Should you build or buy your AI? A primer for business leaders
Multimodal AI is transforming how businesses in media and entertainment extract value from their visual content. But any business wanting to leverage AI for its media assets faces a pivotal decision: whether to build tools in-house or buy a ready-made solution.
While building a custom solution offers control, it demands extensive resources and expertise. Alternatively, existing platforms can simplify implementation and accelerate time-to-value.
The decision is complex, and will ultimately depend on your unique business circumstances. To help you navigate these decisions, and kickstart the conversation internally, we’ve put together this primer for business leaders assessing the AI landscape.
In this quick overview, we’ll cover:
- The key considerations behind the build-vs-buy decision
- The business impacts of each stage
- A framework for evaluating your next steps
(If you’re a technical leader, we recommend checking out the full whitepaper)
What is Multimodal AI, and Why Does It Matter?
Multimodal AI uses machine learning models to analyze unstructured media content in new ways, and deliver insights that haven’t been possible before. Three key aspects are:
- Interrogating different data formats together: you can now search across videos, images, and text sources simultaneously. This can unlock insights that aren’t discernible through traditional discovery methods, where asset types are segregated.
- Greater flexibility for users: you can bring AI powers to generalist teams, through natural language queries, image-to-image search, or combinations of these, with SQL available too.
- Customization and learning: users can teach the AI phrases and ideas in minutes – something that previously required specialist skills, resources, and weeks or months
For Media and Entertainment (M&E) businesses, multimodal AI can transform vast archives of unstructured data into actionable insights and monetizable assets. It has the potential to unlock untapped opportunities in content discovery, monetization, and customer experience.
"Most visual content is invisible to traditional search tools due to patchy or non-existent labels—a problem multimodal AI is uniquely equipped to solve."
— Sergey Astretsov, Head of Product at Coactive AI
We’re seeing M&E and retail use cases like:
- Discovery: natural language search is helping generalist users to find niche assets faster, accelerating business activities like content licensing and marketing campaigns
- Moderation: automating better trust and safety outcomes for visual content
- Analysis: correlating engagement data (e.g. ratings, viewer demographics, social shares) with the success factors (semantic metadata) of visual content – to inform future content strategies
Multimodal AI is the engine driving a new era of efficiency and innovation. But choosing and implementing the right solution is a huge challenge. Both the technologies and the use cases are evolving fast—so designing for flexibility is key.
Each layer of the process—preprocessing, model selection, storage, APIs, and user experience—comes with critical challenges. Let’s explore these layers and their business implications.
1. Preprocessing: Building the Foundation
Preprocessing converts raw images, videos, and other assets into AI-ready formats. This includes encoding files, handling diverse formats, and extracting metadata.
The challenge: Poor preprocessing can create inefficiencies that ripple across the system. For example, many teams try to run AI analysis directly on their high-resolution assets, like 4K videos. This slows processing times while inflating computational costs.
The solution: Preprocess your 4K files into lightweight 480p proxy versions before running AI operations. This simple step reduces file sizes by up to 90% while maintaining the accuracy needed for search and analysis tasks. Teams can now analyze 10x more content with the same computational budget.
2. Choosing and Maintaining AI Models
Foundation models are the backbone of multimodal AI. These pre-trained systems generate embeddings (representations of media content) that enable semantic search and other AI-driven tasks.
The challenge: Organizations often select foundation models based on a single strength, like object recognition, only to discover critical gaps later. For example, a sports broadcaster may invest months implementing a model that excels at identifying objects, only to find that it struggles to distinguish between people—a crucial capability for sports content. For any business, it’s also unclear which foundation models will endure. Dependence on a single model creates a critical vulnerability in your AI operations.
The solution: Implement a flexible model architecture that allows for easy switching and knowledge transfer. When better models emerge, your team can upgrade without losing the custom training and fine-tuning they've already done. This preserves institutional knowledge while keeping the system competitive and accurate.
3. Storage and Retrieval: Speed Matters
Storing millions of visual assets and their embeddings requires careful planning. Decisions about storage architecture directly affect the speed of retrieval, influencing user satisfaction and operational efficiency.
The challenge: A production company with 10 million video clips stored everything in single-tier cloud storage to minimize costs. The result? Multi-second search delays that disrupted creative teams racing against deadlines.
The solution: Implement a tiered storage architecture tailored to your usage. Keep high-priority embeddings on fast SSD storage, while less-accessed assets could move to cost-effective cloud storage. This strategic approach delivers millisecond-level search speeds for frequently used content while balancing overall storage costs.
4. APIs: Bridging AI with Business Applications
APIs are the connectors that allow AI-powered functionality, such as search or tagging, to integrate seamlessly with business applications.
The challenge: Maintaining an API can be tedious, and is often overlooked by in-house teams when planning their resources. For example, a media licensing platform built a basic internal API for their AI tagging system, but couldn't maintain it effectively. When users requested essential features like batch processing and real-time feedback, the dev team was tied up with other priorities. Critical API improvements took months or never materialized.
The solution: Use production-ready APIs backed by dedicated support teams. You’ll have core features from day one, helping your teams to leverage AI insights faster. Having dedicated tech support also means you can add capabilities as your needs evolve.
5. User Experience: Simplifying Complexity
For AI tools to deliver value, they must provide end users with a great experience. This means your backend AI systems must integrate seamlessly with a user-friendly interface (UI).
The challenge: Say a video streaming service invests in sophisticated AI tools for content tagging and categorization. But the interface is complex, so adoption stalls among generalist teams, and the tool goes underutilized.
The solution: Prioritize an intuitive UI that aligns with existing workflows. When AI tools feel familiar and accessible to generalist teams, adoption increases naturally. This accelerates content operations across your organization.
The Bigger Picture: Build or Buy?
The decision to build or buy a multimodal AI solution hinges on evaluating your organization’s priorities, resources, and time constraints. Building in-house offers control but comes with high costs, longer timelines, and increased risks.
On the other hand, an external solution provides a faster path to implementation but requires great trust in your chosen partner. Here’s what Coactive enterprise client Emplifi has to say:
“Our brands have been able to clear their content backlogs, making their workflows much smoother. We aim to make sure they are spending time on our platform efficiently, and these results show we’re achieving that. It’s like night and day.”
– Heidi Eggert, Senior Product Manager @ Emplifi, a Coactive enterprise client
Delaying or mismanaging AI adoption can result in lost opportunities, while competitors gain the edge by unlocking the full potential of their media assets. Coactive AI is here to partner with you, and help bring the powers of multimodal AI to your teams.
Next steps?
- Share the full whitepaper with your technical colleagues for a deeper dive
- See the Coactive platform in action – get in touch for a demo
