Opportunities in Building AI Platforms

Okay. As soon as I clicked on go live, it said, oops, connection lost. And, but now it says that we are live. So I don't know what happened, but let's hope that we have our viewers. I guess we are live somewhere. that's, that's okay. Awesome. We can, we can begin. Hello and welcome to Momentum Office Hours. My name is Yash and I'm joined by my co-founders Jay and Kaushik to discuss topic of the week.

which is opportunities in building AI platforms. Our goal with these sessions is to provide you with actionable insights and practical strategies that you can apply to your own businesses. And throughout the session, we encourage you to engage with us by asking questions and sharing your thoughts. This is a fantastic opportunity to learn from each other and to drive your digital initiatives forward. Let's get started. Jai, Kaushik, how are we doing today? Doing great. Good. Nice. Good. I think we're for this.

Hectic start of the week, but going great now. Yeah, Mondays are generally hectic. And as hectic as they are, they are a good sign that we are creating some value out there in the world, that we are required in two places at once, if not more. So it is always a good place to be in. But coming to opportunities in building AI products, right? So one of the things that the three of us

wanted to do was essentially just take all of our viewers and listeners through some of the, like how should you approach building products in the age of AI? And while we can have a lot of conversations and a lot of back and forth and we can answer your questions as well, but we thought that we, of everything that we've learned, we'll sort of try and put together a deck which we could go through. So it just offers some amount of structure to the,

to the conversation that we going to have. So I just add that deck to the view. hope it's visible to everyone. you're able to see the deck? OK, awesome. So we're basically going to talk about building products in the age of AI. And before we begin, though, when we go to the next slide. So before we begin, though, one of the things, we can go to the next slide,

The goal of the conversation is essentially to look at what other people are building. A lot of the times we can sort of find ourselves confined in the bubble of our own LinkedIn following or some influencers on YouTube or Instagram and think that and either overestimate the capability of AI or underestimate. But just looking at examples of things that other people are building is tremendously helpful and valuable to sort of

set our benchmarks and see what things are possible. And the second is that this is not an answer to all the questions. This essentially just helps us identify what are the questions that we should ask. And so think of this conversation as just a starting point for going in the direction of off-breeding AI products. And so let's talk about AI. We'll go to the first piece, is all the terminologies. We'll go to the next slide, Prashan. So all the terminologies that you generally find yourself.

hearing, is where we talk about, so, you know, autopilot and copilot, like, what's the core difference, verticalizing LLMs, what does it mean when people say narrow AI, what are AI wrappers, you know, remembering that when we are building a new product, other people also have AI and just AI doesn't get funding, distribution has gotten even more difficult. And the last piece, which you may not be able to see, but it essentially says,

solve for accurate data, and then AI can solve for another. So autopilot and copilot will start first. So the difference between the two of them is essentially when if a person is required as a part of accomplishing a task, that is largely called as a copilot, which is where AI is helping the person accomplish the task, better accomplish the task, faster accomplish the task with higher quality. If something like that is happening, then it is called a copilot, which is where AI is

sort of an assistive, isn't more of an assistive rule. Autopilot is where a person may not be required, right? So where you sort of set it and forget it. And so once you've set the context and once you've tested it out, is the AI engine or the AI agent is ready to sort of go out there in the real world and then execute the tasks. The second bigger opportunity is in verticalizing large language models.

LLMs, all of us are aware of this. There's an LLM by OpenAI, which is Chad GPT, largely. There's an LLM by Meta, which is called Lama. There's an LLM by Perplexity, and so on and so on. So there are a few LLMs that exist out there in the real world, which are trained on the data that is available on the internet. Certain LLMs are better at research. Other LLMs are better at generative AI-related use cases.

But there's a huge opportunity that exists in configuring LLMs for a particular vertical. So for a particular legal, a particular vertical like health care or financial services and so on and so forth. Which is where the LLM has significantly more context about a particular industry or a particular use case such that it is able to generate significantly better outputs than your general purpose LLMs. So a good way to think about it is this is the exact difference between a resident doctor and

and then MD of surgery. So it becomes more and more specific and verticalized and better as and when it learns very specifically about the particular thing. The other piece is narrow AI, which is building, it could include building AI wrappers, it could include building rag model, whatever it is that you're building, but building only for a very specific use case. So building not even for an industry, but for a department within an industry.

