How to get to product market fit in AI SaaS, crawl walk run guide for Indian founders

Going from Zero to Product Market Fit is like getting out of a Bermuda Triangle. For every direction that works out, an equal and opposite direction works too.

Rare few get to product market fit on their first try, the majority of folks struggle. It is shocking how few get to product market fit. Only 15 out of 100 SaaS companies get to a million dollars in revenue. When you have it, it is an awesome feeling; everything is going smoothly for you. It also feels scary and out of control. You push one unit, and ten units of progress happen. But when you don’t have it, you push ten units of effort, and not even a unit of progress happens. It feels like a Sisyphean punishment. Moving a boulder up the mountain.

The most actionable definition that I have found is “When you have conviction about how you can find a lot of customers, convince them to adopt what you have to offer, keep them happy with your offer, and retain them from switching to others, then you have product-market fit”

You can’t write a flowchart or playbook to get to product market fit. You can make some thumb rule kind of statements that can act as a compass. Now, when you follow a compass, it gives you calm and confidence, even if it is a broken one. One crude compass is to say that you spend some time crawling before walking and walking before running. Done that way you manage expectations well, including yourself.

Product market fit is not a degree or a trophy you keep forever; you can lose it at any time. Whenever a technology shift changes the market, product market fit could be lost.15 years ago, a Canadian company released such an addictive product that people called it ‘crackberry.’ However, the company failed to adapt to technology and market shifts. The value contracted from $45 billion to $1.5 billion. No one would even care to remember it. RIM & Blackberry are forgotten now. Generative AI is a technology earthquake. It is bigger than the one that the Canadian company faced. Such earthquakes change a lot of things. Those that have product market fit, have to ask themselves if they continue to have it. Whatever you had going for yourself may get wiped out. This is the reason every founder, and technology exec is spooked.

Application developers often overlook changes in the underlying technology platforms. The Gen AI earthquake every layer of the most known tech stack. Some argue that when building applications, the focus should be on the core functionality, like how musicians concentrate on the music itself. However, this is a faulty analogy. The businesses providing the playback formats (e.g., cassette tapes, CDs) often become obsolete. Companies like Sony Walkman and Moser Baer, once leaders in their respective eras, have faded into obscurity. As technologies advance, market demands shift. Developers cannot afford to ignore these changes in the underlying platforms. Those tied to outdated formats and business models risk becoming irrelevant, much like the forgotten cassette and CD manufacturers of the past.

Gen AI is not the same as AI. It is like saying alternating current is the same as power. Alternating current is a type of power, but a different technology architecture. It is the generative part of what it does that makes it interesting, but Gen AI is due to new technology architecture. Any piece of AI that uses a transformer or post-transformer architecture, which has been beating state-of-the-art (SOTA) benchmarks in several cognitive (i.e., human) tasks, is Gen AI. In tech, things change slowly, and then they change suddenly. It is like children; every day, they grow slowly, so one doesn’t notice the change and one day, they say the children are at the same height as them. Anything before this architecture is traditional AI, and regardless of the money invested in it, they are carcasses now. This is one thing that is hard for the human mind to accept, especially for first-time founders who have never seen a technology disruption cycle.

Gen AI brings AI below the API. No more does one need to stand up a team in ML, spend a few million dollars, and say, ‘I want to ensure that Hindi to English translation is taken care of,’ or ‘Voice transcripts must be accurate, or people won’t use it.’ GenAI now assures that.

Crawl phase — Are you on the right mountain?

The initial idea is very important. Derek Sivers made an apt reference, calling an idea “an execution multiplier.”

Startup success = idea X execution

Execution is table stakes. But those who don’t spend enough time initially thinking through their ideas subject themselves to the punishment of spiraling down pivots and iterations. Ask any second-time founder and they will say, “In my next product/startup, I will spend way more time researching the idea, understanding the problem, talking to customers, and competitors, grasping why the business model and industry value chain is structured as it is. I will ensure I’m going after a large problem and not get stuck in tiny niches.”. That’s why the starting idea is far more important than most people think. Because if you’re climbing the wrong mountain, it doesn’t matter how quickly you’re climbing it.

Now the question is how do you come up with ideas? There are many ways. Infinite ways. It is always easy for the human mind to deal with such options when provided in finite buckets. There are many ways to do it, such as market mapping, horizontal tools, vertical tools, foundation tools, etc., or even talking about the domains. One way we can do it is by putting out a tech stack with multiple layers to it.

