Implementing User Co-Creation: Close the Loop on Requirements First

Implementing User Co-creation: Integrate Feedback into Scheduling First

I spent the weekend tinkering with connecting user feedback back into the production system and hit quite a few pitfalls. Recently, Huxiu published an article covering everything from Ford to Xiaomi, with the core theme being how user voices re-enter production after a century of manufacturing. I lead a team of thirty, building B-end tools. We aren't qualified to build cars, but customers recently raised scenario-based needs similar to native electric pop-up roofs. Simply put, the roof rises electrically after parking, creating an extra layer of usable space inside. User co-creation depends on whether feedback actually enters product decision-making. These needs sound like configurations, but actually involve structural parts, supply chains, certifications, and after-sales service. So, I split user co-creation into two approaches and ran them through once.

The original process was very ordinary. Customer Success recorded feedback, Sales included requirements in weekly reports, and Product Managers filtered them weekly. I imported last week's customer interviews, tickets, and app store comments into a spreadsheet, roughly dozens of entries. From user quotes to entering scheduling, my tests averaged two days. The bottleneck lay in inconsistent descriptions. Some wrote "hope the roof lifts," others wrote "parking space insufficient." Product Managers first merged and deduplicated, then asked Sales if it was a real need, then asked R&D if it was feasible. A whole morning vanished. The advantage was clear responsibility: whoever proposed, filtered, or decided had signatures. The disadvantage was that users were too far from production; many needs died in row three of the spreadsheet.

These past few days, I've been testing a lightweight AI Agent capable of automatic summarization, classification, and task field generation. I threw the data into a new task in the WorkBuddy desktop version. In the interface, it broke each piece of feedback into three columns: user quote, AI summary, and risk tag. Here, I used Zhipu models for Chinese summarization and ZCode for generating structured fields. Combined with a supply chain dashboard I'd been using for about a month, I broke requirements down into software, structural parts, materials, certification, and after-sales.

For example, for needs like "parking pop-up roof," AI shouldn't just judge "feasible" or "infeasible," but break it down into roof structure, seals, lifting mechanism, interior space, camping scenarios, regulatory certification, and after-sales spare parts. For the same batch of feedback, tag generation took about ten-plus minutes, and manual review took about half an hour. Errors fell mainly into two categories: identifying casual user wishes as mandatory features, and treating sales pitches as genuine user needs. Later, I added exclusion words, using them for about two weeks to filter marketing terms like "ultimate," "leading," and "disruptive," resulting in cleaner outputs.

The surprise was that scattered feedback turned into a discussable table. Product Managers, R&D, and Sales could argue on the same table for the first time. There were also plenty of pitfalls. AI is good at induction, but it doesn't know if materials are in stock, how long certification takes, or how to manage after-sales spare parts.

AI can structure user voices, but it cannot automatically fill in production feasibility. This approach suits teams with stable user feedback entry points, product review mechanisms, and the ability to coordinate with supply chain or hardware colleagues. It's unsuitable if you only have a Demo without real users, treat user co-creation as marketing rhetoric, or lack scheduling and delivery owners. My biggest takeaway is that cleaning tags takes more time than generating them. The path taken by Xiaomi Auto is worth watching; they delivered their first car in 2024 and placed community, delivery, and smart manufacturing in one chain. Ford CEO Jim Farley has driven a Xiaomi EV for over half a year; traditional automakers are seriously dissecting this approach too. For us startups, don't learn the slogans; integrate feedback into scheduling first.