So as an example, building AI tools or platforms that air traffic controllers can use, which is for a particular department within a particular industry and extremely specific. And that could be a mixture of drag, it could be a wrapper, could be configuring or fine tuning in LLM, whatever the case may be. But building absolutely for a narrow, narrow use case, which is where you'd be able to charge a premium and have a very few set of customers.

The fourth one are AI wrappers. So wrappers are things that are built on top of largely LLMs. So most of the software that we see out there that are talking about AI piece within the platforms are wrappers. And these are significantly easier to build. And you will see more and more of these being built as well. And they also carry the...

downside of not giving you a huge advantage in terms of the technology. So because if you can build a wrapper, I also have access to the same large language models and I can build a wrapper as well. the world will be changed by, like largely the world will be changed by wrappers that are being built. Most of the software that you will use are going to be wrappers on top of large language models. A good example of a wrapper that you would see is

is image generation as an example within Canva is a great example of a wrapper or the ability to write captions within Hootsuite, which is a social media scheduling tool, or Buffer, which is another social media scheduling tool. All of those are wrappers that are built on other LLMs that exist. Some fundamentals to remember when you're building in the space of AI is that other people also have AI. And just having AI as a part of your deck, doesn't get you

get you equity. is sort of born out of a lot of people that I've met who have shared with me that hey, you know, they are an AI company and still are not able to get funding, even though AI space is extremely hard. And so it's important to remember that AI now, if you're building in the tech space, is a necessary condition, but not a sufficient condition. And still the moat stays as distribution, right? So the amount of customers that you are able to get, the amount of people who are using your product, that the

capturing of the market that you do is getting even more difficult and it still stays as the mode in this space. We'll move to the next one, which is where I want to make sure that there are certain opportunities. This is not the whole world of all AI opportunities that exist out there. I want to make it extremely clear that this is not an exhaustive list. Here are some things which are not included, even though some things are not exhaustive. So there are a lot of things that are not included.

So please don't think that this is sort of a say all, end all list of everything that's possible in the space of AI if you want to start building. But certain things that I don't know anything about are hardware, like AI opportunities in hardware, AI opportunities in IoT, AI opportunities in blockchain and Web3, and AI opportunities in AR and VR. So these are some things that we don't understand well enough. don't know what are the opportunities over there. These are just largely AI opportunities in the software space or in the SaaS.

space that we are talking about. So please don't think of this list as an exhaustive list. We'll move to the next one. And so one of the things that like two large approaches. So one is that we want to look at how can AI be used in service delivery versus product. So in service delivery, it's fairly simple, which is how we also use it at Momentum, which is our code quality has gotten significantly better. We are able to automate some amount of QA and testing. We are able to write engaging posts.

On social media, if you're a digital agency, can use that. We are able to do quicker design iterations. So largely, the turnaround times have gotten significantly smaller. And at the same time, our quality of output has become significantly better. So that's broadly an approach of towards using AI as a part of service delivery. So if you're an agency founder who's digital agency or an IT technology services or technology consultation or research and development agency or market.

Sorry, just one second.

or as an example, like market entry consultation. Then service delivery, you'd be able to use AI as a part of your fulfillment of projects that you're able to close. And then approach of integrating, implementing, or building AI capabilities as a part of a product or just native AI capabilities. That is the AI in products. And we'll talk about those in the following slides. So one of the things, and Kaushik, we'll move to the next slide. So one of the things that we've done is that we've sort of

broadly categorized as to what AI can do. And then we've listed down some examples of how other companies have already accomplished them to look at it as an inspiration of what can be done. So one of the very important piece is building native AI experiences. So most large companies did not start thinking about what AI is doing today. So if you think of companies like Notion, ClickUp, Office 365, Google Workspace, all of them.