‘Own the outcome’ provides unique advantages and expands TAM

In the context of B2B, the first place to think about this is in the context of productivity. Gen AI-powered apps are like productivity on steroids. If new productivity on steroids is going to be promised, folks are going to be intrigued, and new tools will be built. A new workflow will be built. The infrastructure supporting the new tools will be built up. One area that a lot of experienced founders are looking at is owning the outcome.

Here, it is like the services company or what would be called Professional/IT Services. Those who have built SaaS for Indian geography can completely relate to this easily. Suitable for other risk-averse, low-trust geographies. When you build an offering such as transcription, you can add the human into the loop. Traditionally, this is where the product founder in you may feel uneasy, but the boundaries between services and products are going to disappear as Gen AI shapes up. The difference between services and products was the margin structure and the need for people to scale the business but that is blurring. This will lead to the rise of Software and Services instead of Software as a Service and lead to a massive expansion of TAM. The entire B2B software put together is $250 billion in revenue. But the entire IT services put together is $1.6 trillion. TAM expands when you build Software and Services instead of Software as a Service. Most founders tend to get obsessed with products versus services. With AI, the boundaries between product and service are merging.

However, if you still feel very strongly against this, then go build the tool, the product.

Prompt to outcome horizontal tools are going to be wide

but eventually be crowded areas,

Two major paradigms are shaping up. The utopian promise of AI is that every tool will become an outcome. The dominant UX that caught everyone’s fancy is a prompt or even called a co-pilot. In previous software application tools, it was a command or a click that would generate an output that you end up manipulating.

In Gen AI, it starts with a prompt and results in an outcome. Prompt to Outcome is a major design pattern. The other one that is not very explicit but still useful is automation agents.

In either case, there are many ways to pick your starting idea. You can copy something. You can ask yourself the question of what is something that has not been done before that is enabled by this new technology.

Then there is the question of vertical versus horizontal tools. One risk that you will run here is that of being called a thin wrapper. It is not a problem as a starting idea, but if you stay that way forever then yes it may be an issue. Ignore that comment anyway; most people who have built businesses, do not make that argument. After all, SaaS is a wrapper around SQL. On the horizontal side, a lot of small, prosumer-type tools can be built. On the vertical, find unsexy, boring verticals that Silicon Valley startups or a big incumbent are not interested in. This you can find by doing 50 prospect conversations, 20 competitor conversations, and 10 other stakeholder conversations before you build anything.

Unpack those conversations, Pick a hair-on-fire, high-frequency usage problem.

Understand how they buy.

Learn what their current tool stack is.

Decide to build your startup in that area.

Doing this will help come up with one of the most steadfast ways of generating an initial idea. In B2B business, it is all about how someone makes a buying decision. You have to map your selling to that buying decision. The deeper you understand how that buying happens, the better you will be able to build a business. There is no replacement for that. There might be another disruption that happens, but this is not going out of fashion

Builder tools will go through a complete rebuild.

Now, if you are an engineer, you can think of coming up with a builder tools idea. It looks easy, but it is not straightforward. First, the bar for developers to pay is high. You have to be as good as GitHub Copilot. Again, to make a million dollars at a 1% free-to-paid conversion of $100/year, you need a million users. If you are selling to developers where they have to get others to adopt it within the enterprise, then the buying process gets complicated. If it touches any of their data, then they won’t want to adopt anything proprietary. They would want an open-source version so that they can see what is going on. You can eventually have a higher ACV instead of $100, but open-source requires a lot more upfront investment. It is like playing a dual game of baseball and golf. For the first 4–5 years, you have to make your open-source the most successful open-source, hit home runs like in baseball, and then you will have to spend the next 5 years monetizing where you have to play an ace game of golf.

How you fundraise for this has to be done in geographies that understand how to fund these types of ideas. Move to Silicon Valley. Is there a potential to build another AppDynamics-type company, absolutely, yes, but it will have to be built very differently.

What is also interesting is that the entire ML development process has changed dramatically.

Earlier, you would have a data scientist, and there would be an ML engineer. You would build a toolchain that would involve the presence of a MATLAB model followed by Python coding on low-level ML models like clustering or other supervised learning techniques.

