Custom Software In Retail & Logistics: The gaps between Design and Reality

Retail and logistics performance is increasingly shaped by how well operational systems and digital tools work together. Higher service expectations, tighter capacity, and more complex networks make it harder to manage performance through local fixes alone – teams need stable ways of working, disciplined daily management, and reliable information flows that support fast decisions at the frontline.

At Kaizen Institute, our core work is to improve performance by designing and deploying end-to-end operating systems: aligning strategy and execution, standardising the way work is done, strengthening daily management, and building the culture and capabilities that make improvement stick over time. In practice, that work often intersects with technology – not as a “tech-first” agenda, but because visibility, data, and system handoffs increasingly determine whether operational routines are actionable, scalable, and sustainable.

With that context, we interviewed an Exaud Software Engineer to capture practical lessons from the gap that often appears between software design and operational reality. Exaud is a software engineering company with experience building tailored solutions and integrating them into existing enterprise landscapes. The interview below focuses on what tends to go wrong (assumptions, edge cases, messy data, integration constraints, adoption), what works in real deployments, and what operational and tech teams can do differently early on to avoid expensive rework and accelerate impact.

1. Who Exaud is and how they approach retail and logistics projects

Can you start by introducing Exaud and how your work intersects with retail and logistics operations?

At Exaud, we don’t just write code; we build digital tools that act as the ‘nervous system’ for complex operations. My role as Head of QA and Engineering Manager is to be the professional skeptic. I spend my time anticipating everything that could go wrong in a warehouse or on the road so that when the software reaches the user, it just works. In retail and logistics, a software glitch isn’t just an error message but it’s a delayed truck or a missing pallet. Our work is about ensuring that technology never slows down the physical movement of goods.

What kinds of projects does your team typically support — are they more about visibility, workflow automation, or system integration?

We usually support a mix of all three, as they are deeply connected. For instance, in our work with Ituran, a global leader in telematics, the project moved from simple vehicle tracking to a full-scale Decision Engine. We focus on:

• Visibility: Creating live dashboards (like our Dataran platform) so managers see exactly where their fleet is and how it’s performing in real-time.
• Workflow Automation: Using AI to handle “heavy lifting,” such as automatically converting call centre audio into searchable text or extracting key data from massive CSV files without manual effort.
• System Integration: Connecting different ‘clouds’ (AWS, Azure, Google) so that data flows smoothly from a sensor on a truck to a manager’s tablet.
  
  

Which teams inside the client organisation do you usually collaborate with — ops, IT, data, or in-store teams?

Because I’m a ‘micro-manager’ in the best sense, obsessed with details and deadlines, I make sure we talk to everyone. We bridge the gap between the IT/Data teams, who care about security and scalability, and the Operations teams, who care about speed and usability.

For projects that we have in progress we worked closely with their Innovation and Data leads to ensure our AI tools weren’t just ‘cool tech,’ but actual solutions for their fleet customers. My goal is always to plan for every scenario, so the IT team feels secure and the Ops team feels empowered.

2. What problems show up repeatedly

What are the most common operational bottlenecks you see (stores, DC, transport, returns)?

The biggest bottleneck isn’t usually a lack of data; it’s data paralysis. Companies often have massive amounts of information flowing in from trucks or warehouses, but it’s trapped in ‘silos’ or complex files that no one has time to analyse. I’ve seen operations slow down because managers are stuck doing manual data manipulation in spreadsheets just to find a simple answer, like which routes are underperforming or where safety risks are highest.

Beyond the obvious inefficiencies, what hidden factors — like process variability or unstructured data — make these problems hard to solve?

Variability is the silent killer of efficiency. In logistics, you aren’t just dealing with ‘clean’ data; you’re dealing with unpredictable weather, varying driver behaviour, and hardware that might send inconsistent sample rates. As someone with a QA background since 2008, I know that ‘unstructured data’ like voice recordings from call centres or messy CSV files is where the real problems hide. If the software isn’t designed to normalise these variables and clean the ‘noise’ automatically, the insights you get will be flawed.

3. Translating operations into software

How do you translate day-to-day reality from shop floors or warehouses into clear, usable software requirements?

I’m a firm believer in the ‘Gemba’ approach “you have to see where the work happens”. I translate warehouse reality into requirements by obsessing over the ‘what ifs.’ If a driver is using our app, I’m not thinking about the perfect office scenario; I’m thinking about them being in a rush, with poor lighting and gloves on. We build ‘Decision Engines’ that turn complex telemetry into simple, actionable alerts so the user doesn’t have to think – they just act.

Where do specifications usually go wrong — is it assumptions, missing edge cases, or misalignment between tech and operations?

