Best full-service eCommerce agency 2026
When most eCommerce merchants hear "AI", they think of chatbots, product recommendations, and virtual shopping assistants.
But one of the more interesting AI developments from the latest earnings season came from Reformation, and it had nothing to do with customer-facing technology.
According to Reformation's Q3 2026 results, net revenue grew 19.1% year over year to $173.6 million, while active customers increased 23.4% to 1.25 million.
During the quarter, the company also introduced an AI-based planning engine designed to improve how it evaluates product performance and forecasts demand for recut orders.
The distinction matters.
Reformation hasn't said the AI tool caused its 19% revenue growth. In fact, management describes the technology as being in the early stages.
What it has explained clearly is where AI fits into its merchandising model: helping teams identify changing demand, decide what to reorder, and keep more of the right products available at full price.
That is a different AI opportunity from adding another interface between the customer and the storefront. It applies AI directly to the decisions that determine inventory availability, speed and margin.
The demand sensing advantage
On Reformation's Q3 earnings call, CEO Hali Borenstein said the company introduced AI-enabled tools to better predict demand for recut orders and help allocate orders across factories in its global supply chain.
The tools monitor demand signals every day, not just sales signals, using data from Reformation's tech-enabled stores, eCommerce funnel, and other digital customer touchpoints.
Management says the planning engine is intended to improve three things in particular:
-
Product-performance evaluation: identifying which products are gaining traction;
-
Recut forecasting: estimating where additional inventory is needed;
-
In-stock and full-price selling: replenishing products that are working while avoiding unnecessary inventory commitments.
The company is also using AI to help determine where products should be manufactured based on speed, cost, and sustainability requirements.
For eCommerce merchants, this is a useful example of AI being applied to a decision that already has measurable commercial outcomes.
-
Did the right products stay available?
-
Did the business reorder the right SKUs?
-
How much product sold at full price?
-
How much inventory remained when demand moved elsewhere?
These are operational questions that existed before generative AI. The technology changes how quickly and consistently teams can process the information behind them.
Reformation already had the operating model to act on the signals
The AI engine matters partly because Reformation already runs a merchandising model built around rapid testing and replenishment.
The company describes its approach as launching limited quantities, reading customer response, and making more products once demand has been proven. Its own sustainability and production overview says it releases limited-edition collections every week and uses sales, pre-orders, and waitlist data to decide what to produce again.
That operating model showed up clearly in Q3.
Borenstein highlighted the Olina Cropped Pant as an example. After identifying the style as a top performer in May, Reformation recut the SKU in 21 days and subsequently expanded the fabric into seven additional silhouettes.
Management says around 90% of apparel revenue comes from products with proven performance behind them, while approximately half of its product can arrive within 60 days or less and recuts can take as little as 15 to 45 days.
That speed is important.
A more accurate demand signal has limited value if the supply chain can't respond until the season has already moved on.
Reformation can test in relatively small quantities, identify what is working and then commit more inventory after receiving actual customer signals.
AI strengthens that existing process rather than replacing it.
Better in-stock decisions without betting heavily upfront
This is where the model becomes relevant to merchants beyond fashion.
Traditional merchandising often forces teams to commit inventory before they know exactly how customers will respond.
Reformation's approach reduces that initial bet.
Instead of trying to forecast the entire season perfectly, the company can place smaller initial buys, observe demand, and recut proven products.
AI can make the middle of that process faster by helping teams monitor more signals and identify products that warrant another inventory commitment.
For a merchant, the potential benefit isn't simply "better forecasting".
It is improving the balance between two expensive outcomes:
-
Too little inventory: demand exists, but the customer can't buy the product;
-
Too much inventory: demand doesn't materialise, leaving excess stock that eventually needs discounting.
Reformation specifically says its new planning engine is intended to bolster in-stock rates and full-price selling.
That doesn't mean the company has already quantified how much AI has improved either measure. The tool was introduced during Q3 and management described the work as being in the "early innings".
But the KPIs it is designed to influence are commercially clear from the start.
Why operational AI deserves as much attention as customer-facing AI
Customer-facing AI has obvious appeal.
A shopping assistant is visible. Merchants can launch it, customers can interact with it and the business can demonstrate a new AI capability immediately.
Operational AI is less visible.
A customer never sees the system that helps a merchandiser decide whether another 500 units should be ordered.
But the commercial outcome can be easier to define.
For demand planning, merchants can monitor:
-
in-stock rate;
-
full-price sell-through;
-
markdown rate;
-
stock cover;
-
inventory days on hand;
-
forecast accuracy;
-
lost sales from stockouts;
-
excess inventory.
That doesn't make demand planning automatically more valuable than conversational commerce, and Reformation's results don't prove that every merchant should prioritise backend AI over frontend AI.
It does mean operational AI can be evaluated against business metrics that merchants already use.
Customer-facing AI can be harder to isolate. A shopper might use an AI assistant and complete a purchase, but determining whether the assistant created incremental demand or simply became another path to an existing purchase can require more careful testing.
There are also customer-experience risks when AI recommendations rely on incomplete information. Blue Yonder's 2026 Consumer Retail Returns Survey found that 46% of US shoppers who had purchased based on an AI recommendation had returned one of those items in the previous year.
That isn't an argument against customer-facing AI.
It is a reason to consider whether operational problems such as inventory availability, product information, and merchandising deserve equal attention in an AI roadmap.
Reformation's model shows why AI needs an operating system around it
One important qualification is that Reformation's planning approach isn't simply an AI tool layered onto a conventional retail model.
The company already has several structural advantages that allow it to act quickly on the output.
High product velocity
Reformation releases limited collections frequently and tests new ideas continuously.
That creates regular signals about which styles, colours and categories customers are responding to.
