Demand Forecast
How my research helped shape an AI-enabled product's design and development
Retail
Decision support system
User Research
Summary
Challenge
My role
Result
What I've learnt
Context
Inventory is one of the crucial investments in any retail business. Storing more stock than the customer demand - it ties up cash. Storing less stock than the customer demand - you lose the money you are supposed to have. The right amount of inventory stored is the amount that meets customer demand.
This forecast task has been assigned to the Procurement team. Every month they have to predict the demand, and simultaneously, plan the amount of stocks for procurement & replenishment.
Tiki (tiki.vn) is an e-commerce platform with a retail business (Tiki Trading) operated by Tiki itself. Tiki Trading generates the majority of Tiki’s profit.
Tiki Trading warehouse
Challenge
The manual approach works in the early days where we have not so many products and cost optimization is not a big concern. However, when our trading business starts growing, the Procurement team then needs to cover a larger number of products (100,000+ SKUs) and stricter requirements for accuracy / optimal level (bad stock, cash flow, etc.)
With all those constraints, the manual approach did not work well anymore (for both coverage and accuracy) - and that’s where software solution adds value.
This is the workspace the Procurement team uses to perform the task.
Research
First, we need to analyze the current methods: how the team do the forecast? the logic behind the prediction? any space to improve? We started with a User Research – which I took lead.
I performed Contextual Inquiry, which requires me to observe the users while doing the task from start to finish and interview (semi-structured) them during the process.
There were about 12 people in total. Each one in charge of different product type (Mom and baby products, Health & beauty products, Electronic devices, …)
Questions to be answered
How is the demand forecasting task currently performed? What the process is like? What calculation they make?
Why they do what they do?
Identify the problems or challenges which could lead to low coverage and accuracy.
We work remotely so most of the interviews was conducted online with the users staying at their ‘natural environment’ - their home, on their computer.
Findings
After collecting the data, my thematic analysis reveals the methods & process being used to forecast customer demand, as well as the potential risks it could cause to the task’s accuracy & coverage. The findings, as well as my recommendations for next steps, are put together in the below report.
Based on my report, together with our team (PM/DA/Dev), we discuss further to determine the key takeaways, presented in the following slides.
Key takeaways drawn from the research
1. Some parts can be automated
Although different product types (e.g., food versus electronic devices) may follow different processes due to their unique characteristics, there are key similarities that could enable a unified decision-making process. For example, at some steps, they all:
Rely on similar data references (e.g., historical sales data)
Share common goals (e.g., identifying historical sales trends),
And use comparable methods to achieve these goals (e.g., calculate average sales per day).
As I was able to synthesize all of those patterns, these serve as a knowledge base for the data/dev team to build their forecasting model/algorithm/set of rules for our AI product.
2. However, there are also parts that could be a challenge to automate
Certain ad-hoc or unstable factors require users to rely on intuition and market experience rather than calculations. For example:
Seasons (e.g., Christmas holidays, back to school season)
Products that are ‘on trend’ or ‘out of trend’ (e.g., via Tiktok/social media) (could have unusual future demand)
New product → no historical data to refer to.
Unusual cases like supply crises (e.g., material shortages or rising gas prices)
These are areas where the system struggles due to limited data and predefined rules, making user input, audit, and review essential.
Back to school season
Example of 'on-trend' and new product
3. Many risks our product could prevent
During the forecasting process, users must navigate multiple Excel files and tools spread across different sources, leading to wasted time switching between platforms and performing repetitive tasks. A significant portion of their time is spent collecting and centralizing data into a single Excel file, followed by manual calculations to derive necessary metrics.
Additionally, human errors and biases frequently arise, such as mistakes from copying and pasting or irrational decision (e.g., relying on overly short time frames for analysis). These issues compromise the accuracy of the forecasts.
Automating these steps would streamline the process, reduce errors, and enhance the reliability of forecasting outcomes.
what i've learnt
Understanding the users while talking to them and also after that, in this research, became more challenging than anticipated, because:
There was no standard process or method that applies across all categories. The prediction method varies based on different factors. Being able to generalize them into one general process was difficult.
To predict the demand, they also have to use personal intuition or market experience in their specific domain as mentioned. They may have some sort of implicit rules or method for this, but they could not explicitly state them.
Our users often struggled to articulate the true underlying reasons and goals. Sometimes, their intentions were deeper and more complex than what they conveyed.
And lastly, all of these things are insanely complicated to a "muggle" like me 😀
Therefore, in order to really understand what's happening, to ensure that I understand the fundamental essence of what they said or did, it required me to:
Quickly get what they mean/calculate/think using my data skill & domain knowledge. But at the same time not shy away from asking more.
Breaking down the problem into smaller chunks, and some other times zooming out from the details to generalize the bigger picture.
Always be… skeptical! Although they are expert in their field, I believe they could not always be rational and sometimes make mistakes. Being skeptical had helped me to identify many problems with what they are doing to forecast the demand.
After all, designers - especially when designing/researching these type of products, I believe, must know more than just design.
Solution
For the first phase of the project, we will design a decision support system for more efficient demand forecasting, that:
Generates predictions using simple time series forecasting methods*.
Users can then review these predictions, using the reference data displayed on the UI, and further refine them by applying intuitive factors to finalize the numbers.
*Futher Machine Learning methods are applied in later stages: https://engineering.tiki.vn/into-the-demand-forecast-of-tiki-operations/
Result
We achieve very good result for this first phase although the system's forecast method was rather simple. Our system has help the Procurement team to improve the accuracy significantly from a huge rate of overestimation to only around 10% overestimation.
As for the product adoption / transition from previous tools (excel) to our new tool, despite the unfamiliar interface and the complex nature of the task, the adoption curve was very low – very few users (out of total 30) required our assistance on how to use the new products.
My contribution
My research help analyze the current detailed methods & identify their problems.
The findings served as a knowledge base for the development team to develop a set of rules for our product’s decision-making process.
The research also informed the design of the user flow and user interface, enabling users to audit and improve system suggestions with ease. By understanding which data should be displayed, how data points are interconnected, and how users perform calculations, I ensured the designed flow and interface supports their needs effectively.
