An interview with Jose Ignacio Orlando, by PM Damian Calderon.
Introduction
In AI product development, the focus often centers around Proof of Concepts (POCs). POCs are essential for validating the feasibility of AI solutions, particularly in high-risk projects where there is uncertainty regarding model accuracy or technical feasibility. They provide a structured approach to assessing whether an AI-driven solution is viable before committing to large-scale implementation. However, POCs are primarily designed for validation rather than deployment, often requiring further development to achieve scalability and seamless integration into business workflows.
While demonstrating this is crucial, i.e. to understand if available data is (good) enough for training custom ML models, it doesn’t hold very well in an era where it’s less likely to have to train a model from scratch but instead leveraging an existing GPT-like network. The true measure of success might usually lie in these cases in delivering tangible value to users and stakeholders. This is where Proof of Value (POV) comes in, a concept that prioritizes aligning AI solutions with real-world needs and outcomes.
We recently spoke with Dr. José Ignacio Orlando, Director of our AI Labs and one of Arionkoder’s machine learning experts, to dive deeper into what POV entails, why it matters, and how it’s been applied in groundbreaking projects like a Public Pharma Company and OncoRX. Here’s what we learned.
Interview: Exploring Proof of Value in AI
Q1: In your own words, what is Proof of Value (POV), and how does it differ from a traditional Proof of Concept (POC)?
Ignacio: Proof of Concept has always been about answering one main question: “Can we build this?” It focuses on technical feasibility—validating whether the model we train, the algorithm we design, or the system we craft, can function as intended with the data that we have. But a Proof of Value takes it one step further, as with it we ask ourselves: “Should we build this?” and “How does this solve real user problems?”
In a POV, we aim to prove what would be the actual impact of the AI solution, before even starting to craft it. It’s not about showing how the technology works; it’s about showing how it creates measurable value for the user or organization that will ultimately adopt it. This includes aligning that future product with current or future business objectives, understanding user needs and pain points, and exploring how that future system would integrate seamlessly into existing workflows.
In a nutshell:
| Criteria | Proof of Concept (POC) | Proof of Value (POV) |
| Objective | Validate technical feasibility and functionality of an idea or technology. | Demonstrate that the solution delivers measurable business value and/or ROI. |
| Scope | Focused on showing that the concept can work in principle, often through a prototype. | Broader in nature, encompassing business impact and market potential. |
| Number of AI Features | Usually only one, the most impactful one, which is explored through experimentation with a focus on accuracy | As many as possible, covering most of the AI use cases envisioned as relevant for the desired tool |
| Deliverable | A functional prototype with a basic UI and the best AI solution to the use case accomplished during time-boxed experimentation | A fully operational MVP-like software, with a finished UI and multiple early versions of the AI features implemented and working. |
| Target Audience | Primarily aimed at technical teams and decision-makers. | Involves business stakeholders, customers, and investors who are interested in the value. |
| AI Accuracy | All ML effort is put on achieving the highest possible accuracy for the selected AI feature, so the solution (if feasible) will be very accurate. | Multiple AI features need to be delivered, so experimentation is constrained for each one (in general one or two sprints top per feature). This has an impact on accuracy. |
| Team and Time | Most of the effort is invested in ML development and experimentation and in the implementation of a robust, scalable backend architecture. | The team is larger, including UX/UI designers for UI research, frontend developers for implementing a UI, QA specialists to ensure the application works as expected, and DevOps engineers for infrastructure procurement and CI/CD monitoring. |
Q2: Why is POV especially critical for AI-based products?
Ignacio: Well, AI comes with unique challenges. Unlike traditional software, AI often involves training predictive models, performing regular updates of them based on changes in data streams and user feedback, and dealing with outputs that can feel opaque or unreliable to users, like probabilities, explainability maps, etc. That’s why we prefer to run POVs for AI projects. Because we want to understand:
– User Trust: AI systems need to be intuitive and transparent. If users don’t trust the outputs, adoption will fail, no matter how accurate the model is. A POV aids us to better understand that and steer model design, prompt and agent development towards the right direction.
– Contextual Relevance: AI works best when it’s tailored to the specific context of its deployment. POVs ensure the solution is relevant and effective in the intended environment, without spending too much on the actual implementation of it. We can make a first version of those components, tailored to that 80% of the scenarios that already create value, and once the value is proven, then we can move forwards and improve it for dealing with edge cases.
– Iterative Validation: AI systems evolve over time, they’re not static. POV emphasizes ongoing evaluation to adapt the solution as user needs and conditions change.
In short, POV is essential for bridging the gap between AI’s potential, rapid (and cost-saving) experimentation, and its practical, real-world application.
Q3: How do projects like the ones we pursued in the past demonstrate the importance of POV?
Ignacio: Our collaboration with a Public Pharma Company and OncoRX are great examples of how focusing on POV transforms an AI project from a technical exercise into a meaningful solution.
We for example pursued a project for a public pharmaceutical company that wanted us to develop the building blocks of a contract management system designed to streamline how their legal team was handling this documentation. The core challenge wasn’t just building the AI’s capability to process contracts but ensuring that the system was intuitive and aligned with the workflows of legal and administrative teams. Our partner, who has experience in AI and is an advocate for its adoption in the company, came of course with some suggestions of AI components to incorporate to the tool. But it wasn’t until we involved users in the iterative design and let them interact with the tool that we identified the correct needs that they had. For instance, comparing contracts against their original templates, or being able to search using specific keywords, was something that was initially not considered as part of the project, and we had to scope it in to gain trust and advocacy from the final users. As a result, we created a tool that enhanced efficiency and clarity, enabling faster contract approvals and reducing errors.
With OncoRX, we shifted our focus by adopting a Proof-of-Value approach to revamp an existing AI workflow. The challenge was clear: integrating diverse lab reports for each new customer was a major bottleneck, slowing down product adoption. We kicked things off by spending one day testing ChatGPT to see if we could streamline the process. This was kind of a basic POC that enabled us to validate accuracy. When that quick test showed promise, we dedicated three more days to build a HIPAA-compliant, LLM-based agent module. By zeroing in on the most pressing issue, our POV-driven strategy not only removed a key obstacle but also made the tool far more adaptable for its users. And the result was a solution both technically solid and remarkably user-friendly.
Use POVs to speed up AI Development
The shift from POC to POV is not just a buzzword—it’s a practical method to accelerate the creation of useful AI products, bridging the gap between feasibility and meaningful impact. POV forces us to focus only on what truly delivers value, helping ML and product teams cut corners by prioritizing usability, relevance, and impact. This ensures the solutions we create are not only feasible but also highly meaningful.
As AI continues to permeate industries, focusing on POV will be essential for driving adoption and achieving real-world outcomes. Are we, as product shapers, ready to prioritize value over novelty and align innovation with the needs of people who use our solutions?
At Arionkoder, we believe this is the only path forward. Projects like the ones we referred to have taught us that technology is only as good as the value it provides. The future of AI isn’t just about building what’s possible—it’s about building what truly matters. Let’s build it together: contact us to explore how we can help.