Subscribe
Sign in
Product Recommendations AI — Personalized software recommendations based on your stack
Personalized Product Recommendations AI answers the question "What software product should I use?
A machine learning solution that can be used at scale to make software recommendations tailored to each user. Powering the models is data on nearly 33,000 products and over 375,000 companies that use and recommend them.
29
Replies
@ggiaco Thanks for creating a useful product. Looks like it works best for certain roles than others. For example, I don't see all the products I use when I choose the Developer role. Also it looks like a database of products that go together and I don't see why you need ML for this usecase.
@sasajeev Thanks Sajeev! Great to see that you're finding it helpful. You should be able to add any product in our DB (33,000 strong now). Is there one we're missing? Let me know and we'll get it added. Or if you prefer you can submit it here: https://siftery.com/submit-product
The ML comes into play in the Advanced Recommendations. At a high level, the model gets better on its own by reacting users marking recommendations as Useful/Not Useful.
@ggiaco good job guys, I really like how you're leveraging the core product to create other adjacent value for users, and at the same time continuing to build the core data set that's powering Siftery. Great virtuous cycle 👏🏽
@_pulkitagrawal Thanks Pulkit! That's exactly how we're thinking about it - "How can we create value first?"
Any ideas for what you'd like to see (for Recos AI or more generally)?
@ggiaco Am I doing it right? I got an account with Siftery, spent some time and added 84 products that we use in our company 
The banner suggests I should get recommendations. Cool! I'm in, that's why I probably registered in the first place.
But when I select a role, let's say it's Support and this is what I got there:
I'm NOT seeing my items on the list and prompted to add products AGAIN.
What's the catch? I've got 84 categorised products in my profile, why adding them again?
The banner suggests I should get recommendations. Cool! I'm in, that's why I probably registered in the first place.
But when I select a role, let's say it's Support and this is what I got there:
I'm NOT seeing my items on the list and prompted to add products AGAIN.
What's the catch? I've got 84 categorised products in my profile, why adding them again?
I gotta be honest, these recommendations are literally the opposite of what they should be: 
Also, this is misleading - when it suggested JIRA as a backtrace replacement, I was surprised that JIRA had added crash tracking and analytics until I realized the engine was just arbitrarily grouping bug-tracking-related tools. This is **bad** because I already field requests from management to use tools that are not good fits. Imagine how much worse my day could be if they now had "AI recommendations" for tools that don't even do the job they think they do.
If the goal is to use incumbent products maybe this would be useful for companies trying to estimate size, but without quantification it's not even that useful in that sense. The "AI" just appears to show me the most used product. I would posit that it's a flawed thesis to correlate "most used" and "best". There are also some clear misses (suggests modernizr as a replacement for react native, which doesn't even come close to making sense).
It looks like a lot of work was put in to the product, which is unfortunate given that the underlying models are fundamentally wrong, and ultimately the tool is most useful as ammunition for management that is willfully ignorant or resistance to change.
Also, this is misleading - when it suggested JIRA as a backtrace replacement, I was surprised that JIRA had added crash tracking and analytics until I realized the engine was just arbitrarily grouping bug-tracking-related tools. This is **bad** because I already field requests from management to use tools that are not good fits. Imagine how much worse my day could be if they now had "AI recommendations" for tools that don't even do the job they think they do.
If the goal is to use incumbent products maybe this would be useful for companies trying to estimate size, but without quantification it's not even that useful in that sense. The "AI" just appears to show me the most used product. I would posit that it's a flawed thesis to correlate "most used" and "best". There are also some clear misses (suggests modernizr as a replacement for react native, which doesn't even come close to making sense).
It looks like a lot of work was put in to the product, which is unfortunate given that the underlying models are fundamentally wrong, and ultimately the tool is most useful as ammunition for management that is willfully ignorant or resistance to change.@jesseditson Thanks. Harsh feedback is often the most helpful, but I also hope you'll give this its fair shake.
From your screenshot, it looks like you're going through the Basic recommendations. The Advanced Recommendations flow is going to be much better - literally 4x more models at work. We cannot do much to personalize recommendations for your company if we don't know to which company you're affiliated.
In the Advanced engine you'll also be able to mark recommendations as "Not Helpful", which is a super valuable signal for us. Example of that button below:
Based on the feedback we're seeing here and in the data, it looks like our Developer role recommendations need the most work, while others are getting better-than-expected feedback. I also want to set reasonable expectations: We see this engine as a starting point in the product discovery journey, not the end. Given that there are so many products being created (36k in our database now, more to come soon), it's becoming increasingly difficult for most people to be even aware of everything that's out there.
While most discovery resources are still focused on maintaining an increasingly-long list of products, this is our effort to help winnow that list down for most users in just a couple of minutes.
Anyway, I do appreciate you taking the time to share these thoughts!!
Based on the feedback we're seeing here and in the data, it looks like our Developer role recommendations need the most work, while others are getting better-than-expected feedback. I also want to set reasonable expectations: We see this engine as a starting point in the product discovery journey, not the end. Given that there are so many products being created (36k in our database now, more to come soon), it's becoming increasingly difficult for most people to be even aware of everything that's out there.
While most discovery resources are still focused on maintaining an increasingly-long list of products, this is our effort to help winnow that list down for most users in just a couple of minutes.
Anyway, I do appreciate you taking the time to share these thoughts!!
Does it work mainly for front end tools?
How does it pick up my current stack?
Is it only one time recommendations or on going?
Can I request optimization for cost, performance, or scalability?
@prasanna_says All good questions!
For our model to recommend a product, it first needs to be in our DB (we have about 33,000 now and growing) and then we must have some data on usage (i.e. which companies use or have used a product) and sentiment (i.e. do its users recommend it or not). We have more data on FE tools, but BE is also in the house.
You can come back and get recommendations anytime! We're working on proactively pushing recs (set it and forget it!), but that's still upcoming and will be guided by the feedback we get.
For optimization on cost, performance, and scalability - we don't address them directly, but if you go through the Advanced flow you will see that recommendations are often adjusted for company size, industry, geo, etc. (in the background we pull in this information for your company/domain from public sources). Therefore, you might get a recommendation because a particular product is more popular with companies of your size (e.g. an HR solution for the Enterprise). Indirectly, this can account for those variables.
This is awesome! I've been an early user and it's great to see this new seamless way to get recommendations. One of the companies I contract with was able to make a better decision on their marketing software thanks to Siftery!
@kkkosariya I feel like soon that's all that will happen. Or maybe software will just pick what software to use and automatically start using it..
It's much funnier to discover new products and tools with Siftery rather than other websites 😀
@jajoosam Hey Samarth, yeah - looks like this happened because one of our models is based on popular alternatives within a category. In this case both are grouped as "Languages". We'd solve this by getting more granular categorization (we have roughly 750 categories at this point).
Did you find any of the other recommendations useful? If you bookmark or mark any of the recommendations as "Not Helpful" in the app, the algorithm uses that data to make better recommendations for the next user.
Clara