
What it gives the team
Data-input component library
Every input in three states, plus a section header with progress.
Master-detail form template
A reusable way to break any long “add an item” form into sections.
Complete “Create Person” workflow
All 13 sections designed with real data, and a rebuilt selector.
Grid density study
Evidence that 8-point rows fit about 18% more, whatever the font.
Nested-grid redesigns
Two real screens, with annotated slides explaining each change.
Typeface comparison
Measured savings from Inter, with a recommendation to test with users.
What I’d do differently
If I improved one thing now, it would be colour contrast. I would align the whole project with WCAG so every screen is easier to see. Many Radfords users work on tablets in bright packhouses and orchards, where low contrast is a real barrier.
Checking the palette against WCAG AA (4.5:1 for normal text, 3:1 for interface elements) shows where to start:
Colour
Contrast on white
Result
Primary green #26A146
3.35 : 1
Fails behind white button text
Dark green #2A8040
4.93 : 1
Passes, so it could take over as primary
Light green #67BD7D
2.29 : 1
Fails, even for interface elements
Slate grey #818E97
3.36 : 1
Fails for helper text
Light grey #D2D5DC
1.47 : 1
Fails as an input border
I would also make it a rule that meaning never depends on colour alone, so red and green always come with a label or an icon.
What I took away
Real data beats placeholders
The real 13-section form changed my direction and led to the role-first idea.
Isolate the variable
Testing typeface, size and row height separately showed which one actually saved space.
Know where your role ends
Measured comparisons and annotated slides let the team make the final calls on evidence.




