The Codex

Could video answer what product photos left out?

A few studio photos and basic attributes could show the packaging without telling shoppers what came inside or how to prepare the food. We already made TikTok Shop videos that showed those details. I wanted to see whether bringing them to our storefront would help shoppers understand unfamiliar products and give them a direct way to shop.

I worked with our UI team to integrate a third-party platform that hosted our videos and provided the player and product placement features. We used it on the homepage and product pages.

Decisions I led

Regional playlists the team could maintain. I used my knowledge of store operations, the storefront, and how the provider’s platform was configured to organize homepage playlists by region. We could rotate the videos while accounting for differences in what stores carried and had in stock. Grouping playlists by region kept the configuration manageable.

Different layouts for the website and app. On the website, I worked with UI to give featured products space beneath the video. On the app, we used circular previews so shoppers could scan the topics without giving up much of the home screen.

What the one-month pilot taught us

I selected a mix of supplier and in-house products to feature in homepage videos. I looked at repeat purchasing by video engagement and compared sales for featured brands with the prior month.

The results gave us a reason to test more placements. Comparing shoppers who chose to engage and comparing sales across months leave other factors in play. A follow-up could assign shoppers to storefronts with and without video during the same period to better estimate its effect.

AI FAQ: solving the catalog bottleneck

I prototyped a way to prepare product information in batches for a product-page FAQ.

The prototype addressed manual data entry. I would evaluate answer accuracy before expanding its use.

Bring in-store payment access online

Customers could already use EBT in stores. Starting on the marketplace let us extend access online and learn from live orders before scoping our own checkout.

Online SNAP was already expanding nationally. USDA reported more than 260 participating retail chains in September 2023 and more than 390 in September 2024, representing thousands of stores. USDA's FY 2025 notes (opens in a new tab) and FY 2026 notes (opens in a new tab) report those dated counts. The USDA retailer directory (opens in a new tab) lists participating retailers by state.

Start with the marketplace

As marketplace manager, I saw that our partner's existing infrastructure gave us a way to offer SNAP EBT online. I worked with the marketplace and a SNAP EBT processor to launch through that channel first. This gave us a way to observe real orders and see how many shoppers were new to us on the marketplace before deciding whether to invest in our own checkout.

Align the approval and catalog handoffs

I coordinated the launch with the marketplace, payment provider, and internal teams.

Before rollout, we checked the eligibility labels shoppers saw online against our product records.

What the marketplace launch showed

The launch gave us a way to observe how shoppers used SNAP EBT online and what to evaluate before adding it to our own checkout.

Turn feedback into a prioritized backlog

Setting up campaigns and promotions could mean repeating the same steps across stores. I interviewed back-office users to identify where work repeated, where the process invited errors, and where the software itself was the constraint. I used those distinctions to decide what Operations could address and what belonged on the engineering roadmap.

How I escalated back-office requests

I started with the lightest fix that could solve the problem. If it fell short, I moved the request down to the next level.

1. Adapt the existing workflow. If the back office could already support the task, I worked with operators to find a simpler way through it using the tools they had.
2. Standardize the process. When teams did the same task differently, we agreed on the most efficient procedure. I wrote SOPs and user manuals in a shared knowledge base so new hires and colleagues in other departments could follow the steps independently.
3. Automate repeatable work. When a task followed consistent rules and could run in batches, I used AI to help build scripts or dashboards that Operations could use without changes to the platform.
4. Prioritize platform changes for IT. When the problem required a change to the platform itself, I ranked the requirement alongside other requests and handed it off to IT for development. This kept roadmap discussions focused on work that needed engineering capacity.

Selected examples of hands-on automation

Promotion setup. I built a batch tool to replace one-at-a-time entry and checked that each discount matched the intended item.

Sales reporting. I helped consolidate recurring reports into a dashboard the team could use to look up results by date and region.

Overall operations result

Locate the delay at pickup

I traced delays to the pickup step and used the share of orders with a tip as a baseline for the checkout change.

