Building a Smarter Product Merchandising Engine
The Challenge
How do you display products on an ecommerce website in an order that serves the business, while helping customers find the products that are right for them?
Effective merchandising is more than simply putting the best-selling products first. When businesses have control over how products are merchandised, they can balance customer needs with business priorities and surface the products that matter most.
Think about it as analogous to a large chain grocery store - the store needs the necessities to be easy to reach, but there's a trade-off and that’s the reason that the milk, eggs and other similar high-volume goods tend to be at the back of the store. Just like these grocery chains, who have spent hundreds of millions of dollars on the psychology of merchandising, you want your store to have the chance to tempt the consumer into additional purchases as well as to give them what they want and need.
This becomes more challenging when ecommerce represents only one part of the overall business.
In the case of the client that inspired this post, and a large subsection of all B2B brands, their ecommerce store did not represent the full picture of their sales. Though their platform's default 'Bestsellers' sorting was able to accurately reflect online orders, it did not account for products that perform particularly well through offline sales.
In fact, some of the client’s best-selling products were primarily ordered offline and were significant contributors to the business’ turnover. The question was:
How can the business bring that complete sales picture into ecommerce merchandising?
Our client’s existing solution did not provide a clean, extensible, or future-proof way to do this – and to be honest, no CMS is able to directly out of the box.
Sales performance was also only one part of the equation here, and several other business factors needed to be influencing how product were ranked:
Gross profit
Product lifecycle and end-of-life status
Quick-ship availability
Featured status
Margin tier
Online and offline sales performance
Other business-specific merchandising priorities
Much of this information lived in the client's Enterprise Resource Planning (ERP) system rather than the ecommerce platform.
The challenge was bringing these different data points together into a merchandising solution that the business could actually control.
The Existing Approach
The ecommerce platform provided a Featured sorting option that was controlled through a sort-order value assigned to each product.
However, the problem was not the sort-order itself, but maintaining it as the underlying business variables changed.
Previously, our client had been forced to rely on a complex, spreadsheet-based process to calculate and maintain these values. While the process worked, it had several limitations:
It was not easily understood by everyone involved.
It was difficult to change.
It required significant manual maintenance.
It was difficult to extend as new business requirements emerged.
There were also no suitable third-party applications available that addressed the client's specific requirements.
We quickly realized they needed a different approach.
Aysnd’s Approach
We started with the business problem rather than the platform's limitations.
By working through the system from the source data to the final ecommerce experience, we found an opportunity to bring the client's merchandising logic into a dedicated solution built specifically for their business.
The goal was not simply to build another merchandising tool (though this is now a customizable solution if you find yourself in the same position). We wanted to create a central place where the business could combine its data, define its merchandising strategy, apply human judgment, and safely publish the results to the ecommerce store.
We also explored where AI could help accelerate the implementation while keeping business decisions and important controls in the hands of the people responsible for merchandising.
The result was a solution built around four key principles.
1. A Single, Joined View of the Catalog
BigCommerce holds the product catalog, while Acumatica holds sales and cost information.
We match the two systems using the Stock Keeping Unit (SKU) and combine online and offline sales data to calculate metrics such as gross profit and units sold for each product.
This created a much more complete merchandising signal.
Instead of asking, "What sells best online?", the business can now ask, "What products are most important to the business overall?"
That distinction matters to all businesses, but especially when ecommerce is only one part of the sales operation.
2. Rules the Merchandiser Can Change
The weighted merchandising factors are treated as data rather than code.
Each rule has a configurable weight and can be reordered or turned off directly through the interface.
For example, if the business wants to place greater emphasis on margin during a particular quarter, the merchandising team can make that adjustment directly without waiting for a developer to change the application.
This transforms merchandising from a developer-dependent process into something the business can actively manage.
3. Manual Overrides on Top of the Model
Not every merchandising decision can or, necessarily, should be automated – and this can be the problem with several AI solutions currently on offer, which prioritize only velocity or other partial signals.
There are times when a merchandiser knows that a particular product needs additional visibility, or that an entire category should receive a boost.
Aysnd’s custom solution allows merchandisers to boost or pin individual products and boost entire categories.
These overrides are applied on top of the calculated score, allowing automation and human judgment to work together, with AI given the power to augment, but not to override.
The goal is not to replace the merchandiser; it’s to give them better tools.
4. A Publish Process Built for Trust
Any automation that writes directly to a live ecommerce store needs to be predictable and safe.
Each publish follows a controlled process:
Preview the exact changes before publishing.
Write only the values that have changed.
Record the publish run.
Maintain a history of the changes.
Allow changes to be reverted when needed.
A nightly job can refresh the underlying data and publish changes automatically.
The ‘history’ function also records the previous and new values for every product that was changed.
This makes automated merchandising transparent and auditable rather than another black box.
5. Data Quality Is Part of Merchandising
A merchandising engine is only as reliable as the data behind it.
Sales records that cannot be matched to a catalog SKU are surfaced for review instead of being silently ignored.
This provides visibility into data quality issues and ensures that the business understands when its merchandising decisions are being made with incomplete information.
What We Learned from Building With AI
AI played an important role in accelerating the implementation, but the project also reinforced where human judgment remains essential.
AI Executes Fast
AI can move very quickly when the desired outcome is clearly defined.
We used tests before implementation, faithfully translated the existing spreadsheet logic into the new solution, and identified issues such as a VLOOKUP bug during the process.
The speed was valuable, particularly when working through an existing set of business rules that had already been refined over time.
AI Defaults to Familiar Patterns
AI has a tendency to fall back on common design patterns.
Early in the process, we found ourselves with what looked like "the same website every AI builds."
That was a useful reminder that good implementation is not just about generating working code, it also requires context, design direction, and a clear understanding of the client's existing experience, expectations and business needs.
We pushed the implementation to properly use the provided pattern library and align with the client's design system.
Architecture Still Requires Human Judgment
One of the key architectural decisions was to keep the data flow simple:
Fetch once → Cache → Work from the cache
Rather than repeatedly calling external APIs, the application retrieves the required data, caches it, and performs its calculations against that cached data.
AI helped us move quickly through the implementation, but the decision about where to draw these boundaries remained a human one.
Avoiding Over-Engineering
Another important lesson was to use the data that already exists.
Acumatica already provides many of the fields required by the merchandising logic. Rather than re-deriving those values and recreating formulas in our application, we use the data provided by the ERP wherever possible.
This keeps the solution simpler, easier to maintain, and less likely to introduce differences between systems.
Sometimes the best architecture is simply the one that does not create work that another system has already done.
The Bigger Picture
Product merchandising is often treated as a simple matter of sorting products.
In reality, it can represent a combination of sales performance, profitability, inventory strategy, product lifecycle, operational priorities, and human judgment.
For businesses that sell through multiple channels, the challenge becomes even greater because the ecommerce platform may only see part of the story.
By bringing those signals together and giving the merchandising team control over how they are used, businesses can move from a static sorting strategy to a merchandising system that reflects how the business actually operates.
For us, the most important part of this project was not simply automating product rankings.
It was building a foundation that allows the business to continuously evolve its merchandising strategy without having to rebuild the solution every time the business changes.