Many new AI-powered search solutions have been released this year, and each promises to provide great results, but as a technical or business leader, how do you select the one that is right for your organization – not just for today but also for your future growth stages? As a leader with longstanding  expertise in this industry space, we’ve published a new buyer’s guide checklist to help understand and compare the different technology approaches for AI search. 

Broadly speaking, the solutions can range from plug-and-play to do-it-yourself (DIY). They include:

• Search-as-a-service API solutions
• Ecommerce solutions
• Build with open source solutions

Capabilities can vary a lot, and your choice would also depend on factors such as time to market, ability to deploy developers and the IT infrastructure of your organization. Do you want a solution that includes only AI retrieval, or also AI-powered ranking and query processing? How much manual control do you need over results? Or do you just want the raw foundation to build something very customized for your business situation? Another critical element that needs to be considered would be the time, effort, and developer resources you would be able to commit — all of this would determine how you answer the question of whether to buy an off-the-shelf solution versus building one from the ground up. 

Our new guide can shed some insight into the differences and similarities between solutions. Here’s a preview of the guide and some top-line considerations that can help organizational leaders and teams make an informed choice when comparing solutions out there in the market. 

What does AI search refer to?

AI-powered search refers to search engines that use artificial intelligence (AI) algorithms and machine learning techniques to provide more accurate and relevant results for user queries.

Traditional search engines rely on a set of pre-defined rules and statistical-based algorithms to match user queries with the relevant content. However, these rules may not be able to capture the full meaning or context of a user’s search, which can result in irrelevant or incomplete results.

In contrast, fully-featured AI-powered search engines use a variety of techniques for query processing, retrieval, and ranking, including:

• Natural language processing (NLP) algorithms powered by machine learning AI to analyze and understand the meaning behind user queries 
• Vector embeddings for AI retrieval to identify results from the search index that match query intent 
• Learn-to-rank AI algorithms (such as reinforcement learning) to order results from most to least important. Over time, an AI search engine will refine and improve results automatically 

An important point with AI search is that it actually works better with traditional full-text keyword search. The combination, sometimes called hybrid search, offers the best of both worlds: AI search for managing concept-style queries, and keyword search for precision matches. 

What are the pros and cons of each solution?

Search as a service API

With an API-first approach, businesses, such as Algolia, can provide AI search instantly. In this case, the SaaS vendor manages the backend index, infrastructure, and application, and provides customers with APIs for deploying search onto a website or as an app. In the case of Algolia, the service includes end-to-end AI combined with full-text keyword search so it works well for either exact matching searches or broader query types. 

Ecommerce solutions 

With a focus primarily on ecommerce applications, these vendors may apply a vector-based search to just individual parts of the search process (thereby not being true end-to-end AI solutions) in order to minimize zero results pages. But this capability is expensive and engaged sparingly, e.g., only for minimizing null results, slowing the experience. These providers might be using some AI but typically not upfront for retrieval, e.g. they have an add-on for reranking. AI retrieval matters because if you use AI only to rank results, the reranking is sub-optimal and affects downstream conversions. 

Build with open source solutions 

Solutions built with open source often appeal to large engineering-driven organizations who prefer DIY projects. In this case, to get AI parity with a SaaS solution, customers would need to combine both an open source search project with AI search projects to cobble together a solution. Open source is free, but can be very expensive to host, manage, and update. Moreover, it requires technical machine learning expertise to thoroughly understand and optimize results. Additionally, it will require unremitting maintenance by an in-house dev team.

What about using ChatGPT for search?

As you are considering AI search solutions in your discovery process, the idea of using generative AI solutions that utilize Large Language Models (LLMs) such as GhatGPT may come up for discussion. It is important to understand that Generative AI solutions like these aren’t delivering search results; they’re writing new answers based on input, and they suffer bias based on however they’re trained. This can be complementary to search, but it’s not the same as search. Results from generative AI can vary greatly from query to query, and for now at least, they can’t be customized as easily as more traditional search. Furthermore, ChatGPT cannot provide critical search functionality like recommendations. You might also check out our blog about why ChatGPT won’t replace search engines anytime soon. 

The bottom line

Download our free search buyer’s guide checklist to help guide your planning for launching AI search on your site or app.