Consider what makes an order worth picking up

From a driver’s perspective, expected earnings can affect how attractive an order is. We hypothesized that making tipping easier to find would help customers who already wanted to tip and give drivers more opportunities to earn tips. If that made more orders attractive to pick up, we might see a change in fulfillment as well.

Make the option easier to find

We updated checkout on the app and website with four groups in mind.

CUSTOMERS

Make it easy to leave a tip if they choose. The service goal was more reliable delivery.

DRIVERS

Give drivers more opportunities to earn tips on the orders they pick up.

BACK OFFICE

Use the order and overtime data the team already reviewed to see how checkout behavior and fulfillment changed after release.

BUSINESS

See whether a checkout change could support better service and reduce overtime, which matters to the cost of fulfilling orders.

Track checkout and fulfillment separately

I tracked tipping and fulfillment separately after launch. For a follow-up, I would measure time from order-ready to pickup, on-time delivery, and checkout completion.

Test how Cerenity would fit an existing habit

We wanted to know whether U.S. consumers saw a need for a portable toilet-seat spray and how it would fit into the way they already used public restrooms. Our hypothesis was that people would be more willing to use the spray alongside a seat cover than to replace the cover with it.

We combined restroom journey observations, one-on-one interviews with U.S. consumers, a Qualtrics survey with 59 responses, and secondary research on the category and competitive landscape. The observations and interviews helped us understand why people kept using seat covers. The survey let us compare reactions to the product concept, formats, and pricing.

Position the spray alongside the seat cover

Most respondents (54 of 59) did not currently use a toilet-seat sanitizer. That made it especially important to show how the spray would fit into a routine shoppers already followed.

Translate the findings into a go-to-market plan

Product and pricing. We recommended testing a quick-drying, travel-size spray. About 95% of respondents preferred a spray format, and 78% said they were somewhat or extremely likely to buy a travel-size version. We recommended a follow-up pricing test.

Distribution and acquisition. We recommended grocery stores and e-commerce, the survey’s two preferred purchase channels. We proposed product demonstrations on YouTube and partnerships with creators in TikTok’s cleaning community to explain how the spray fits into an existing routine.

A separate B2B opportunity. We also saw a possible route through businesses that provide restroom amenities. Some interviewees said seeing a sanitizer in a stall could make them question how often the restroom was cleaned, so we treated venue placement as a hypothesis to test separately.

What we would validate next

We recruited largely through personal networks, and many respondents were college students from the same area. The study helped us narrow the positioning and launch options. A broader sample across U.S. markets and tests of actual purchases and repeat use would help establish whether those preferences translate into product-market fit.

Turn the brief into a playable concept

The assignment gave us a theme, “Big,” and ruled out dice-based movement and RPG stat systems. During ideation, we turned BIG into the name Business in Grey and used large financial stakes as part of the game: one money token represented $5 million. Players ran competing fictional hospitals and had to decide how much risk to take while keeping up with debt payments.

Design the mechanics and build the prototype

I helped design how the game’s mechanics worked together. Patient and action cards gave players different ways to earn money, while inspections and increasingly frequent debt collections put pressure on their choices. I created all the game assets, including cards and tokens. We brought the rules and components together in Tabletop Simulator so classmates could play through the full game.

Playtest the pace and the rules

We playtested with 20 classmates and used their feedback to assess pacing, rule clarity, and whether the mechanics produced the decisions we intended. One session ran over time and stopped in phase seven. The draft allowed up to 11 phases, so the team shortened the game to eight in the final rulebook.

Players also found that earning money could be too easy and had trouble interpreting some trap cards. That feedback surfaced two different design problems: how to make the risk-and-reward choices matter, and how to explain the rules well enough for players to act on them.

Pitch and present the game

We pitched and presented the game to the class. The final prototype showed how the “Big” theme shaped the game economy, and the rulebook gave players the setup, turn sequence, and win conditions needed to play.