Specifications usually fail because of optimistic assumptions. Most people design for the ‘happy path,’ but my job is to design for the ‘messy path.’ Specs go wrong when they miss the edge cases: What happens if the file upload is interrupted? What if the AI extracts the wrong keyword from a noisy audio call? I’m quite a micro-manager when it comes to these details because I want to plan for and minimise negative impacts before they reach the frontline. Anticipating 90% of the problems through rigorous QA reduces the impact of the 10% we couldn’t see coming.

4. Systems and integration realities

What usually exists already (ERP/WMS/TMS/OMS/POS) and what’s hardest to integrate?

Most clients aren’t starting from scratch; they’re sitting on a ‘legacy stack’ of ERPs and warehouse systems that have been around for years. The hardest part isn’t the new tech, it’s building the bridge to the old stuff. Integrating real-time data from vehicle sensors into a rigid, older database is a challenge. I’m always looking for ways to ensure these systems talk to each other without crashing, using scalable cloud architectures to handle the heavy lifting.

What are the most common integration mistakes, and how can teams avoid them early?

The biggest mistake is ‘over-promising’ on how fast systems can sync. Teams often underestimate the ‘dirtiness’ of data coming from different sources. To avoid this, I insist on early ‘stress tests.’ Don’t wait until the end of the project to see if the API can handle 100,000 entries. I’m a controller by nature, so I want to see those integration tests running in week two, not week ten. It’s about anticipating the bottleneck before it stops the business.

5. Data and governance

What kinds of data quality issues most often block progress?

Inconsistency is a huge blocker. You might have one system recording ‘trips’ and another recording ‘engine hours,’ and they don’t always align. We often find anomalies in large data files that a human would never catch. My approach is to build ‘Privacy Engines’ and audit-ready controls from day one. If you don’t bake data governance and privacy into the architecture, you’re just building a house of cards that won’t pass a GDPR audit.

6. Adoption, change, and day-to-day use

Why do frontline teams still struggle to use well-built systems?

Because ‘well-built’ often means ‘technically perfect’ but ‘operationally annoying.’ If a call centre operator or a driver has to jump through hoops to find information, they’ll stop using the tool. I’ve learned that enterprise readiness depends on role-specific dashboards. If I’m a manager, I want the big picture; if I’m a driver, I just want to know if my behaviour is safe or if my truck needs maintenance. If it doesn’t help them make a decision in 5 seconds, it’s a failure.

What rollout or change approaches reduce resistance — phased pilots, internal champions, or training patterns that work?

Phased pilots are my go-to. I don’t believe in ‘Big Bang’ launches. We start by moving from infrastructure to intelligence in stages. This allows us to find internal ‘champions’ who see the value of proactive alerts and real-time intervention early on. As a manager, I’m constantly planning these phases to ensure we can pivot if the first pilot reveals an edge case we didn’t expect.

How do you design user interfaces for operators who may not be tech-oriented?

I keep it incredibly simple. Operators are busy, and I’m obsessed with making the UI intuitive. We use advanced algorithms behind the scenes so the front end stays clean. For example, instead of a complex search bar, we might use an ‘Ask the AI’ feature where they can just type a question like they’re texting a friend. It’s about taking the ‘manual’ out of data management.

7. Measuring outcomes (neutral, evidence-driven)

Which metrics genuinely reflect improvement — service level, productivity, cost-to-serve, returns?

While I’m a fan of hard data like ‘reduction in fleet downtime’ or ‘increased safety scores’, the metric I value most is Time-to-Insight. In a fast-moving operation, it doesn’t matter if you have the data if it takes a week to understand it. True improvement is when a manager can ask AI a question and get a reliable answer in seconds. That speed allows for real-time intervention, which is what actually saves money and, more importantly, keeps people safe.

8. What’s next (industry overview)

Where do you see retail and logistics tech leaders shifting their focus in the next 12–24 months?

I’ll be honest: everyone is talking about AI and Cybersecurity, but I suspect nobody truly knows what’s coming next. In this context, the focus is shifting from ‘long-term planning’ to radical agility. I’m a planner and a controller by nature, but I’ve realised it’s becoming harder to predict the next 12 months when we don’t even know what the next 3 will look like. The goal for tech leaders now isn’t to build a 1 to 3-year roadmap; it’s to build systems that are resilient enough to pivot overnight. We need ‘Privacy-First’ architectures that protect us in a world where data is a weapon, and AI that helps us make sense of chaos (or install it).

If you could change one industry mindset about “custom software,” what would it be?

I’d change the idea that custom software is a ‘destination’ or a one-time purchase. In a world this volatile, software is a living tool. It’s an ongoing commitment to quality and adaptation. Stop looking for a silver bullet that solves everything forever. Instead, look for a partner who helps you stay prepared, minimises negative impacts when things go wrong, and ensures that even when the future is blurry, your operational data remains a source of truth.

Nuno Matos – Head of QA

With over 10 years of experience as a QA Engineer, Nuno is responsible for establishing and evolving formal QA processes, respecting and implementing good practices to achieve products and clients’ goals.