When a tech firm’s inside processes are inefficient, the signs aren’t all the time instantly obvious. On the floor, issues appear to operate: orders get processed, routes are calculated, help tickets are resolved. However behind the scenes, inefficiencies can quietly construct up — extreme handbook work, redundant steps, and inconsistent efficiency. As the corporate scales, these hid bottlenecks begin to hinder progress.
Fixing the issue requires greater than piecemeal fixes. It requires rethinking the structure, redesigning workflows, and guaranteeing that techniques work together seamlessly. This calls for rigorous evaluation, shut consideration to operational element, and the power to implement modifications with out disrupting steady processes.
Manu Ajith is the co-founder and CTO of ReturnQueen, a US-based startup that handles product returns for on-line purchases from platforms like Amazon and BestBuy. On this interview, he explains how the group identifies factors of friction of their operations, automates high-effort processes, and implements change in a method that helps progress — with out breaking what’s already steady.
When a System Works — however Holds You Again
ReturnQueen, a US-based startup, helps clients return on-line purchases from marketplaces like Amazon and BestBuy. The method is simple: customers schedule a pickup via the app, and a courier collects the merchandise from their doorstep. The corporate launched rapidly throughout the pandemic, when e-commerce volumes had been surging globally.
ReturnQueen was based in 2020 by Manu Ajith, along with companions Daphna Englard and Dasya Katz. At this time, the startup operates in over 10,000 zip codes throughout greater than 20 U.S. cities — with plans to increase into further key markets. By optimizing the post-purchase expertise, ReturnQueen boosts buyer satisfaction and drives long-term retention — reshaping what customers count on from on-line buying.
At first look, the system appeared to work properly: returns had been processed, couriers arrived on time, and buyer help dealt with inquiries effectively. Nevertheless, behind this obvious stability had been inefficiencies that made scaling troublesome. Many duties — particularly these involving return labels — had been dealt with manually. Customers needed to find and add labels themselves, whereas help workers regularly assisted with discovering the right portal or troubleshooting add points.
“Our early evaluation revealed that about 20% of our help group’s time went towards label-related issues — from guiding customers via retailer portals to resolving add points. These had been hidden prices that didn’t appear important initially,” Manu remembers.
Logistics posed one other problem. The dispatch system didn’t account for geographic proximity or present routes, resulting in redundant journeys. Drivers had been despatched to the identical neighborhoods a number of occasions a day, considerably rising mileage per pickup. The underlying points solely turned clear after the group related Metabase to their operational information and analyzed person habits via PostHog.
How ReturnQueen Decides When to Make a Change
At ReturnQueen, modifications aren’t made on a hunch — they’re pushed by three key inputs: quantitative information, person habits insights, and suggestions from the operations group. Solely when all three point out the identical downside does the group transfer ahead with evaluation and potential updates.
One instance is how they addressed return delivery labels. Initially, customers needed to manually add labels by navigating to the retailer’s web site, downloading the file, after which importing it to the app. Later, ReturnQueen launched a function that robotically extracted the right label from the order affirmation e-mail. Information from Metabase confirmed that this automation decreased failures — like corrupted or mismatched information — by about 15%. Buyer help additionally famous a lower in associated tickets, whereas PostHog information confirmed that customers counting on handbook uploads had been extra more likely to repeat the step or abandon it altogether.
“We don’t simply depend on metrics. Generally the numbers change, however the group doesn’t discover an actual distinction. Different occasions, we catch customers struggling at a particular step via session information, even when it doesn’t present up on dashboards. That’s why we take a holistic strategy — combining information, on-the-ground workload, and group suggestions to find out when and the place to step in,” Manu explains.
Why ReturnQueen Needed to Automate Information Assortment
At ReturnQueen, automation wasn’t launched in a single day — it advanced as the corporate confronted rising operational calls for. Initially, whereas the service operated in only one metropolis, many duties had been dealt with manually: verifying orders, monitoring return deadlines, and coordinating pickups. This handbook strategy labored to start with, however because the group deliberate to increase right into a second market, it turned clear that doubling the amount would require excess of simply doubling the sources. To scale effectively with out considerably rising headcount, the group needed to automate probably the most labor-intensive elements of the method.
“We estimated the manpower wanted to keep up our present pace, and the mathematics didn’t add up — scaling manually simply wasn’t sustainable,” Manu remembers.
Step one was automating information assortment. The group developed an ETL course of that, with person consent, built-in with Gmail. This technique extracted order affirmation emails, recognized which objects had been returnable, tracked return deadlines, and robotically triggered the right label or routing move — eliminating the necessity for handbook information entry.
This automation wasn’t nearly person comfort — it was a strategic shift. With out it, dealing with hundreds of returns per day would have required a considerably bigger group.
“This integration wasn’t merely about enhancing person expertise. It was important for constructing a streamlined, automated pipeline that would scale effectively,” Manu explains.
How the Crew Assessments Modifications With out Disrupting the System
Whereas automation enabled ReturnQueen to scale, it additionally heightened the dangers. Any mistake within the logistics engine may end in missed pickups and erode buyer belief, notably when dealing with excessive volumes. The problem was clear: the way to implement modifications with out jeopardizing the present system.
