A practical market-scanning workflow for small consumer products

Finding a small consumer software opportunity is not the same as finding a popular topic. A market scan can turn up thousands of mentions and still miss the useful question: is there a focused job people struggle to get done, and does it fit the business you can build and operate?

Veronata describes its studio as using software agents to identify, build, and operate consumer products. Its stated identification step has agents survey markets, score opportunities against a portfolio-fit model, and select a product to build. The company also says a human curates the agents and portfolio. It does not publish the scoring model or scraper configuration in the information available here. So the workflow below is a practical framework for structuring agentic market research, not a description of Veronata’s internal setup.

Start with a decision, not a scrape

Write down the decision the scan must support. For example: should a small team investigate a consumer product for a specific recurring task? Keep the question narrower than “what is trending?” A trend can be large, short-lived, or already well served. A useful scan should help compare several candidate problems on consistent terms.

Set boundaries before collecting data: the intended customer, the task, the product format you could plausibly support, and any areas you will not enter. For sensitive topics, include a human review requirement from the start. Market research can point toward health, financial, or other high-stakes needs; popularity is not evidence that an automated product is appropriate.

Collect signals with limits

Use agents to gather evidence from sources you are permitted to access. Keep the collection focused on public descriptions of problems: repeated questions, complaints about existing approaches, workarounds, and requests for a specific outcome. Follow each source’s access rules. Do not treat a scraper’s ability to collect material as permission to do so.

Store the source, date, short excerpt, and the reason a signal was included. Deduplicate near-identical posts and separate first-hand descriptions from speculation or reposts. This makes the evidence easier to inspect and reduces the chance that one widely copied complaint looks like many independent customers.

Ask an agent to summarize patterns, but retain links or source references in the research record for human checking. The summary is a navigation aid, not proof. Review a sample of the underlying material, especially when the agent labels a problem as frequent, urgent, or poorly served.

Score the opportunity, not the buzz

Use a short rubric with the same scale for every candidate. A useful starting point is to rate each item from one to five:

• Evidence of demand: Are there independent, recent descriptions of the same problem?
• Problem clarity: Can you describe the user’s task and the unsatisfactory alternative in one sentence?
• Competition and substitutes: Are existing products addressing the task, and what remains unresolved?
• Portfolio fit: Does the idea make sense alongside the studio’s current products and operating capabilities?
• Build and operating burden: Can the product be made and maintained without requirements the team cannot meet?
• Risk: Are privacy, safety, regulation, or reliance concerns manageable?

These are suggested criteria, not Veronata’s disclosed scoring factors. If you assign weights, write them down before scoring. Otherwise, a compelling idea can quietly change the rules in its favor. Keep a note beside each score explaining what evidence supports it, and mark missing evidence as unknown rather than filling the gap with an agent’s confident guess.

Use portfolio fit as a real constraint

Portfolio fit is not just a question of whether an idea sounds different from current products. Veronata’s live portfolio spans iOS, Android, web, and physical formats. Its products include a voice health companion, a nicotine-quitting guide, a baby-monitor system built around two iPhones, and a physical tongue scraper. That range makes the practical question broader than category overlap: can the studio support the product’s format and ongoing work?

For a candidate concept, write down what it would need after launch: customer support, monitoring, content updates, privacy handling, and any physical fulfillment. Compare those needs with the capabilities and constraints you can actually sustain. Do not award a high fit score simply because an idea shares a feature or audience with an existing product.

Make the shortlist earn its place

Have the agent produce a ranked list with evidence, score explanations, uncertainty, and the strongest argument against each candidate. Then have a person inspect the top few source trails and challenge the ranking. The next step is not automatically to build. It may be to collect better evidence, interview prospective users, or reject the idea because the operating burden is too high.

Keep the scan repeatable: use the same rubric, record when the evidence was collected, and preserve rejected ideas with their reasons. Over time, this gives strategists a way to compare automated product discovery runs without confusing more data with better judgment. The aim is not to make agents declare a winner. It is to make the next human decision clearer.