William Blair analysts highlighted this advantage in their analysis reported by Retail Dive, noting that Reformation can read and react to demand faster than traditional apparel peers and that AI-enabled merchandising could further improve its ability to localise assortments and chase demand.
Merchants with slower product cycles can still use forecasting technology, but the operating model and decision frequency will be different.
Direct access to customer signals
Reformation remains heavily DTC-led.
Of its $173.6 million in Q3 revenue, $152.8 million came from DTC, up 16.7% year over year. The company also ended the quarter with 77 stores after opening seven during Q3.
Those channels give Reformation direct information across both digital and physical customer journeys.
Borenstein specifically described the stores as tech-enabled and said the company combines that data with information from its eCommerce funnel and other digital touchpoints when evaluating demand.
For merchants considering AI planning, this is an important prerequisite.
Before asking whether the model is sophisticated enough, ask whether the underlying inventory, customer and product signals are reliable enough to support the decision.
A supply chain that can respond
Reformation's speed also matters after the forecast.
Management says recuts can arrive in as little as 15 to 45 days, with around half of products arriving within 60 days or less.
That means a demand signal can still influence what reaches customers during the relevant selling period.
If a merchant requires six months to replenish a product, identifying a demand spike tomorrow doesn't produce the same operational advantage.
The AI recommendation and the organisation's ability to act on it have to be evaluated together.
What eCommerce merchants should take from Reformation
Merchants don't need Reformation's product cadence or 77-store footprint to apply the underlying principles.
Start with a specific inventory decision
Don't begin with "we need AI demand planning".
Begin with the operational question that needs improving.
For example:
-
Which SKUs should we reorder?
-
Which products are likely to stock out?
-
Which variants need inventory moved between locations?
-
Which categories are accumulating excess inventory?
-
Where are forecasts consistently missing actual demand?
That gives the technology a defined job and gives the team a measurable baseline.
Focus on products with enough signal
A high-volume category with frequent orders gives a planning system more current information to work with than a product sold only occasionally.
Start where forecasting errors already have a measurable commercial cost.
That might mean high-volume evergreen products, seasonal categories or SKUs where stockouts and markdowns regularly affect margin.
Clean the operating data first
AI doesn't eliminate the consequences of inconsistent inventory and product data.
Before introducing a forecasting layer, review whether the business can reliably answer basic questions such as:
-
How much inventory is actually available?
-
Which variants are interchangeable?
-
How quickly can each supplier replenish?
-
Which orders are returns, replacements or genuine new demand?
-
Are product IDs consistent across channels and locations?
Better models won't compensate for basic inventory records that teams don't trust.
Measure the outcome, not the AI activity
The important metric isn't how many forecasts the system generates.
Track whether the decisions improve:
-
in-stock rates;
-
full-price sell-through;
-
markdown dependency;
-
forecast accuracy;
-
inventory turns;
-
days of inventory;
-
stockout frequency.
That makes it possible to compare AI-assisted planning with the process it replaced.
Check whether the supply chain can act
This may be the most important lesson from Reformation.
Its planning model is connected to an operating system built for quick reaction.
If the AI identifies demand but the merchant can't reorder, move or produce inventory quickly enough, the insight arrives without an action.
Before investing in more sophisticated forecasting, identify where the supply chain can actually respond differently when the forecast changes.
The peak-season implication
Reformation's model is particularly relevant during periods when demand changes quickly.
On the Q3 call, Borenstein described the company's agile merchandising model as a particular advantage during the holidays because it can identify trends and respond quickly during an important trading period.
For merchants entering peak, introducing an entirely new planning platform at the last minute may create more risk than value.
But the principle can still be applied now.
Actions to take:
-
Identify high-risk stockouts: focus monitoring on products carrying significant peak demand;
-
Review replenishment speed: know which SKUs can realistically be reordered during the remaining selling window;
-
Monitor demand signals more frequently: shorten the gap between identifying a change and making an inventory decision;
-
Protect proven winners: make sure fast-selling products aren't losing revenue because teams are relying on a weekly or monthly planning cycle;
-
Track markdown exposure: identify inventory accumulating faster than current demand supports.
After peak, those same workflows provide a useful starting point for deciding where AI-assisted planning could improve the process.
The bigger picture
Reformation's Q3 results don't prove that demand-planning AI produced 19.1% revenue growth.
The company reported growth across channels, categories and geographies, while active customers increased 23.4%, international revenue increased 24.7% and wholesale and other revenue increased 40.8%.
The AI planning engine is one part of a much broader growth model.
What makes it interesting is the job Reformation has given it.
It isn't there primarily to create another customer interaction.
It is monitoring demand, improving recut forecasts, supporting factory allocation and helping the business respond more quickly to what customers are already telling it.
That makes AI part of the merchandising and supply-chain decision process rather than a layer placed on top of the storefront.
Conclusion
The most useful lesson from Reformation isn't that every eCommerce merchant needs an AI forecasting platform.
It is that AI investment should start with a commercially important decision.
For Reformation, that decision is whether demand justifies another inventory commitment, where that product should be made and how quickly it needs to return to stock.
The company already has weekly product testing, direct customer signals and a supply chain designed to chase proven demand. AI gives that system another way to process those signals at scale.
Customer-facing AI will continue to grow.
But before adding another chatbot or shopping assistant, merchants should also look behind the storefront.
If AI can help put the right product in stock, reduce unnecessary inventory and support more full-price selling, the customer may never see the technology.
They will see the result.
About On Tap
On Tap is a growth-focused eCommerce consultancy helping mid-market and enterprise merchants improve their digital commerce operations, from merchandising and product data to platform strategy, integrations and ongoing performance optimisation.
If you're evaluating where AI can create measurable value across your eCommerce operation, get in touch.
