The "Build vs. Buy" dilemma in software product development has been a central decision point for organizations for decades. This choice involves deciding whether to build software solutions in-house or to purchase pre-built software from external vendors. Each option comes with its own set of advantages and challenges, and the decision has evolved with changes in technology, market demands, and business strategies.

Early Days of Computing

In the early days of computing, during the 1960s and 1970s, the "Build" approach was predominantly the only option available. Organizations often had to build their software in-house due to the lack of commercially available software solutions. This period saw the rise of custom software development, with companies investing heavily in creating tailored applications to meet their specific needs. The process was time-consuming and resource-intensive, requiring skilled programmers and substantial financial investment.

There were no out-of-the-box alternatives to this, as the technology itself was new. Additionally, a distribution network for software was nonexistent. It was difficult to build software that would work across all systems of the same company, let alone multiple companies. Hence, there was no other option but to hire an expensive development team and get the job done if you saw the ROI.

Rise of Commercial Software

The 1980s and 1990s witnessed the rise of commercial off-the-shelf (COTS) software. Companies like Microsoft, Oracle, and SAP began offering robust, scalable, and reliable software solutions that could be easily integrated into existing business processes. This era marked the beginning of the "Buy" approach becoming a viable option. Businesses saw the benefits of purchasing software, including reduced development time, lower upfront costs, and access to vendor support and updates.

During this period, the software industry expanded rapidly, and specialized software solutions emerged for various business functions such as accounting, customer relationship management (CRM), and enterprise resource planning (ERP). The availability of these products led many organizations to reconsider their in-house development efforts in favor of purchasing pre-built solutions.

All software available during this period was on perpetual licenses, meaning customers simply needed to pay an upfront fee, and they could keep using the software with the same features for life. Such software did not need an internet connection and could be deployed on-premises for the customers. This was both a good thing and a bad thing. The good thing was that the customers owned the entire stack and could customize it, ensure privacy, and have predictability in the software behavior for the most part. However, the bad things outweighed the good ones by a lot, and these were solved in the SaaS era.

Internet Era, SaaS, and Subscription Pricing

The advent of the internet in the late 1990s and early 2000s brought significant changes to the "Build vs. Buy" decision. The rise of Software as a Service (SaaS) transformed how software was delivered and consumed. SaaS solutions, hosted in the cloud and accessible via the internet, offered numerous advantages over traditional on-premises software. These included lower total cost of ownership, scalability, and the ability to access software from anywhere.

Companies like Salesforce, with its pioneering CRM platform, demonstrated the potential of SaaS to deliver powerful, scalable solutions without the need for extensive in-house infrastructure. This shift further solidified the "Buy" option for many businesses, especially small and medium-sized enterprises (SMEs) that lacked the resources for extensive software development projects.

There was a nuance here. The software “could” be accessed from the cloud on any device, which benefited the customers, but the software “needed” to be accessed over the cloud, benefiting the seller companies. This enabled the companies to strategize a new form of pricing—subscription pricing, inspired by newspaper vendors and other essentials that a person needs routinely.

Since the software could only be accessed over the cloud, the seller companies could now cut your access to the software if the subscription wasn’t renewed. This gave the seller companies a source of recurring revenue, which added stability, freeing up bandwidth to innovate quickly without worrying about keeping the lights on.

Did the consumers of SaaS benefit from such models? Most of them did; others were left with no choice. A major benefit of a SaaS model for consumers is that new features get pushed quickly, support is readily available, and they needn’t spend a large one-time fee on purchasing the software. The SaaS model is still very commonly used by all Tech Product-based companies, but there was another shift that happened.

Modern Era: Agile, DevOps, and Open Source

The "Build vs. Buy" debate was influenced by new development methodologies, the rise of open-source software, and the growing importance of agility and innovation. Agile development and DevOps practices made in-house development more efficient and responsive to changing business needs.

Open-source software has also played a crucial role. Platforms like Linux, Apache, and MySQL, as well as numerous libraries and frameworks, provided organizations with powerful tools to build custom solutions without starting from scratch. The open-source community fostered collaboration, rapid innovation, and cost savings, making the "Build" option more attractive for many organizations.

