First Impressions on the ChatGPT Marketplace from a SaaS Perspective
ChatGPT marketplace as the next SaaS distribution platform?
Last week OpenAI finally unveiled its GPT marketplace which enables users to explore and find GPTs created by third party developers. I tested a couple of them and here are my current thoughts on their potential for the SaaS industry.
Under the Hood
GPTs are basically ChatGPT instances with enhanced capabilities and predefined behaviors that anyone can create.
Creating a GPT involves customizing the standard ChatGPT bot in two key ways: first, behavior modification and second, enhancement with unique actions.
Behavioral customization is achieved through prompt engineering, where instructions are written in natural language to define the bot's interactions with users and how it should behave.
These GPTs can also be enhanced to perform tasks beyond the scope of a regular ChatGPT, such as accessing third-party APIs or private databases.
I read their documentation and created two/three GPTs myself. And I have to admit that creating a GPT is an intriguing blend of traditional coding and the more esoteric art of prompt engineering.
For instance, in the OpenAI documentation you have an example of behavior specification where you can find sentences such as “YOU MUST GO THROUGH ALL OF THESE STEPS IN ORDER. DO NOT SKIP ANY STEPS.” or “YOU MUST CALL update_behavior on gizmo_editor_tool with parameters "context",” that basically scream at the bot so that he really does what they want. A bit like when my wife gives me instructions that she wants to be sure that I follow :-).
These instructions can sometimes be funny to read.
My Experience from Testing GPTs
I experimented with approximately 10-15 GPTs, both consumer and B2B.
Consumer GPTs included use cases such as dog training advice, MBTI profiling, and trail recommendations for hiking.
B2B GPTs were more focused on professional tasks like presentation creation, coding assistance, and PDF document analysis.
In terms of quality of results, I noticed that GPTs delivered better results when the tasks conducted were precise. For instance for clearly defined tasks such as asking for hiking trail recommendations or identifying my MBTI type, the GPTs I tested were quite good. But when the tasks were more complex, such as making a presentation which requires more input from the user and more creative thinking in the solution, they struggled more. Or at least they didn’t struggle to give me a result, but the result was not that great.
In terms of the UI/UX experience I have to admit that doing everything through a text based interface can be a bit tedious and boring. So yes, I got some good results for the trail recommendations, but to receive the results in the form of a bullet point list is probably less attractive (and practical) than going on a dedicated mobile app where you can see everything on a map with attractive pictures.
Same with the GPT bot that helps you generate code based on what you need. It’s cool, but when you compare the UI/UX with an IDE that integrates natively GenAI, you feel a real difference.
It’s still very early, but at the moment most GPTs I tested and checked looks more like “nice-to-have” than “must-haves”.
This is why in terms of my personal usage I see more ChatGPT as a Google on steroid than as a B2B tool at the moment. It has already replaced many of the interactions that I had through Google (search, translation etc…). And in this perspective, I see the scenario where ChatGPT is kind of a “super app”, but more in a consumer context than in a B2B context. But I might be wrong.
Recurring vs. Punctual Tasks:
These tests lead me to think that at the moment GPTs are better suited for occasional tasks rather than regular, routine activities for which dedicated services are probably still better.
In my case GPTs could be ideal for punctual needs like hike planning or MBTI assessments, or when I need GenAI to generate code for a script to import and transform data from one service to another for example. Then I would probably use a GPT for it. But if my job is to develop on a daily basis, chances are that I would prefer GenAI features directly integrated in my IDE.
Consequences for SaaS Companies:
This is why after less than a week of tests (so very little time), the biggest opportunity that I see on this marketplace for SaaS businesses is lead generation.
Free GPTs could serve as a way to attract and deliver value to users, potentially converting them into paying customers down the line.
B2B GPTs are in a way almost freemium stand alone products useful to attract users and convert them into customers later. It’s especially true that there is a distribution aspect baked in, as SaaS companies can target and reach millions of ChatGPT paid users through the marketplace interface and ranking.
The other play I see for B2B SaaS is to offer a GPT for their users who really want an integration on ChatGPT and are ready to pay for it. There is no native monetization on the GPT marketplace yet, but you can ask the user to be logged in to access third party APIs. So you can ask a ChatGPT user to be a customer of an external service or source of data before he can use a specific GPT.
At the moment I see more this marketplace as a “plugin” type of marketplace rather than a marketplace where you can build a real B2B software.
Going Where the Puck is Going to Be:
Obviously my take will be outdated in a couple of weeks only as OpenAI will probably greatly enhance its service. So I don’t expect the GPT marketplace to stay as it is today.
For example in terms of UI/UX they could provide GPT creators with a greater number of visual components such as maps, tables, graphics, a PDF reader or dashboards that could be integrated directly in the ChatGPT interface (a bit like the native code editor that you can already see).
I am not sure that they will go down that path, but I would love them to let developers create ChatGPT clients. Like TweetDeck was for Twitter. I’m sure we could see very interesting takes from third party client developers.
Also as Sam Altman mentioned, we should already think about the next iteration of ChatGPT (GPT 5) and the fact that many of the current problems and limitations will be solved in the near future.
So even in terms of results coming from these GPTs, I’m sure that we will see improvements. The presentation that was generated for me through GPT 4 was not that great, but it could potentially become great with GPT 5.
That being said there is also a chance that this GPT store gets shutdown, similar to what happened to ChatGPT plugins.