B2B SaaS Foodservice Supply Chain Management:
AI Inventory Forecasting MVP
Inventory is what drives foodservice companies. Too much and you're wasting money, not enough and you're losing money and potentially a customer who won't return.
Forecasting to predict demand and plan for inventory is described as "both a science and an art". Companies have algorithms they currently use, which are mostly based on historical data. Procurement buyers are constantly managing spreadsheets to look at past data, current data, and then get in touch with distribution centers and stores. Different items all have different lead times, which need to be accounted for in stocking.
Our goal is to create an AI forecasting model as well as a UI that a user can interact with to make adjustments to the model and visualize actionable information.
ArrowStream: Foodservice Supply Chain Management
Company and Users Overview
Arrowstream is a B2B Supply Chain Management SaaS product. It is a multi-persona platform with Brands, Distributors, and Suppliers utilizing different products. It is complex and data-heavy, ingesting and mapping data from the distribution center, supplier feeds, and user uploads. It helps users make critical decisions in their supply chain with focused reporting.
Foodservice supply chain management is a complex domain. Due to that complexity, users often have worked in their roles for a long time and get attached to certain workflows and processes. Each brand has its own workflows and required data. This is a field that runs on interpersonal relationships, and a personal phone call or email to a distribution center or supplier is how most communication is handled over automated notifications.
Project Context
-
Procurement is focused on making sure distribution centers and stores have the inventory they need
-
Items each have varying levels of priority for alerting, lead time to receive, usage, and base levels of stock expected
-
Forecasts are affected by lots of factors:
-
Geographic location
-
Seasonality
-
Item Attributes like shelf time
-
Ability to be substituted
-
-
Forecasting is extremely important with little margin for error
-
This project is an additional feature in our core product
The Team, My Role, Timeline, Constraints
The Team
2 teams I worked across
-
AI Team: AI Engineer, PM, and me
-
Central: PM, 5 engineers, and me
My Role
-
End-to-end design
-
Qualitative user research
-
Collaborative scoping of product requirements
-
Experience design
-
Prototyping and Usability Testing
-
Dev Handoff
Timeline
-
This is an MVP release and an ongoing project
-
1 month on Discovery, and User Research
-
1 month on Ideation, Scoping, Design, and User Validation
-
2 months on Development
Constraints
-
This was a fast timeline pushed by leadership initiative
-
Scope shifted from a standalone UI to integrating this model into an existing UI
-
Discovery work was done with the AI team only
-
The central team was missing some context with the handoff between PMs
We wanted to push the product forward by integrating actionable AI insights. Forecasting is a natural space to build a new model.
Forecasting on a more granular level
Previous inventory alerting in the product was based on weekly usage with set models of 7, 14, & 28 days.
AI model calculates based on daily usage providing a more granular and hopefully accurate forecast.
More control over inputs that affect the model
Before forecasts were created and updated quarterly, which didn't allow responsive pivots.
AI surfaces anamolous data to the user for them to exclude.
Future state will allow the user to add APIs that track things like weather or large events in the area for more responsive forecasting.
Actionable Inventory Alerts for teams
The new AI model drives our current Inventory Alerting which provides actionable insights like potentional runout and when to place Purchase Orders to avoid.
UX Research
I organized external qualitative interviews and materials, as well as facilitated sessions with brands.
I also attended internal meetings around model building, testing, and team handoff.
3 External Brand Interviews
Goal: Understand brands' current forecasting methods and goals
-
At what level are brands forecasting and how are they generating that forecast?
-
What is their current forecasting accuracy?
-
What are their biggest forecasting pain points?
-
How would they track effectiveness of a new model?
12 Internal Meetings
Goal: Check in on the progress of the AI model using customer data and deal with handoff between teams
-
What is the difference in accuracy we can get by using the model?
-
What are the limitations we have in user inputs?
-
How are we dealing with handoffs between teams?
-
What are each internal team's responsibilities?
From our research, we broke the user needs into MVP, future state, and not feasible.
1. MVP
-
Beat the average accuracy of 85-90% based on brand
-
Model ingesting historical data and creating a daily usage forecast, this is limited to the top 100 items due to server and data constraints
-
Surfacing anomalous data for the user to act on, reducing the need for manual review
-
Provide actionable insights in our existing UI
2. Future State
-
Allow users more control over which items are forecasted on
-
Integrate APIs around weather or event information
-
Show users more insight into the logic behind the forecast
-
Side-by-side comparison between AI & traditional models
-
Allow users to set more custom alerting thresholds
3. Not Feasible/Potential Workarounds
-
Forecast completely new promotions, items, and stores.
-
Could have the user find a comparable if available
-
-
Forecast on Tail Skews (Low volume items)
-
This one is tricky with so little data, needs more research
-
MVP
We integrated the AI model into our current Inventory Alerts dashboard.
We need to work within the constraints of the current UI. All Confidential Information has been redacted.
-
The AI model is available on the Configuration Settings, and "Powered by AI" is displayed when in use
-
The notification bell surfaces anomalous data to users to exclude from the model
-
Potential Issues: Icons that link to the relevant record in the system are colored based on severity. An actionable suggestion is listed.
MVP Release
Our model accuracy is currently 5-8% higher, depending on the brand. We launched our MVP for all customers, we have 6 early adopter brands and are working on gathering feedback for the next iteration.
Key Learnings
Dealing with scope changes and inter-team handoffs.
Documentation of user research is critical for projects where there are inter-team handoffs with the designer as the only constant.
Being able to pivot as directions change on scoping takes deep knowledge of users' priorities.