SERVICE
AI feature discovery
I help find where AI could add meaningful value to a product, then shape a feature people can understand and use. The work connects the user task with the capability, interface and evidence needed to decide whether it is worth developing.
A useful AI feature starts with something people need to accomplish. It might help interpret information, prepare content, find a relevant answer or carry out part of a task. We investigate where that assistance would matter and compare it with other ways to improve the experience.
I bring product discovery together with human–AI interaction design, making the opportunity tangible through concepts and appropriate experiments. The result can be a prioritised opportunity, a designed feature or a working demonstration of the behaviour that needs investigating.
FROM AN AI IDEA TO A USEFUL PRODUCT FEATURE
Find user tasks where AI could make a meaningful difference
Identify opportunities through people's needs, current effort and the limitations of existing approaches. We examine how people use the product, what they are trying to achieve and where the work becomes difficult.
Research and available feedback can reveal tasks involving repeated effort, complex information or content that needs adapting to different circumstances.
I help compare possible AI uses with simpler product changes or existing capabilities. A clearer interface, a better default or a rule-based step may address some needs more directly. We prioritise opportunities according to the value they could create and the assumptions involved, giving the team a reasoned starting point for exploring a feature.
Check whether the capability fits the product and its data
Investigate the information, model behaviour and connections the feature would depend on. A feature's usefulness depends on the inputs it can access and the conditions in which it operates. We identify what information is available, how current it needs to be and which constraints affect its use.
Depending on the idea, that may involve retrieving material from an approved source, interpreting an image or connecting a model with an existing tool. I help make those dependencies understandable and investigate them through a focused prototype or PoC. We consider response quality, time, cost and the consequences of mistakes, with technical collaborators involved where the question requires specialist implementation.
Design how people understand and control it
Make the relationship between AI assistance and human judgement visible in the experience. People need to know what a feature has produced, what remains uncertain and what they can do with the result.
I design the relevant interaction, including how outputs are presented, edited, accepted or challenged and how the experience responds when the capability is unavailable or unsuccessful.
In Zipli, an important design distinction is between information suggested by AI and information confirmed by the kitchen worker. Other products may need a review queue, a comparison view or approval before an action takes place. We develop the interface around the actual responsibility and task, with suitable controls for understanding, correction and oversight.
Test whether the feature earns its place
Evaluate realistic behaviour and user experience before committing to a wider implementation. We agree what a useful result would look like for the task, then investigate the feature with relevant inputs and people.
Where model behaviour matters to the conclusion, a working experiment helps reveal variation that a simulated interface would conceal.
Evaluation can consider task completion, output usefulness, correction effort and the ability to recognise a poor result, alongside cost or latency where relevant. I bring the findings into a recommendation for the next step: develop the feature, adjust its role, investigate a dependency or pursue a different approach. The decision stays connected to the value created for the user and the work needed to deliver it.
CASE STUDY
Where should we start?
See contact optionsYour message is on its way. I’ll reply to .
It didn’t go through. Copy your message and send it tohello@mareksevcik.com.