Shift4
In progress
A redesign of handheld ordering at Shift4: fewer tabs, clearer required choices, and an assistant that builds orders from a sentence.
The problem
Dine Handheld looked more modern than Duo, but it was harder to use. Departments, seats and item actions were hidden behind horizontal scrolling, and items were added with no confirmation. One participant called it “clunky, too many steps, too much room for error.”
A newer interface hadn’t made ordering better. Servers work one order at a time at a table, so every extra step and every hidden control costs them speed and accuracy.
Core controls were hard to discover
Items and seats were easy to get wrong
Visual polish wasn’t translating into preference
The short version
I’m redesigning the entire Dine experience at Shift4, starting with ordering on the handheld. The new flow keeps servers on the menu, puts quantity and guest on one item sheet, holds Add until required choices are made, and confirms every addition. An assistant can build an order from a sentence. It’s grounded in a baseline test against the legacy product and a competitive review, and it’s now a working prototype built in Claude Design, with a retest to come.
Where it started
The approach
We measured first, then researched, then prototyped, so the redesign can be judged against evidence rather than opinion.
Baseline test: Participants used A/B prototypes of Duo and Dine Handheld, so we could see what the current product gained and lost over the legacy one.
Research: We split ordering into four areas: order entry, menu navigation, item building, and order assembly and send. For each we reviewed Toast, Square, Lightspeed, TouchBistro and SpotOn, plus consumer ordering in Uber Eats, Taco Bell and Keeta. A merchant survey on SkyTab, the handheld’s former name, also kept pointing to faster menu navigation.
Prototype and retest: The findings went straight into a working prototype. The plan is to keep cycling through prototype, test, prioritize, revise and retest.
Team: I lead the work with a product designer and a senior UX researcher. It started with one PM on ordering, then added a second PM who owns ticket management, and each PM’s team brings its own developers and leads.
Understanding the problem
Seven participants worked through ordering tasks on both prototypes: A was Duo and B was Dine Handheld. The goal was to establish baseline measures before any redesign, so we could later answer the question that matters most: did it actually make the product better for users?
The overall result was clear. All 6 participants who stated a preference chose Duo, citing consistency and a gentler learning curve. The seventh rejected both. Dine Handheld only won on visual appeal and engagement.
Alongside the test, we split ordering into four areas and studied how the market handles each, using product documentation, demos and reviews. That covered Toast, Square, Lightspeed, TouchBistro and SpotOn, plus Shopify POS and consumer ordering in Uber Eats, Taco Bell and Keeta.
| Feature | Square 7/9 | SpotOn 6/9 | Toast 5/9 | Lightspeed 4/9 |
|---|---|---|---|---|
| Open ground | ||||
| Dedicated substitution flow | ||||
| Where products pulled ahead | ||||
| Staff-facing search | ||||
| Guided required modifiers | ||||
| Nested modifiers | ||||
| Multi-select on the ticket | ||||
| Combos as their own item type | ||||
| Allergens at order entry | ||||
| Swipe actions on ticket items | ||||
| Table stakes | ||||
| Table-first, with no order-type step | ||||
Department scrolling isn’t discoverable. Four participants got stuck moving between departments, and two fell back to search.
Search felt fast, but it was masking a navigation failure. Three participants used it to route around browsing.
Favorites only help if they ship preset from top sellers and specials, not as an empty tab to set up.
All 7 expected a quantity selector. Two reached for a stepper over Duplicate once both existed.
5 of 7 expected one sides section with checkboxes, not separate Side 1 and Side 2 radio groups.
Selecting an item added it instantly with no confirmation, so two participants couldn’t find it on the ticket.
Seats were the most consistent failure. Everyone who tried them on Dine Handheld struggled, and all handled them quickly on Duo.
Remove and Duplicate sat behind a horizontal scroll with no sign it could scroll. Two participants never found Remove.
Color was Dine Handheld’s clearest win, with a caveat: two participants also called it busy.
Mapping it out
The review showed where the industry had settled and where it hadn’t. Four areas shaped what we did next:
Order entry: Every POS we reviewed opens from the table, with no order-type step. Square restaurants created fake To Go tables as a workaround, leading to a billing dispute and an estimated 5% daily mistake rate at rush.
Menu navigation: Well-reviewed products pair persistent search with browsing. Square’s manual grid slows servers when it doesn’t match the printed menu, and Lightspeed has no search in Quick Service mode.
Item building: Square walks servers through required sets automatically, and Keeta’s required state is the clearest at a glance. No one handles substitutions cleanly.
Order assembly: Toast carries color-coded send status through to the floor view. SpotOn has the fullest gesture set, and Uber Eats and Taco Bell fold removal into the quantity stepper.
We kept some questions open for testing, because no competitor answered them: how often servers use subgroups, whether the menu should stay open after adding an item, and how to move items between seats.
I mapped how ordering works today, where more than a dozen item options sit in tabs, then a proposed flow built around these patterns: keep the server on the menu, confirm every addition, and move quick edits onto the item line.
Seeing the system
Required modifiers were a recurring source of error. Items could be added before a server had finished choosing, and participants lost track of what was on the ticket.
Each required set is now labeled, and Add stays disabled until they’re complete. The status updates as the server chooses, so what’s missing is visible at a glance.
Trying it out
I built a working prototype in Claude Design, using a Claude design system connected to Bone Zebra, our Material-based system for native Kotlin. The team can react to real interactions rather than static screens, and each baseline finding maps to a specific change:
Add is now an explicit button, with a toast naming the item and guest
Quantity sits beside Add, inside the item sheet
Guest assignment stays in view on the item sheet
Modifiers stay on the item sheet, with required sets labeled
The ticket is organized by guest, with Send and Pay pinned to the bottom
The prototype also includes some workstation versions of the redesign, alongside the handheld.
Bringing it to life
The prototype also includes an assistant that works on the screen the server is on. Say “two cheddar bacon burgers medium for guest 1,” and it builds the order, asks for any required choice it’s missing, and waits for confirmation before adding items.
It holds to the same rule as the rest of the flow: nothing goes on the ticket until required choices are made. Servers can type or speak, which suits a job where their attention belongs on the guest. We used xAI to power the agent for testing.
What changed
What the redesign changes:
Item option tabs go from 13 to 2, with other actions moved to the ticket
A steak with required choices takes 4 taps, ending in a confirmation
Servers can start an order in one line to the assistant, typed or spoken
Tickets are organized by guest, with Send and Pay always in reach
The same patterns carry across handheld and workstation
These come from the prototype. Task success and speed will come from the retest against the baseline.
Looking back
This project is still in progress, so the most useful result so far is the baseline itself. Without it, we could easily have assumed the newer product was the better starting point.
Next, we’ll put the prototype in front of servers and measure it against the same baseline: task success, effort, ease and confidence.