Guiding principles for AI development at Dexory
January 8, 2026
By Divya Gautam, Head of AI & Chris Coote, Director of Product
At Dexory, AI is the translator of physical reality into digital insights. While it opens up competitive ways to solve complex problems and unlock efficiency, we believe AI is never an end in itself. Adding ‘intelligence’ to products without a clear strategy creates confusion, not value.
That’s why we put together this piece: to highlight the core principles we use to build AI-powered solutions like DexoryView, and the standards that guide our AI & Science organisation as we tackle the industry's hardest problems.
Start with the problem, not the solution
We do not build AI for AI’s sake. At Dexory, every project begins with a rigorous scientific assessment: Is AI the most robust tool to solve this specific problem?
We follow a structured process that mirrors scientific discovery, moving from hypothesis to validation:
- Exploration: what is the customer problem? In logistics, the problem is often clear, but the data isn’t, which is why we rely on Physical AI to generate new ground-truth where legacy systems fail.
- Research: what is the most robust solution that solves the problem efficiently? We prioritise simplicity unless the complexity of the data demands a deep learning approach.
- Prototype: build fast, test, iterate, and gather evidence
- Pivot & scale: Be ruthless about stopping projects that don’t prove value, and scale the ones that do.
Knowing our North Star keeps us on the right track. The goal isn’t perfection on day one, instead, it’s velocity, learning, and rapid improvement through feedback
Be anchored in the product strategy
Although it may sound obvious, many companies fail to do this: build a clear product strategy.
A well-defined roadmap is the foundation for everything that follows. Knowing what you want to achieve, over what timeframe, and what problems need solving along the way keeps you from being distracted by hype cycles or technology searching for a problem.
A practical way to assess opportunities and prioritise them is using the ICE framework:
- Impact - If we do this, how much value will it create, in terms of, for example, revenue, retention, or customer satisfaction?
- Confidence - How sure are we about the impact and the effort, based on the data we have, customer validation and past learnings?
- Ease: How easy is it to deliver, how long will it take and what other factors does the deployment depend on?
As we operate in the Deep Tech space, this process has to be adjusted by adding another critical criteria - Defensibility. Prioritising Ease can lead to building commodity features, so choosing the “hard” path - like building our own vision models - creates uniqueness.
Strategy acts as the vector for our engineering efforts. In Deep Tech, applying AI to a vague strategy accelerates failure. Clear strategic bounds ensure that when we apply the power of AI, we are scaling our competitive advantage and not just amplifying noise or complexity
Build only if it makes sense
Building an AI-powered solution might sound appealing - it’s new, exciting and there are many business opportunities to be unlocked. But does it always make sense to build your own?
The way we see it, a solution is worth the investment of time and resources only if it creates strategic advantage - for example, we build our own Machine Learning models when we have a unique data moat (like our Lidar and Vision streams) that off-the-shelf APIs can't compete with.
If the use case is already commoditised, we will always go with buying existing solutions instead of reinventing the wheel.
Get proper buy-in
When proposing a new solution, technology, product to a customer, make sure you understand who will use it, how, and what problems they have. In industrial environments, AI must be explainable. We don't just give an answer, we show the evidence e.g., the image of the damaged rack. Successful AI adoption is about making people’s jobs easier, faster, and more efficient.
To get buy-in:
- Build trust and make the experience relatable, so people understand the logic and the value behind it
- Show how your solution will solve problems, what KPIs will it achieve, what unique value it can create
- Create real change for people at all levels - if your product is used by other employees than the exec team you sell it to, get both bottom-up and bottom-down buy-in. Bottom-up engagement will beat top-down enthusiasm every time.
Stay close to the customer
At least 50% of a product team’s time should be spent with customers, understanding their pain points, workflows, and the jobs they’re performing. Product, design, and engineering must operate as a tight trifecta, not separate silos.
Too many layers between the builder and the user create blind spots. Direct exposure to customers motivates teams and grounds development in reality - it bridges the gap between the solution developed in the office and the real needs and pains of the customer.
Embrace uncertainty, but reduce it fast
Building with AI means navigating uncertainty at many levels - scientific, market, and customer uncertainty. The trick is to reduce it quickly and systematically.
Research and validation remain critical, but shipping in the era of AI happens faster. We identify problems, test solutions, learn and deploy iteratively.
We know what we want to achieve, so we can easily answer the question: what does good look like?
Build the right team and culture
Last but not least, all of these principles can be put into practice only with the right team and the right organisational culture. Curiosity, resilience, combined with engineering rigour and a strong mission is what advances any product or solution.
AI - and Agentic AI specifically, are powerful amplifiers that can work in both ways - they can scale clarity and productivity, or chaos. To use this new technology to our advantage, in everything we build, we choose to use it deliberately, to bring more visibility to our customers’ data and better, faster, workflows.
We also do it by being transparent about what we do and by investing heavily in communicating our vision and the benefits we’ve identified. To learn more about how we see AI shaping the warehouse of the future, read our latest white paper on Adaptive Warehouses.