How a virtual customer can provide continuous insight throughout the delivery lifecycle
Imagine your last Sprint Review. You present a new feature to stakeholders, but as usual, none of the key customers are in the room. Now imagine Mr. X – a “virtual customer” – sitting quietly in the corner, asking tough questions, giving real-time feedback, and representing the voice of the actual customer.
Research shows that only about one-third of developed features deliver the value intended by the Product Manager. One of the main reasons is the lack of continuous and meaningful customer feedback during development.
Fast and ongoing feedback is a core principle in developing new products and solving complex problems. The goal is to get input as early as possible, with customer or end-user feedback at the top of the scale. In reality, customers are rarely available early in the process, and often do not attend even the Sprint Review.
This article explores how AI can be integrated into the team’s workflow to consistently represent the customer’s voice – from early design all the way to the Sprint Review.
Although the examples here use GPT, the same principles apply with any enterprise AI module, depending on organizational policy.
The main component for enabling continuous feedback is the “AI Customer Expert” – a virtual customer that is always with the team. It is designed to accumulate knowledge over time, learning continuously about the company’s main customers.
To start, an AI project should be defined – a common term in GPT as well as in other AI solutions – into which all relevant customer and persona information is loaded. If product and marketing teams have not yet mapped the key personas, this becomes the first step, which AI tools can also help with.
In addition, the model can be enriched with company and product data (e.g., official website content), as well as real customer feedback from interviews, surveys, or service calls. This knowledge can represent different customer types or focus specifically on the company’s primary customers.
Over time, the AI’s knowledge can be refined by evaluating the quality of its feedback. This way, the model becomes more accurate and valuable for the team.
Looking ahead, the natural evolution is to create an Agent that works continuously and automatically. Unlike a one-time GPT query for feedback, an Agent can integrate directly into the tools the team already uses (ADO, Jira, Figma, Slack), track context across time, flag issues, and provide near real-time feedback. This makes the virtual customer not just reactive, but a constant participant in the process.
With the virtual customer available, it is important to involve it at every workflow stage.
We all know Spotify, which we will use here as an example to illustrate the concept.
In the early design phase of a new feature called “Shared Mood DJ”:
To go beyond conceptual discussion, we also moved to rapid prototyping:
Created a basic screen in Figma and exported an interactive mock-up.
Uploaded the prototype into tools such as Base44 / Lovable to run it against the virtual customer.
“We asked the virtual customer, acting as the persona (Dana), to simulate a simple journey: invite friends and start a shared playlist.”Result: Confusion around button labels and lack of clarity about how new participants could join.
Team Value:
By combining knowledge loading → focused prompts → prototype testing with standard tools, the team was able to identify critical issues before development started.
The insights were logged back as backlog items for clarity and transparency improvements.
The virtual customer can also be activated as the team completes stories. Each completed story can be reviewed for quick feedback. This supports deeper communication with the Product Manager, exposes risks, and can even lead to feature adjustments before work is finished.
The key is to use AI critically: prompt it to challenge assumptions, suggest alternatives, and highlight risks, not only to confirm existing ideas. Continuous, critical feedback helps catch problems earlier and reduces wasted effort.
At the Sprint Review, the feedback gathered from the virtual customer throughout the sprint can be combined with input from stakeholders. This adds depth to the discussion and helps identify gaps earlier.
Stakeholder reactions also serve as new data for the virtual customer, improving its understanding of customer needs. Most importantly, real feedback from actual customers must remain the top priority, feeding both the team and the AI for continuous improvement.
Throughout the development of the “Shared Mood DJ” feature, the virtual customer – Dana – stayed with the team as part of the process:
By the time the Sprint Review arrived, many usability issues had already been resolved.
At the Review itself, stakeholders raised a business concern around scaling the feature for large groups and Premium pricing.
This concern was also run through the virtual customer, enriching the perspective with the customer’s voice.
Team Value:
Stakeholder input was fed back into the AI knowledge base, strengthening Dana’s future feedback.
This created a continuous learning loop where the virtual customer improved over time and stayed relevant for upcoming features.
This approach creates a continuous learning loop, where each cycle improves the quality of team feedback. By combining real customer input with virtual customer input, the gap between what we think customers want and what they actually need becomes smaller.
Integrating AI in this way delivers clear business value, with measurable ROI through a higher percentage of features that truly provide value to the organization’s customers.
This article was developed and written by Oded Tamir, edited and presented with the assistance of GPT.
Request for additional information and prices
Subscribe to our newsletter, and stay updated on the latest Agile news and events
