SeniorWard is an AI-based decision security layer designed to help banks protect older customers from scams, starting in Italy and with phone calls. A small device listens to a phone call and analyses it alongside the bank transfer the customer is about to make. It then sends the bank a green, yellow or red signal, together with the information behind it. The final decision remains with a person at the bank.
My role
The idea for SeniorWard came out of our first brainstorming sessions. I suggested starting with a market I know well, Italy, and with two trends that make the problem particularly relevant: an ageing population and the number of young Italians moving abroad. Phone scams then emerged as the most promising angle because they create a gap between what happens during a call and what the bank can see when a customer authorises a transfer. With my teammates’ input, we turned this intuition into a defined proposal.
From there, my main contribution was defining the strategic direction: the problem, the business opportunity and the decision to make banks our paying customers. I also worked on the interactive prototype of the bank console, which shows how the system connects a phone call with a transfer. Towards the end, I put together the final presentation and coordinated the integration of my teammates’ contributions, helping with other parts to keep the proposal consistent.
Decisions and iterations
This shaped the main challenge: deciding how much of our vision we could realistically deliver as a starting point.
Decision 1: Start with a physical device and focus on phone calls in Italy. Our initial vision was a software solution covering calls, social media and SMS. However, this would require cooperation with banks, social networks and phone manufacturers from the beginning, which is difficult for a startup. We therefore narrowed the first version to phone calls and a physical device that needs only the older person’s authorisation, without relying on social networks or phone manufacturers. In the longer term, I see the device as a way to enter the market. Once we have credibility and the right partnerships, we could move towards software only, covering more channels.
Decision 2: Make the bank the paying customer. Next came the question of who pays. We initially considered a product that older people would pay for themselves, and later moved to a model in which banks pay an annual fee for each protected older customer. Banks have a direct interest in reducing fraud and meeting their regulatory obligations, and they are essential to the product: they hold the transfer information needed to assess risk and can act on an alert. The bank is therefore both the paying customer and a key part of the service.
Decision 3: Leave insurance coverage for a later stage. The same logic applied to insurance. We originally planned to offer insurance coverage when the system gave a green signal. We realised this would add complexity to a product whose core mechanism still needed to be established, so we left it for later. A possible future opportunity is to provide insurers with aggregated data, subject to privacy safeguards.
What failed or surprised me
The biggest surprise was that protecting customers from manipulated decisions can matter as much as protecting their credentials. In Italy, 74% of the value of fraudulent transfers in the first half of 2025 came from payer manipulation. The customers authorise the payment themselves, often because a scammer has persuaded them, so strong authentication alone cannot address the problem: the manipulation takes place outside the bank’s systems, often during a phone call. This is why SeniorWard connects the call with the transfer.
I also had to revise an assumption of mine: I took for granted that the device would always hear the call clearly. Since this is not guaranteed, the prototype follows a simple rule: if the call is not heard well enough, the signal is never green. I underestimated how many measures we needed to evaluate. At first, I focused on false greens: scams that the system classifies as safe. When we tested the prototype in an AI-assisted simulation using scripted scenarios rather than real data, the picture became more complex. We also had to consider unnecessary holds on legitimate transfers, the time the bank needs to decide after an alert, the money protected and the proportion of calls the device captures. These measures can conflict: a system that catches every scam but blocks too many legitimate transfers, or overloads the bank’s fraud team, would not be useful. It is not enough for the model to predict accurately; the whole process must work for the older person and the bank.
How feedback and evidence changed the work
These findings and feedback during the lessons pushed us to be more precise about what our first version should do. Rather than covering every channel used by older people, we focused on one workflow our prototype could show: matching a phone call with a bank transfer.
Informal conversations with older people who could use the service also influenced us. A physical device seemed more appealing to them than software alone, which supported keeping the device in our initial concept. These conversations were informal, though, and would need to be followed by more systematic user research.
What I learned from collaboration
Agreeing early on roles, expectations and in particular on processes makes a significant difference. A concrete example was the change in who pays. The move from the older person to the bank was not clearly shared with one teammate, and for a while our work went in different directions. Once we noticed, we realigned. This taught me that processes need to be defined from the start, especially communication ones, and that a change of this size has to be shared explicitly with the whole team.
How my view of AI and strategy changed
These experiences also changed how I see AI. Before this project, I thought of AI as a technology that could create value on its own. I now see that value comes from a whole ecosystem of processes and people around it. SeniorWard shows this: the technology can work, but the ecosystem that determines its success still has to be built. Some of our most important decisions were about how to enter the market, how to gain credibility and how to create partnerships with banks. More than the model itself, these are the variables that will decide whether we succeed.
The project also changed how I think about strategy. AI is a tool that makes us more productive in repetitive and technical tasks, such as building an Excel file or a PowerPoint presentation, and it helped us explore and communicate our ideas faster. However, the strategic decisions with the greatest impact must be taken by a human: which problem is worth solving, who should pay, which partnerships are essential and which features should wait. These choices require judgement and responsibility that a tool cannot take on for us. I came away convinced that AI is most valuable when it frees up time for people to make those decisions well.
Davide Famos (828176 – Team 36)
