With the amount of competition in the SaaS space - it can be rather daunting to find that one killer idea that helps you stand out.
Furthermore, CPC advertising is crazy-expensive and almost impossible to get right as a new founder. What you need is a solid idea and a good product- something that has little competition and is relatively easy to sell. Since AI is such a hot topic, now is the perfect time to ride this hype wave.
Before you go and build something, I strongly suggest doing some market research first to see if there is a demand for what you are building. I discussed how to go about this in my earlier article here.
In this article, I will run through some of my SaaS ideas that you can "steal" and make your own. I have not done any market research therefore there is no guarantee that these are viable products but hopefully, they can serve as a starting point.
Let's talk about AI first
We tend to use "AI" loosely, however, machine learning is just a subset of AI. Anyway, the point is that AI has been a hot topic since 2023.
In my opinion, it's mostly just hype but that does not mean that there are no opportunities in this space.
There are plenty of practical use cases for machine learning, such as:
Classification. From binary (e.g. spam filtering) to complex multi-class classification, this is one of the most common use cases.
Summarization and QA. Think ChatGPT and training chatbots on large PDFs or books and then being able to ask the bot questions on the subject material.
Recommendations engine. Think Netflix or YouTube. They customize your video feed according to your preferences.
Idea 1: Build a smart personalized offers model
This concept will work well for e-commerce merchants and may also be suitable for other industries as well.
You would feed the model with shopping data and predict future sale potentials.
Example: The model can analyze product types being bought by a customer and determine what other products go well with the purchased product.
Thereafter, send the customer a customized coupon to entice them to make additional purchases.
Idea 2: Product tagger
E-commerce merchants commonly submit their products to Google Shopping and other shopping portals. Thus, taxonomy and attribute requirements vary for each portal.
You can train a model to recognize product attributes, tags, and taxonomies. Then provide an API to access this information. The API would return a list of attributes, taxonomies, and tags based on the product description, image, and title.
This will then allow merchants to categorize, tag, and syndicate their products more efficiently.
Idea 3: Super CRUD
CRUD is annoying if you have been in the web dev industry for some time now yet it's a vital component of any web application.
There are many such generators around, however, they tend to introduce further abstraction layers on top of the web framework you are already using. This usually makes it difficult to customize.
Take for example Laravel; fine-tune a model so that it generates standard Laravel code but is aware of the entity and its relationships.
If you prompt the model with something like the following, it should then generate all the models, controllers, and views using the UI of the project.
php artisan crud:build Products
I am aware of Copilot and other such tools that already do something similar, however, you can offer this model as a managed service and hone in on specific niches where a general LLM doesn't do well.
For companies/individuals who are privacy-conscious, they can just pay for a maintenance contract where you provide support only but the model is hosted on their servers.
Idea 3: Article optimizer
This involves fine-tuning an LLM like Mistral 7B to scan through various social media posts, directory listings, Google search results, and so forth.
There are already many of these sorts of products on the market but there is still plenty of opportunity in niches as most of these are general in scope.
A writer can then feed their article to this model and get feedback on tips to improve the title and article so that it can perform better on various platforms without losing its original meaning.
Idea 4: CV optimizer
This one is slightly more difficult to do since you may need to partner with recruitment agencies to get sufficient data.
The goal would be to scan a person's CV and determine, based on the training data, a quality score and tips to improve.
You can then add to your dataset all kinds of metrics like salary level, which skills are more in demand based on region, and so forth.
Furthermore, you can generate reports that provide useful refactoring advice. An example, "If you structure your bio info this way... and List these skills ... you will increase your chances of getting a job by 30%".
Idea 5: Smart recruiter
This might be a privacy infringement but then again could make both recruiter's and developer's lives easier - so maybe not a bad thing.
LinkedIn is a powerful source to find great candidates. You could build an ML scraper to scan through profiles (I am not sure if LinkedIn allows this) on LinkedIn and look at their job position history.
Does this person change jobs every 2 years? What kind of jobs do they like?
For example, based on this information, if the user changes jobs on average every 2 years and they've been at their current job for 18 months, perhaps they would be on the market again.
When matches are found, simply alert recruiters with the necessary information.
Resources to build your AI SaaS product
Great! Now you have some ideas on what to build, but how do you get started?
To help you along, I have curated some resources below:
1> Model: You are going to need some sort of model to use as a base. At the time of this article, the best open-source model around is Mistral:
To run mistral without a GPU, you can learn more here:
Run Mistral on LLama2
2> SaaS boilerplate: To build any kind of SaaS, you are going to need billing, teams, user accounts, and so forth. There are plenty of great paid and free SaaS boilerplates out there. Here are a few you can try:
- Wave - Open source and based on Laravel.
- Bullet Train - Open source and based on Ruby on Rails.
- SaaS Pegasus - paid boilerplate, based on Python.
- Gravity - paid boilerplate, based on Node.
3> Tool to help analyze search traffic so you can target and build content around keywords: Wordstream
Conclusion
Building any SaaS business is hard and market saturation with heavy competition is always a huge problem. However, since the AI/ML hype is still very new, there is a lot of interest in this space.
Thus, this provides a great purchase-happy boost to help enthuse potential customers.
Be careful though - there is plenty of madness going around where companies just throw in OpenAI and call it AI. Cheap gimmicks are not going to work - you need to add value and by doing so, customers will appreciate and use your products.
Top comments (0)