But now, that changed. You need tools for the world of makers, shapers, and takers. Makers are those relatively few people who will be the Large Language Model/Foundation Model makers. Shapers are folks who are going to use the base infrastructure, build custom data, and fine-tune it. Takers are the rest of the folks consuming this using an API. This MLOps 2.0 toolchain is almost completely different from MLOps 1.0.

Platform and model layers are best picked for sovereign reasons

You can also work on a platform idea. But know that platform economics are very different. They can be very expensive. You will have to capitalize yourselves well, which means you have to raise the right kind of funding.

Keep in mind:

• Platform economics are very expensive.
• Unless barred by regulation, platforms are global oligopolies. You are competing with Google and OpenAI.
• Google of India is Google, and Facebook of India is Facebook.
• To train a model, it costs $50 million.
• Sovereign reasons can drive country-specific platforms.
• Example: India Language LLMs

Be sure of the reason why you are picking and want to build a startup in this layer.

Pick areas where you gain unique strength

When you are picking an idea, also keep in mind that incumbents have an advantage over you. Silicon Valley startups are going to have an advantage over you with funds they have access to. Pick areas where you can uniquely win. Platform ideas are, therefore, a more difficult space. AI builder tools are long cycles. Picking Vertical AIs where you can build deep domain understanding is a better alternative or even better own the outcome. Don’t pick an area where Silicon Valley would already want to invest in it. Will set you up for too much competition.

Staff a right-sized AI team

An experienced founder gets this right: you need a small but complete team. If you are building tools, you want to ensure that you have brought on board a kick-ass designer. In the age of AI, UX, and UI are what will help differentiate and even help build moats. You also need AI Engineers. What Gen AI has done is that it has moved AI below the API. Now, to build AI applications, you don’t need the knowledge of ML underneath, but knowing that will come in very handy. AI Engineers are not ML engineers. These are engineers who can build AI applications; they have some knowledge of ML but, most importantly, they know how to build great experiences. Most likely, Python application engineers will upskill as AI engineers. Marketer: you need a great marketer that can educate the audience. There is so much jargon, a lot of FUD, and paranoia in the space. A great marketer can help position and message as the landscape shifts. If you don’t have a good designer and marketer, you are at a disadvantage

Build a simple lovable complete product

Next, based on whether you arrived at the idea by copying a prompt to the outcome because you zoned in on a “hair on fire” problem in a specific niche, or you decided to build some builder tools, you need to build a simple yet lovable and complete product. You don’t want to think about it in terms of minimum, but rather as simple and complete. The key idea is that it is also a lovable product. In Gen AI, it is extremely hard for you to sell vaporware. Since it is a technology leap, it will not be a considered purchase, at least not for some time. You cannot sell what is not built and experienced by customers. Users are going to have to try before they can make up their minds. Design a great trial experience; a lot of it is going to be latent needs discovery. Whether they need this solution is going to be based on the mind-blowing demo they experience. When building, one of the things that is important to keep in mind is that the outcomes are going to be nondeterministic, which means that the same input and output cannot be expected in each iteration. That makes creating the output challenging How you think about the design is going to be different. So far, all computer systems have been like tools that can be used to create a certain output. In Gen AI design, it is input to outcome; the role of the designer becomes one of a curator. Both these changes mean that in development, the process becomes one of iterating and curating through a lot of generated options.

While we spend a lot of time thinking about the starting point of the idea, it is not the one that you eventually land on. Everyone goes through multiple iterations. While you are going through this, you want to make sure that you escape competition and disruption due to bigger incumbents who have an advantage in this technology shift. It is like playing a Pac-Man game, where you will have to do a lot of backtracking and escaping getting chewed up by competition in the journey of going from zero to PMF.

Oh yes! In this stage, there is absolutely no need to look at procuring GPUs or toying with open-source models. In this stage, you are looking to strengthen your conviction. You are making sure that you are on the right mountain.

Walk phase — Optimize experience, bake responsibility, and strengthen AI blocks

Once you feel that this core loop (of prompt to outcome) or the vertical you picked is the right place, Then you are ready to get up and walk further.

Applications that succeed and take off in GenAI must meet these five key parameters, especially in B2B contexts

• Accuracy
• Reliability
• Latency
• Delightful UX
• And becomes better every day

Every B2B AI SaaS application should track and improve across these 5 dimensions. If there are incorrect responses that are going to show up, then you will end up with a legal liability. In B2B, trust is key; if an application is not going to be reliable, then adoption is going to drop. One of the reasons why developers have looked to switch from OpenAI is cost, but the second reason is latency.