have reacted to what AI has done. And so they have retrospectively tried to fit AI into their own user journeys and their own workflows and things like that. And so there is an extremely big opportunity in building systems of records, systems of engagement, or systems of decision making, natively with AI. So what if you forget how the software was built over the last 10, 12 years, and you just have a fundamental new and native AI

related approach to building your product. And so this is going to require a lot of innovative thinking because all of us are surrounded by software and we take inspiration from things that surround us as well. But this is another big opportunity to think about. So next, if we could go to, is when we sort of look at some of the use cases. So one of the first use cases that we will look at is, Kaushik, if we can go to the next slide, is what are the capabilities that AI has in terms of seeing things?

So the first is AI can see things. So it can see images and videos and convert images and videos into text. And so these are all the examples of what AI can do. as an example, iOS extracts phone numbers from images. ATSs, Applicant Tracking Systems, can scan resume and then do requirement matching. So it can look at like a thousand resumes, figure out which resumes are closest to the requirement that the organization have.

matching can happen. Facebook detects alarming behavior during live streams. if is streaming violence or violent content or if someone starts to stream sexual content, Facebook uses AI to see the content and then it detects alarming behavior and then takes the next steps. So it can convert images or videos into text. This is sort of an umbrella.

capability, one of the umbrella capability that AI has. And then here's how companies are using it. Next one. And so AI can also uncover insights. So since LLMs are trained on all the data that exists on the internet, it knows market, it knows competition, it knows customers. So as an example, one of our customers, Alfana, who's also on our podcast, by the way, it converts one piece of content into 32 pieces of content. So you put in.

like a podcast video or an episode, it converts that into transcripts, into subtitles, into social media posts, into short videos, into a blog post, into all of these pieces. So it can uncover insights from the world, from the internet, or from the content that you have offered, and then it can convert it into all the outputs that you may want to look for. So insurance companies use it to evaluate risk. Banks calculate credit scores based on history.

So it can look at large amount of data and then it can uncover insights. can recognize patterns and it can help you make better decisions or it can help you generate some amount of content as well. The next one is basically AI can continue to uncover insights. There are a lot of other use cases also. So Power BI is an example, uses this piece for natural query. So instead of having a data analyst sort of create pivot tables and convert and connect and do all of those.

jugglery within Power BI. can just very like with a natural prompt, I'm able to sort of create my charts and graphs and get answers that I'm looking for. iPhone automatically triggers focus modes depending on what my pattern is and what I'm doing right now. ArcGIS analysis geographical data to identify trends, predict market movements and showcase new and upcoming opportunities in the real estate space. So how are people buying houses? How are people buying?

retail spaces, how are people buying commercial spaces or businesses buying commercial spaces. This is that it looks at all of that data, all of that trends and then gives insights also on top of it. Can you move to the next one? The other thing that AI can do is it can, this is generative AI, right? So it can create and generate as well. So it can create visual art, videos, voice and even music. This is the fundamental use case that as a consumer or a customer we see often.

And so there are n number of examples. So you can look at Gemini, you can look at GitHub Copilot to write code, Gemini to create lyrics for a song, ChatGP to write a social media post or to write an email, Mid Journey creates digital artwork. it can create and generate. Previously we looked at how it can generate, how it can convert videos and images into text. This is the other way around, which is where you can put in a prompt.

And then it will use its own intelligence to create visual art, videos, voice, and even music. Next.

And so AI can also hear things. So it can understand speech and recognize voice. So there are companies who are, so Alexa is the biggest, best example of this, but there are, you know, N number of language tutors out there which understand what you are seeing and then tell you whether you've, how well you've done on the, on the next, like the second or the third language that you're running. This is like, you know, one of the customers that we have is also implementing this into that.

recollection or dead collection as well. It can make calls, can say things, and then it can hear the responses and then basis that it can make decisions and then respond to that as well. There are a host of use cases within the fact that another umbrella pieces is that AI can hear things as well. Go to the next one.

