An Interview with Pini Usha, CEO and Founder of Buffers.ai – The AI solution for Demand planning
Fashion companies have invested heavily in digital transformation over the past decade, yet many merchandising, planning, and inventory allocation decisions are still managed through spreadsheets and manual processes.
In this interview, Digital Fashion Academy speaks with Pini Usha, CEO and Founder of Buffers.ai, about how artificial intelligence is helping retailers and brands improve forecasting accuracy, optimize inventory, and empower planning teams to make better decisions.
Q. Pini, let’s start with the problem. What are the biggest challenges retailers and brands face today when it comes to inventory planning?
A. The biggest challenge is forecasting demand accurately.
Whether you’re a retailer or a manufacturer, you need to predict future sales and make purchasing decisions accordingly. If the forecast is too high, you end up with excess inventory. If it’s too low, you face stockouts and miss sales opportunities.
Many companies still compare planned sales versus actual sales and discover significant gaps. Some products sell out much faster than expected, while others remain unsold in warehouses. This creates both financial and operational challenges.
Fashion companies know these problems very well. Excess inventory often leads to markdowns, while stockouts mean lost revenue and disappointed customers.
How does Buffers.ai help solve this challenge?
Our platform helps companies make better forecasting decisions using a combination of historical sales data, business events, and human expertise.
We have two main modules:
- Purchasing Planning, which helps companies decide what quantities to buy or manufacture.
- Replenishment Planning, which helps distribute inventory efficiently from warehouses to stores.
These are connected processes. First, you determine how much inventory you need. Then you determine where that inventory should go.
Q. Let’s talk about forecasting. How does the system generate predictions?
A. We use three main sources of information.
The first is historical sales data. We analyze previous consumption patterns to identify trends and seasonality.
The second is business events. The system automatically identifies recurring events such as holidays and understands their impact on sales. For example, if sales increase every year during a specific holiday period, the system learns that pattern and incorporates it into future forecasts.
The third source is human input. Sales teams and planners often have information that doesn’t exist in historical data. They may know about a large upcoming project, a major customer order, or a planned promotion.
Our platform combines these inputs to create a final forecast.
Q. I found the event management functionality particularly interesting. Fashion companies frequently plan around Black Friday, seasonal promotions, marketing campaigns, or store openings. Can these events be included in the forecast?
A. Absolutely.
Users can create future events and specify their expected impact. For example, if a retailer plans a stronger Black Friday promotion than last year, they can adjust the expected uplift accordingly.
Similarly, companies can identify unusual events from the past.
Imagine sales dropped significantly because of a temporary store closure or a geopolitical event. If the system doesn’t understand the reason behind that drop, it may incorrectly assume the same pattern will repeat next year.
By adding context, planners help the AI make more accurate decisions.
Q. One thing that stood out during the demonstration was that the platform combines multiple forecasting models rather than relying on a single algorithm.
A. That’s correct. Different products behave differently.
For one SKU, a forecasting model developed by Google may provide the most accurate prediction. For another product, a model from Meta or Amazon may perform better.
We continuously test multiple forecasting models using back-testing techniques. The system evaluates how accurately each model would have predicted actual sales and automatically selects the best-performing approach.
This allows us to optimize forecasting accuracy product by product.
Q. Forecasting is only one part of the process. Once inventory has been purchased or manufactured, retailers still need to allocate products across stores.
Why is replenishment such a complex challenge?
A. Because it happens every day.
Forecasting may be performed once per season or once per month, but replenishment decisions are continuous.
Imagine a retailer with 150 stores and thousands of SKUs.
Every day, planners need to decide how much inventory should be sent to each location. Doing this manually is extremely time-consuming and often impossible to optimize using spreadsheets.
Our replenishment module automatically calculates inventory distribution at the store level, ensuring products are available where demand exists while minimizing excess stock.
Q. What about new products that have no sales history?
This is a common challenge in fashion, where collections change every season.
A. We call this the “first allocation” challenge.
For new products, we offer several approaches.
Some retailers prefer rule-based allocation, where percentages are assigned to stores based on category performance.
Others use fixed allocation ranges.
Our most advanced approach uses similarity analysis. The system identifies comparable products from the past and learns how they performed in different stores. It then uses those insights to recommend initial inventory allocation for the new product.
Q. Many people assume AI is primarily about reducing headcount. What are you seeing among your clients?
A. Most clients are not buying our solution to eliminate jobs.
They’re buying it because they want better decisions.
Today, many companies have teams of planners spending countless hours collecting data, updating spreadsheets, and trying to create forecasts manually.
AI automates much of this work, but people are still essential.
Instead of spending time manipulating data, planners can focus on higher-value activities such as strategy, assortment planning, and business analysis.
Q. That’s a perspective I strongly agree with. The goal isn’t necessarily fewer people. The goal is allowing talented professionals to focus on more valuable work.
Who are typically involved in the decision to adopt a solution like Buffers.ai?
A. The decision-makers are usually executives such as Chief Operating Officers, Chief Technology Officers, Chief Supply Chain Officers, or CEOs.
However, the people who use the system every day are usually merchandisers, planners, and supply chain teams.
It’s interesting because executives are often excited about innovation and AI, while operational teams want practical solutions that are easy to understand and trust.
Successful adoption requires addressing both perspectives.
Q. Finally, how do companies measure success after implementing a solution like Buffers.ai?
A. For forecasting, the primary KPI is prediction accuracy.
We compare forecasts against actual results and measure forecasting error over time.
For replenishment, companies typically focus on three key metrics:
- Stockout reduction
- Excess inventory reduction
- User productivity and decision-making efficiency
When these metrics improve simultaneously, the financial impact can be substantial.
The Future of Fashion Planning
As fashion supply chains become more complex and consumer demand becomes increasingly unpredictable, the ability to combine AI-driven forecasting with human expertise is emerging as a critical competitive advantage.
Rather than replacing planners and merchandisers, platforms like Buffers.ai are helping them make faster, smarter, and more accurate decisions, reducing inventory risk while improving product availability.
For fashion retailers, brands, and manufacturers, the future of planning may not be about choosing between human intuition and artificial intelligence, but learning how to combine both effectively.