AI-Enabled Searching in Pinterest

In the era of Artificial Intelligence, most products are leveraging AI to stand out and improve their offerings. As a frequent user of Pinterest, I have experienced instances where the platform did not yield the desired results due to the absence of relevant content on the platform. To address this, I devised a feature that can benefit both Pinterest users and the company. The concept is straightforward- leveraging AI in Pinterest’s search feature to generate images based on text prompts, similar to DALL-E 2 or MidJourney. In this article, we will explore the two approaches Pinterest can take to implement this feature- outsourcing the task and using pre-trained models or utilizing their in-house team, considering which method would be more feasible and effective.

Problem

As a fan of LeBron James, I often seek out images of him for motivation, particularly ones where he is wearing a crown like a king. Unfortunately, my recent search on Pinterest for such images has yielded unsatisfactory results. Despite my attempts to browse and scroll extensively, I have been unable to find the specific images I am looking for. This may be due to Pinterest’s search algorithm not being optimized for highly specific or niche results or simply because the desired content has not yet been uploaded to the platform. Consequently, this has resulted in a suboptimal user experience, which can be frustrating and time-consuming.

Solution

Develop a search feature on Pinterest that not only generates relevant content but also displays related content. The proposed solution involves utilizing AI to generate images based on user search queries.

Upon clicking on an AI-generated image, multiple alternate images, also created by the AI, will be displayed underneath.

This solution is inspired by the MidJourney Bot on Discord, which has been used to successfully generate such results. By integrating MidJourney’s image generation capabilities into Pinterest’s search functionality, the user experience could be significantly enhanced.

How to achieve this?

Buy (License/Outsource)

1. Conduct a feasibility study: The first step is to assess the feasibility of integrating MidJourney’s AI algorithm with Pinterest’s search algorithm. This will involve conducting market research, identifying the target audience, and analyzing competitors. The product manager should also work with the development teams of both companies to evaluate the technical requirements and potential challenges.
2. Develop a product roadmap: Based on the feasibility study, the product manager should develop a detailed product roadmap outlining the specific features and functionalities of the AI-generated content feature. The roadmap should prioritize features based on user needs and the potential impact on the business.
3. Develop a minimum viable product (MVP): The product manager should work with MidJourney’s development team to develop an MVP of the AI-generated content feature. This will involve creating a prototype that demonstrates the key features and functionalities of the feature. The MVP should be tested with a small group of users to gather feedback and identify any issues that need to be addressed.
4. Refine and iterate: Based on user feedback, the product manager should refine the MVP, addressing any technical or usability issues. The product manager should work closely with the development teams to ensure that the AI-generated content feature is fully integrated with Pinterest’s search algorithm. The feature should be tested with a larger group of users to validate its effectiveness and usability.
5. Launch and monitor: Once the feature is fully integrated and tested, the product manager should launch it to all users. The product manager should monitor the feature’s performance, analyzing user feedback and usage metrics to identify areas for improvement. The product manager should work closely with the development teams to iterate on the feature, continually improving it to meet user needs and business goals.
6. Measure success: The product manager should establish metrics to measure the success of the AI-generated content feature, such as user engagement, conversion rates, and revenue impact. The product manager should regularly analyze these metrics to track the feature’s performance and identify areas for improvement.

By following these steps, the product manager can ensure that the AI-generated content feature is developed and launched successfully, providing Pinterest users with a more robust and effective search feature.

Make (In-House Approach)

Developing an in-house image generation AI for Pinterest would require a detailed plan and allocation of resources. Here is a plan that outlines the steps and resources needed to build an AI image-generation system for Pinterest:

1. Conduct a feasibility study: The first step in developing an AI image generation system for Pinterest would be to conduct a feasibility study. The feasibility study would involve assessing the technical and financial resources required to build an AI system, analyzing the benefits of developing an in-house AI system as opposed to using a third-party solution, and estimating the time and costs required to complete the project.
2. Develop a project plan: After conducting the feasibility study, the next step would be to develop a project plan that outlines the timelines, milestones, and deliverables for the AI project. The project plan would also define the roles and responsibilities of the team members, and establish the communication channels and reporting mechanisms to ensure that the project stays on track.
3. Assemble a team: Building an in-house AI system would require a team of experienced AI researchers, developers, and data scientists. Pinterest could allocate existing resources from its AI teams to this project or create a new team specifically for this project. The team should have experience in developing image generation algorithms, deep learning, and neural networks.
4. Collect and curate data: The success of an AI image generation system depends on the quality and quantity of the data used to train the algorithms. Pinterest has a vast collection of images, and the team could use this data to train the AI system. The team would need to curate the data to ensure that it is of high quality, free from bias, and representative of the different categories of images that Pinterest users are interested in.
5. Develop and train the AI models: The team would then develop and train the AI models using the curated data. The AI models should be designed to generate high-quality images that are relevant to Pinterest users’ interests. The AI models should also be optimized for speed and efficiency to ensure that the image generation process is fast and scalable.
6. Test and refine the system: After the AI models have been developed and trained, the team should test the system to ensure that it meets the performance criteria. The system should be tested for accuracy, speed, and scalability. The team should also refine the system based on user feedback to ensure that the generated images meet the users’ expectations.
7. Deploy the system: Once the system has been tested and refined, it should be deployed to the production environment. The team should ensure that the system is integrated seamlessly into the existing Pinterest platform, and that it is secure, reliable, and scalable.

To allocate resources for this project, Pinterest could either allocate existing resources from its AI teams to this project or create a new team specifically for this project. If Pinterest decides to allocate existing resources, the company should assess the skills and experience of the AI team members and assign roles based on their expertise. If Pinterest decides to create a new team, the company should hire experienced AI researchers, developers, and data scientists with expertise in image generation algorithms, deep learning, and neural networks.

In terms of budget, Pinterest should consider the costs associated with developing an in-house AI image generation system. These costs would include salaries for the team members, hardware and software costs, and any third-party services required for the project. Pinterest should also consider the long-term benefits of developing an in-house AI system, such as improved user engagement and increased revenue.

Comparing Make vs Buy

There are several factors to consider when comparing the two approaches of using MidJourney’s AI for image generation versus building an image generation AI in-house:

1. Data: One major factor to consider is the data that will be used to train the AI. If Pinterest already has a large dataset of images and user behavior data, building an AI in-house may be more beneficial as it can be specifically trained on Pinterest’s data. However, if Pinterest’s dataset is not large enough, using MidJourney’s AI, which has been trained on a wider range of data, may provide better results.
2. Cost: Developing an AI in-house can be expensive, requiring significant investment in resources, including hiring specialized talent, infrastructure, and ongoing maintenance. On the other hand, using MidJourney’s AI may be a more cost-effective solution, as Pinterest would only need to pay for access to the AI.
3. Speed: If Pinterest is looking to quickly implement an image generation AI, using MidJourney’s AI may be the faster solution. Developing an AI in-house can take significant time and resources to train and test, whereas using an existing AI can be implemented relatively quickly.
4. Control: Building an AI in-house would provide Pinterest with greater control over the development and implementation of the technology. Pinterest would have more control over the algorithms used and the data used to train the AI. However, using an existing AI from MidJourney would provide less control over the technology.
5. Quality: Ultimately, the quality of the AI-generated images will be a crucial factor in deciding which approach to take. If MidJourney’s AI is able to produce high-quality images that meet Pinterest’s needs, it may be a more viable option. However, if Pinterest’s in-house AI is able to produce superior results, it may be worth the investment.

In summary, the decision of whether to use MidJourney’s AI for image generation or develop an in-house AI will depend on several factors, including data, cost, speed, control, and quality. As a product manager, it would be important to carefully evaluate each option and consider how each approach aligns with Pinterest’s overall goals and resources.

Conclusion

In conclusion, Pinterest’s proposed feature of leveraging AI in its search algorithm to generate images based on text prompts has the potential to revolutionize the way users search and find content on the platform. By addressing the problem of inadequate search results and providing related content to users, this feature can significantly improve user experience and engagement on Pinterest. While outsourcing to companies like MidJourney may provide a quick solution, building an in-house team can lead to better customization and control over the feature. In order to achieve this, Pinterest can allocate resources toward building an in-house AI team or collaborate with existing AI teams within the company. Overall, implementing this proposed feature can lead to increased user satisfaction, engagement, and growth for Pinterest.