Gen AI has moved AI below the API. But if you have more data about the domain in which you are operating, Once you have nailed your use case, you should look at acquiring additional data. You can improve your AI by having data for your domain; now is a good time to think about the data. Now is also a time to invest in finding open-source models with fine-tuning to replace OpenAI. The difference between a fine-tuned open-source model and using OpenAI is at least a factor of 30.

Just like how data is going to play a critical role in winning, another critical thing that will contribute to winning in Gen AI is regulation. Unlike older SaaS applications, where unless you were in healthcare or finance, it was heavily regulated, you did not have to think about it deeply. In AI, regulation is going to play a significant role. Not enough work is being done in this area; if you don’t hire a team member for this, you should at least get an advisor who will help you navigate this. Put in guidelines for yourself to develop responsible AI. Once you take care of proprietary data and regulation, and optimize across each layer of your app for a delightful UX and continuous improvement, you will be prepared for the run stage.

Run — Watch your retention, map the influencers, and manage the breakneck pace.

ChatGPT took everyone by surprise; AI has always fascinated us. After many false promises, AI finally seems to be working. This has heightened people’s expectations. Some overenthusiastic marketers build a false reality. All this means that a lot of people want to try out things. One thing to keep in mind is that these customers are like tourists. They come for a few days and then go away.

If you are building horizontal prompt-to-outcome tools, then you should expect a lot of churn. Popular horizontal tool companies are already seeing massive churn. Which is not good news in one way. But if it is looked at in another way, you can think of this as how a consumer application behaves; the 20% that are returning are the real ICP for the product.

Watch your retention curves

Plot your week-by-week retention cohorts. Where you find an asymptote with the x-axis as time passes is the set of people you have product market fit for. Taking this idea further, what you should do is develop a retention-based product roadmap. You may have started with a certain ICP and assumption of building a roadmap, but understanding this retention cohort, doing a deep dive to understand their JTBD (Jobs To Be Done) motivation, and then shaping/changing the product roadmap will be very important.

On the GTM (Go-To-Market) side, starting with messaging and positioning, the key thing to remember is that this is a new category creation in most cases. If you have your names and domain names aligned to the functional use case, then it will do wonders for your traffic. You will have to use existing concepts to explain but immerse people and educate them about new experiences. It is not one of the competing but a new category creation playbook. Have a POV (Point Of View), and tell a story of change. Packaging is going to play as important a role here as it did in regular SaaS.

Influencers are key

On the channel side, the most important seems to be around influencers. Products that get mentioned by Sam Altman, Andrew Ng, Yann LeCun, Allie Miller, etc., do extremely well. These mentions getting slipped into one of the many newsletters and then hopping from one to another is what is adding a lot of visitors to the top of the funnel. Having a thoughtful approach to orchestrating everything from the name to getting the influencer engaged will go a long way. In SaaS, you must add your inbound engine foundation; content marketing and SEO are table stakes. There will be situations where some new platform changes will disrupt you. You should do the pre-work to pick an idea where disruption is not likely to happen because the space is boring.

When it comes to pivoting, one must do a hard pivot. Most founders consider pivoting to an adjacent market. Most pivots are not hard enough. If an idea is not working, one should pivot very hard, moving to a completely different space.

Give yourself at least three ‘lives’ to win in this maze game.

In the crawl stage, an inordinate amount of time must be spent on picking the idea because, in AI, the odds are against you winning. Second is the importance of design & product marketing. Because this is a new experience tool, not red ocean building. New models have to be included in the design to win and have moats. Pay careful attention to data to strengthen your win and don’t ignore regulation (or responsible AI) so as not to get wiped out. Also, ensure accuracy, reliability, and latency are taken care of for usage to happen. And when running, use AI influencers as a way to spread the word and do faster hard pivots. If you are in the crawl/walk, focus on the right idea and the solution, if you skip this stage and don’t give it due attention, you will have to pivot hard and then use another life in the game of product-market fit. OpenAI started with a robotic hand as the first product and after many pivots found success in chatGPT. Perplexity started as a text-to-SQL startup and after many pivots landed as a personal search engine.