And then AI can undertake basic tasks, right? So this is sort of moving from a co-pilot piece to an autopilot piece as well. So Tesla offers a great driving experience based on roads. Power Automate creates automations based on historical data. So it can essentially, LG's smart fridges create groceries for you. And then in some not too distant future, it will also be able to order from Amazon. So these are all the agentic things and basic, absolutely basic tasks.

that AI can do today as well. can help customers perform basic office tasks and assist in daily life as well. This is also what they call as an autopilot or an AI agent as well. The next one.

Some things, again, to keep in mind is that, as I said before, distribution is even more difficult than it used to be before. AI safety, you need to invest your time and energy into making sure that the AI that you're implementing is safe for the users. And so what essentially becomes your motive is how deeply do you understand the use case to implement AI and make it better, faster, cheaper, higher quality? So that's basically...

going to be one of the more that you have when you're building, especially zero to one use cases in AI. Very, very clear definition of who your ideal customers are. And please remember that building any piece of software is a sunk cost. And so you want to try replacing artificial intelligence with natural intelligence first. So just do things yourself. And then once you see that you're buckling under the pressure and whatever it is that you're doing is extremely high in demand, you want to then start replacing it with AI.

Just do things yourself first, get some demand in. Once there is demand, then you start to build software. And then you make the software better that you sort of become replaceable from fulfilling the tasks. And that's a great way to build an AI product or an AI application. And I think that's pretty much it, Kaushik. So just wanted to bring this in front of all of you guys. Hope this was helpful. We can discuss more on AI piece as well. But yeah, so that's.

sort of some thoughts in terms of use cases and some thoughts of what are the categories and umbrella items that exist as far as opportunities in building AI products are concerned. I hope this was helpful. What did you think, Kaushik Jai? Did it give you more questions than answers? Did it give you answers that now you are absolutely clear? What are your thoughts? Answers that would lead to more questions. Yeah, the goal of putting this together was

that all of us have been reading a lot on what is happening in the AI space since a while. this deck essentially just sort of categorizes. I know that we're missing a lot of things within this deck also. But it at least starts the process of categorizing the possible use cases as well. Yeah. I think one interesting thing from this deck as a takeaway that anyone who is watching this could take that is that

important part of deciding your implementation, AI implementation is about scoping what you want to implement. So you have everyone has a job and each job has a task. Not just one task, many tasks. I think the explanation that you just gave would help them identify which task has the needs AI capability or has AI capability, which task doesn't need AI capability.

Because that differentiation is something that we are seeing widespread, people not trying to understand or even make a decision about. Not everything needs to have an AI capability within the task set that they have. So I think that's a takeaway that probably I would have. This is actually an old problem in building zero to one products, right? least for first time founders, one of the challenges that they have is that let's do it because we can.

And when we start to build things just because we can, it essentially ends up being a solution that is looking for a problem instead of looking for a problem first and then solving that. So that's one of the bigger challenges, sort of a hurdle. I went through that and it is extremely painful, building a solution that then sort of goes out there looking for a problem, which is not an ideal place to be.

So just like in traditional SaaS journey, was like a founder just starts with a product, building a product to solve one problem, right? And then eventually they get so much into it that they enter into one more feature trap, just like that. Same way in AI, it's like a lot of capabilities are there and then people forget on what exit problem they're trying to solve. And then again, it focuses on, I they focus more on giving.

confused about what sort of offering to utilize from. So obviously, the scoping needs to be well-defined and maybe then the right step should be gone. yes, Yeah, sorry. I'll just to add on that. So there's a favorite meme that I have on this, which is if you've seen The Office, if people who are watching this, if they've seen The Office, which is a really good series. So there's a meme on Michael Scott just sitting on a bench.

