InHouse Or OffTheShelf? A CSuite Decision Framework For Buying Vs. Building An AI Model
Raj Neervannan is the co-founder and CTO of AlphaSense , a market research and data platform used by the world's leading companies.
Artificial Intelligence (AI) and Generative Artificial Intelligence (GenAI) in particular are changing the business landscape. Despite the constant pressure on revenues, Korn Ferry predicts that enterprise AI budgets will increase by up to 27% this year.
As companies of all shapes and sizes explore and implement GenAI, one key question for decision makers is whether to buy the technology from an existing vendor or develop it themselves. There are many considerations in deciding which approach will work best for your business, and leaders need to understand the pros and cons of both.
Advantages and disadvantages of standard purchasing
The main benefits of choosing a standard generative AI model are simple: time-to-market, cost-effectiveness, ease of use, proven performance, and continuous update support. Software vendors offer pre-built AI models for shared use such as customer service chatbots, where businesses can adopt with minimal time and cost to build a model from scratch.
They are typically tested and proven to work in a variety of situations and reduce the technical complexity of implementation. These models are trained on publicly available customer service content and represent what a typical customer service conversation looks like.
Large companies such as Airbnb and Deutsche Telekom invest in large, commercially available language models (LLMs), including development tools and integrations, that enable different teams within the company to use these models to achieve common goals.
Additionally, purchasing an AI model from a vendor that specializes in GenAI can provide faster access to new features as they roll out, giving you the ability to adapt to your needs.
However, business solutions are designed for different use cases, so the specific functions of each company are not fully prepared, especially if the actual data is different from the conditions in the model. If you're building a product and see it as a competitive advantage, it can be useful to teach the model custom content and tell it to work in a way that meets those requirements.
Licensing costs and supplier lock-in carry additional long-term business risks. An external model may require data to be transferred to a supplier, which increases privacy and security concerns, especially for companies with sensitive data.
Advantages and disadvantages of internal construction
The Genie model ensures that in-house creation is practical and has features that are customized and specific to your business needs. A custom AI model is compatible with your existing platform and technology stack because the engineers who created the model work closely with your developers and team, giving you flexibility and control. Intellectual property can provide a competitive advantage while ensuring the protection of your data.
A proprietary GenAI model may be innovative, but it inevitably requires a large investment. The current cost of building and operating such a model can be from ten to one hundred million dollars, with a long time to market and the risk of failure.
One of the most expensive reasons to build Genie models from scratch is the high cost of hardware and talent required. The high-end GPUs needed to train these models are rare and even when available, they are still very expensive.
Additionally, engineers with relevant research and development qualifications and the leadership skills necessary to manage the process from start to finish command salaries in the seven- to eight-figure range. If your company doesn't have a team of experienced AI engineers, you'll need to hire new employees for these roles. According to data from LinkedIn, it takes an average of 49 days to hire an engineer, longer than many careers, including finance, IT and healthcare.
How we go through the decision making process
At AlphaSense ChatGPT we decided to buy or build GenAI before offering it to the masses. Our vision was to apply GenAI to help financial analysts and other professionals conduct their business research faster and more accurately. We've researched the best AI models available at the time, as well as affordable alternatives, that we've been able to sift through through our own documentation collections and annotation guides. Even the original formal models had limitations, such as some contractual requirements restricting the sharing of documents with third parties (even if the documents are publicly available).
As our use cases evolved from pre-generated responses to multiple documents, search results, and eventually full chat functionality, we recognized the need and challenge to maintain control over intellectual property, privacy, cost, quality, and performance. It was also important to ensure that AI-generated descriptions were relevant and relevant to proven data sources for each user query.
In the end, we decided to build our own models based on flexible open source templates for business, to achieve high levels of performance and privacy, and control costs, extensibility and internal customization.
Conclusions
The decision to buy or build with GenAI is critical. To make the right decision, leaders must carefully analyze the required use cases, evaluate available options and available solutions, be strategic with their resources, and follow a path that is best for their business over the long term.
The Forbes Technology Council is an invitation-only community for top CIOs, CTOs and technology executives. Am I eligible?
