A Three-Step Method for Completing AI Product Requirement Analysis

[baoshi.rao]

The author shares insights from analyzing requirements for ToB image-based AI products, offering practical work experience.

The author hopes to discuss how to complete requirement analysis for ToB image-based AI products. The content reflects the author's habits in practical work and welcomes further discussion.

Requirement analysis is a fundamental skill for product managers. Only by thoroughly understanding requirements can errors in later product design be avoided. For ToB AI image-based products, the author generally follows the following analysis framework.

If market analysis is excluded, business requirement analysis should be the first step in product planning for product managers.

Especially for ToB products, only by deeply understanding users' business needs can the product be tailored to users, making them eager to purchase. The author typically collects user data from the following key points:

1. Business Context: It's essential to clarify the business background of the current requirements. Understanding the business context helps identify why users need the product—what bottlenecks in the current business state require AI solutions. It also aids in gathering industry attributes for later competitive analysis.

2. Usage Scenarios: Despite significant advancements in computer vision, challenges like camera angles, lighting, shadows, and occlusions still affect accuracy. During requirement research, it's crucial to clarify the scenarios where users intend to apply image recognition technology.

   • Security Scenarios: For example, detecting and identifying human attributes in logistics warehouses. Challenges include high-mounted cameras covering large areas, resulting in smaller targets and increased detection difficulty.
   • Fixed Camera Scenarios: Considerations include camera angles, vertical and horizontal distances to ensure clear images for algorithm detection.
   • Special Scenarios: Mobile phone captures and security scanners present unique challenges like inconsistent resolutions and angles, affecting algorithm accuracy.

3. Business Value Analysis: After understanding the 'where' and 'what,' the next step is assessing whether it's worth doing. Business value analysis is critical for evaluating product feasibility and pricing strategies.

   • Direct Analysis: Estimate daily usage volume, a common pricing metric.
   • Indirect Analysis: Assess business value, such as cost savings, efficiency improvements, or increased click-through rates.

Quantitative analysis is ideal for convincing stakeholders.

4. Algorithm Requirements: Product managers must translate business needs into algorithm requirements, including:
   • Scenario Description: Clarify imaging conditions and resolution.
   • Hardware Constraints: Evaluate if current hardware meets requirements; if not, consider alternatives.
   • Performance Metrics: Agree on accuracy and latency benchmarks with users.
   • Deployment Options: Decide between API or SDK based on model and hardware costs.

For example, video analysis may require high bandwidth costs, prompting consideration of image analysis with optimized frame rates and models.

Summary:
AI technology and scenarios should complement each other. Mature AI technologies align with business scenarios, while emerging technologies may require scenario constraints to ensure feasibility.

The product plays a role in driving the project forward by coordinating internal technology and users.

Summary:

Combining the above content with business needs analysis is generally sufficient to propose requirements to the algorithm team, and finally, clarify the timeline for the algorithm solution output with them.

Note: Algorithm requirements analysis requires AI products to continuously communicate and confirm various information between the algorithm team and users (sometimes this can be tedious, as users may not be able to estimate their expectations before integrating AI).

AI products need to break the limitations of traditional GUIs. The product forms provided by AI products externally can include not only complete front-end and back-end GUI systems but also API interfaces and SDKs.

In a previous article, How to Develop an SDK Product, we briefly described SDK products. This article focuses on the API form. The product requirements document should include interface inputs, outputs, algorithm accuracy, false detection rate, missed detection rate, interface performance metrics, and algorithm constraint rules.

Taking image recognition as an example, the core fields that the product needs to define include:

For GUI-based forms, the product team should focus on designing the product prototype, user experience, and user workflows. This is no different from traditional product design, except that algorithm requirements need to be separately assigned to the algorithm team. At the same time, the algorithm rules should be agreed upon. These rules refer to the constraints imposed on user interactions based on the current algorithm capabilities, such as requirements for file formats, naming conventions for uploaded files, and limitations on the recognition results returned to users.

As a product manager, it can be frustrating. In principle, the focus should be on users, uncovering their most natural habits. However, when AI technology is not yet mature, some user experience compromises may be necessary.

The biggest difference between AI products and traditional products is that there is an additional layer—the algorithm—between the product and development teams.

AI products must first clarify algorithm requirements and obtain feasibility verification results before further considering product requirements. Of course, algorithm requirements are also part of the product requirements.