And then it just says on the top that released one feature, now I should get more customers. And this is Michael Scott just waiting on the bench looking at bunch of pigeons and cats and just thinking that I now release the feature, now I'm going to get a lot of customers right away. And so it never happens like that. So here's a good hack that I found. Before, when I was going through this,

Whenever someone would ask me or I would ask myself as to why don't I have enough customers, I would just respond to myself saying, because I don't have a particular feature or because I don't have a particular module or because I don't have a particular workflow within the product. And my life changed when I forced myself to answer that question without naming a feature or a module or a workflow. So why don't I have customers? If I'm not allowed to.

talk about my product, feature, module, workflow, what would be my answer? Is my positioning not good enough? Is my ideal customer's definition is not narrow enough? Is my outreach methods not good enough? Is my research not good enough? Is the problem that I'm solving not big enough? And so on and so forth. So once I forced myself that the answer cannot be the product,

feature, module or journey, then what are the other possible answers to this? And so since then I've had like, I'm not saying that it has made things better for the business, but it has definitely made sure that we're investing in the right directions. So we're investing on honing our skills at marketing, becoming better salespeople, trying new markets, defining the problem better and so on and so forth. But sorry Jay, you were saying something.

Yeah, no, no, definitely that does make sense. I was just trying to understand with respect to SaaS products which are already there. I mean, we are talking about building AI products, so it's more about opportunities we talked about, which are starting from scratch. But the ones which are already there doing great and still trying to figure out on what AI component they should have. Do you have any process or any suggestion on which?

where they can get started from for the products which are not having any exact AI feature yet, but product is being used. It could be in any segment. So is there a way to approach there as well? for sure. So here's a good way to think about it. Look at your most active customers, talk to them and ask them what other AI tools are they using in conjunction with your product. So I'll give you an example. So an example of

So very simple example of the way that I schedule these streams. So I use Restream to schedule these streams. And one of the ways in which I use an AI product, which is separate to Restream, is to write captions and it is to write description for the event or for the live that we are doing. Now, if product manager at Restream would ask me, what are other AI tools that I use in conjunction? And then, sorry. And then we also use a platform called Riverside.

to convert the conversation that is happening over here today into smaller videos. And then we use another platform. I don't remember our social media manager uses that. But we use another platform to convert this live into a blog post. So now if Restream's product manager reaches out to me and asks me that you've been using Restream's live video streaming capability since the last one year or so, and you've been consistently live week after week.

What are other AI tools that you use in conjunction with streaming just for that particular use case? I would give them the list of these three products. Now, if that product manager talks to me and then talks to at least 100 other people like me, they will be able to make a list of five or seven things that are common amongst the 80 % of the customers who are using it, who are using in conjunction with their core use case.

And then starting with the first basic and fundamental use case is the best way to go about it. So, know, RESTream could start with allowing me to auto-generate captions or auto-generate social media descriptions based on the topic and the thumbnail that I have. And then slowly and gradually, it could allow me to generate transcripts, and then it would allow me to generate blog posts, and then it would allow me to generate short videos, it would allow me to generate and so on and so forth, right?

And I would happily overpay because I'm currently paying for four tools for one stream. So I would happily pay more to restream. Now think of this in other platforms as well. So let's say if I'm building a social media scheduler or if I'm building a task or a project management system or if I'm building a sales CRM or a marketing CRM, like in martech space in general.

If I'm developing a website, whatever the case may be, in and around the core use case that my product has for my active customers, what are the additional periphery AI use cases that exist? I may be using Gemini, I may be using Perplexity, I may be using ChatGPT, I may be using different sort of tools and platforms, but they still become part of the same workflow of not being able to go live to being able to go live.

And so that's a great way to go about it. Unfortunately, the answer to this question as much as I want it to be, unfortunately, the answer to this question does not lie in research and technology. Sorry, It does not lie in technology and technology related research. It relies in research and customer related research. So that's where every journey should begin. But I think that's...

It also brings us to the end of the conversation for today. We tried a new format. And if you want us to try other new formats as well, so this is a new format that we did this time, which is where we took you through the deck. If you want us to try other new formats, please encourage us by subscribing to wherever you're watching this, whether it is YouTube or if it's LinkedIn, you can like, can share, can comment, whatever you think is OK. Do consider doing that. And we'll try.

newer, better formats for these office hours streams as well. Until then, thank you for joining in and hope this was valuable. We'll see you next time. Bye-bye.