To mitigate dangers, main updates are first examined in a managed atmosphere. As an illustration, when the group launched a brand new routing logic primarily based on a VRP (Automobile Routing Drawback) algorithm, they rolled it out completely in Austin. Over the following two weeks, they monitored key metrics corresponding to on-time arrival price, route length, and common mileage per pickup. In addition they gathered driver suggestions via inside channels to rapidly determine any potential points.
“If one thing went flawed, we may instantly revert to the earlier model. Function flags make it potential to modify functionalities on or off inside minutes,” Manu explains.
Function flags permit the group to activate particular options with out redeploying code. This strategy helps managed experiments, A/B testing, and fast rollbacks if outcomes deviate from expectations. By following this phased rollout methodology — combining gradual implementation, steady monitoring, and the power to reverse modifications — ReturnQueen maintains system stability whereas introducing enhancements.
As handbook processes had been phased out, the following focus was guaranteeing that the system remained resilient at scale — not simply from a purposeful perspective, but additionally by way of sturdy structure.
When It’s Simpler to Rebuild the Structure Than Patch It
As ReturnQueen’s order quantity elevated, the group confronted a essential problem: the system couldn’t sustain. The platform was initially constructed as a monolith, which means all core processes ran inside a single block. This setup made it troublesome to scale, as any change or failure in a single space — like routing logic — may disrupt different features, together with order monitoring and information ingestion. Even minor updates required deep, system-wide testing.
To beat this, the group transitioned to a microservices structure, breaking the platform into distinct, modular parts: order ingestion, routing, and standing monitoring. Every module was designed with a transparent accountability and could possibly be scaled independently. This shift considerably decreased strain on particular person elements of the system and elevated flexibility. Now, updates could possibly be made to 1 module with out risking the steadiness of the whole platform — an important issue for sustaining quick iteration and protected progress.
How the Crew Measures Whether or not a Resolution Really Works
At ReturnQueen, each change is pushed by clear, measurable outcomes. Earlier than implementing a brand new function or replace, the group defines success standards: Are operations changing into quicker? Has the help load decreased? Are customers navigating the method extra easily?
For instance, when updating routing logic, the group tracks metrics like common miles per pickup, the success price of first-attempt pickups, and the time between a return request and the courier’s arrival. If the person interface is up to date, they monitor what number of customers full the step on the primary strive, the bounce price, and the way usually clients attain out to help.
“We begin with a easy query: what precisely ought to enhance if the answer works as deliberate? That method, we keep targeted and may rapidly spot when a speculation doesn’t maintain up,” Manu explains.
How the Crew Responds to Unintended Aspect Results
Regardless of thorough planning, sudden outcomes are generally unavoidable — particularly when machine studying is concerned. At ReturnQueen, this problem emerged with the return label recognition mannequin. Since these fashions are educated on massive picture datasets, they are often extremely delicate to information steadiness.
One incident occurred when the group launched an replace aimed toward enhancing efficiency on blurry label photographs. Whereas the brand new mannequin dealt with unclear photos higher, it unexpectedly carried out worse on sharp ones. The problem was detected rapidly: the dashboard confirmed a spike in errors in eventualities that had beforehand been steady.
“We instantly rolled the mannequin again to the earlier model — because of function flags, it took simply minutes. Then we adjusted the dataset to cut back the bias towards blurry photos, examined the mannequin offline, and solely redeployed it after confirming constant efficiency,” Manu explains.
This swift detect-diagnose-recover strategy permits the group to deploy even high-risk modifications with minimal disruption, sustaining system stability whereas constantly enhancing efficiency.
Why Not All the pieces Ought to Be Automated
Regardless of ReturnQueen’s emphasis on automation, some processes are deliberately stored human-driven — particularly when belief takes priority over technical accuracy. A first-rate instance is dealing with high-value disputed returns.
If a buyer claims an merchandise arrived broken however the courier insists it was intact at pickup, the scenario not often has an easy reply. Even with photographs, logs, and timestamps, an automatic system may misread context, resulting in a choice that would harm buyer belief.
“After all, we may develop a mannequin to research photos and textual content, however the price of getting it flawed is simply too excessive. We intentionally maintain these edge circumstances underneath human oversight — folks can grasp nuance and talk with empathy, relatively than simply processing information,” Manu explains.
By sustaining this steadiness, ReturnQueen ensures that whereas routine, scalable duties are automated, human judgment stays central the place stakes are too excessive to rely solely on expertise.
How ReturnQueen Evaluates Know-how Choices
For Manu Ajith, efficient product improvement means instantly connecting technical options with tangible, real-world influence — each for the person and the enterprise. Whether or not it’s a brand new function, algorithm adjustment, or infrastructure change, the main target isn’t on complexity however on measurable outcomes.
“The important thing query is: what downside are we fixing, and the way can we show it? It’s not sufficient to consider it really works — we’d like information, person periods, and metrics that clearly present enhancements in simplicity, pace, or readability,” he says.
At ReturnQueen, the benchmark for achievement is utility, not novelty. Each technical change is required to reveal worth via proof relatively than assumptions, holding the group targeted on sensible, user-centric enhancements.
Have you ever learn?
The World’s Finest Medical Faculties.
The World’s Finest Universities.
The World’s Finest Worldwide Excessive Faculties.
The World’s Finest Enterprise Faculties.
The World’s Finest Style Faculties.
The World’s Finest Hospitality And Lodge Administration Faculties.