The Next Generation of Software Building

Starting in 2022 (around the time OpenAI released ChatGPT), a new kind of software product emerged. Some product companies that existed before tried to integrate the powerful technology of LLMs into their existing products to delight their users, while a whole new set of companies built their entire products around LLMs.

There were several use cases that came up: chat and voice bots capable of carrying out realistic conversations with consumers in natural language, text analysis to analyze and organize large volumes of unstructured data, text summarization to compress unstructured data, and text, image, and video generation to aid marketing, advertising, and just about any writing task.

The early models of Generative AI were awe-inspiring but not really usable in most workflows. Most businesses leveraged this and fine-tuned these models for specific use cases or verticals. These models were much better than the foundation models in those specific tasks. However, that changed when big tech players got into an AI race and started training better and better models every quarter. These models all but took away the advantage of fine-tuned models and became better than the fine-tuned models at all tasks.

Although there are still a lot of great fine-tuned models out there, the trend is that general foundational models will eventually conquer them all, and within-context learning simple prompting will be more than enough to get the job done.

The Calculation Behind Build vs. Buy Decision

The Build vs. Buy decision is taken by companies considering two things: the cost/time to build and the value that the software brings to the table. Any decent software product nowadays is the sum of user research, UX design, and engineering efforts. On top of it, there is an opportunity cost of delaying the launch of the product. The value that the software brings to the table depends on the volume of transactions it would reach and the value per transaction.

Now, as long as the cost of building a product is higher than the value it can create within a company, the company will likely purchase the product and pay on a per-transaction or a monthly recurring basis. This will cut their operating costs or increase operating profits without adding any CapEx.

On the other hand, when the cost to build becomes so low that the internal market within a company can provide sufficient value, the company chooses to build a product themselves. Then, they may or may not externalize it for other customers, depending on the overall strategy of the organization or company. A notable example is AWS, which started to serve Amazon alone and then externalized it after seeing how much value it was able to add.

So What is Happening with Gen AI Products?

In Gen AI products, the newer foundation models are so good at completing so many tasks that they have pushed down the costs of engineering and the time to deploy a great product within the market. This has resulted in overall costs decreasing, and as seen in the graph below, the lower the costs, the lower the value that should be added by the product before a company decides to build its own.

With Gen AI, for most use cases, minimal CapEx is resulting in high ROIs for companies instead of purchasing expensive products in the market. Hence, there is a huge trend for building products instead of buying them, and very few Gen AI products have seen success in the market.

Why Aren’t Product Sellers Dropping Prices?

Traditionally in SaaS, 50% or more margins were common. This was because it was very difficult for a new entrant to build what an existing player would have built over the years without spending significant funds. In Gen AI, there are no old players, and everyone is building from scratch. Moreover, the time and complexity to build a set of features is rapidly dropping due to advancements in technology. Now, it doesn’t make sense to use third-party products that charge 50% margins instead of building your own.

So, why aren’t prices dropping? The reason is the operating costs of sellers. In traditional software, operating costs per marginal user or marginal transaction are famously known to be 0, and hence, the software industry was able to scale as fast as it did. But with Generative AI, it is no longer so cheap. Every additional user or transaction adds up to the costs of the seller company, and they must charge high prices to simply break even.

The seller companies could consider dropping margins and go for volume like in commodity markets, but VCs that have funded these companies wouldn’t like that. In software, a revenue multiple of 5 or 10 is common for valuation because 50% or more of that is assumed to be profits. With decreasing margins, it would be more and more difficult for VCs to make returns on their valuation. However, I do suspect prices will rapidly drop in the future for the Gen AI product companies to sustain in the market.

Conclusion

The decision of Build vs. Buy in Gen AI products will eventually come down to the cost of building and the cost of sourcing a product that satisfies all your use cases. The cost of sourcing is usually extremely low compared to the cost of building, but this will change as the cost of development drops.

‍