We live in a world where machine learning and artificial intelligence are now integral to many industries, including traditional businesses all the way through to SaaS technology companies.
Leveraging artificial intelligence to grow product and help supercharge growth strategies is no longer an optional but I would argue an essential requirement to stay competitive in this market.
In this lesson, you'll learn:
The history of product growth
The move to automation
Where we are now, with machine learning and product growth
Mo Jalil:
Hey everyone. My name's Mo. I'm a serial entrepreneur and a lifelong engineer. Having spent many years in finance at Goldman Sachs and I've founded various companies. My last company was awarded at Cannes, was featured on Forbes and Unilever dubbed it as one of the top marketing startups. My current company Metapair deals heavily in AI, natural language processing, and [generative 00:00:25] learning models. We take a very deep product led approach to how we operate. I'm really excited to talk about how you can use artificial intelligence and machine learning to grow your product.
Mo Jalil:
In this presentation, we're broadly going to be covering a few things. Well integrated ML features in product-first companies and how that's led to adoption. And the second being, how you can use machine learning, ML for internal analysis. How to identify opportunities to grow your product. And as we do that, we're going to cover examples, questions that you can consider and things that you should be asking yourself.
Mo Jalil:
So why don't we go ahead and start. If you're tuning into this presentation, you're somewhat familiar with product led growth, but for those that aren't, it's definitely worth a quick summary. Product led growth or we'll refer to the shorthand, PLG. Is a strategy where the product is the primary driver for sales, adoption and growth. In essence, it means growing your business without having to invest countless dollars in advertising or having dedicated sales teams. It is fundamentally different to a marketing led approach or a sales led approach. Companies with a PLG strategy, think Slack, Spotify, Dropbox, Netflix have products where users are able to get value before they start paying. And they're able to grow faster and more efficiently by leveraging their products to create a pipeline of customers.
Mo Jalil:
And even if you're not seen as a PLG company, you don't do SaaS, you don't have a freemium model. You're creating maybe a traditional large enterprise. You can still benefit from the principles of PLG. On the subject of large enterprises. There was a time when selling new products could take months, quarters or even years. And for some of you, I'm sure that's still the case. Now we're seeing that software showing up in the office all on its own, introduced by individuals who love using it. Champions its wider use, and people are finding, downloading a document products without any directive from their managers. And in fact, the models basically flipped. Users are telling their management teams, which software to buy, which then drives the adoption within the organization. And let's use Slack as an example, by using the product, we have eliminated the need to filter through and send hundreds, if not thousands of emails. And on top of that, it's fundamentally changed the way that team members communicate with each other. That in itself is a huge value proposition for the users.
Mo Jalil:
Remember, in PLG, the primary driver of user acquisition is product usage and companies like Slack have taken this approach and have been handsomely rewarded for it. Just take a look at the market, the fastest growing companies in the world, the most valuable private companies in the world are all product led. So what about ML? Almost a decade ago, venture capitalist, Mark Andreessen said, "Software is eating the world." What he meant was software would be critical for all business to operate and software companies would take over large swathes of the economy. Fast forward a decade later, you can see how true that is. And now we're hearing the same about AI and ML. Whether you realize it or not machine learning is pervasive in many of the products that we use. Some popular examples are auto correct, or to complete new emails, your phone notifications on commute times, and recommendations on Spotify, Netflix.
Mo Jalil:
The remarkable thing about these features is that these very complex algorithms are integrated into the simplest product designs. And this is very well with PLG. Many of the core competencies are fundamental to good machine learning. So data on product and product usage, having a deep understanding the customer needs, and having an analytical and experimental mindset. Product led teams all already have these things. These are also fundamental to building and using machine learning effectively. So how does machine learning actually create value for my products? We're going to take a look at some of the most widely used ML products and features, and they probably fit into essentially three buckets. First up, recommendation engines, great for converting customers from free trials to paid subscriptions, upselling, customer attention. A perfect example of recommendation engines, it's not on this slide, but Stitch Fix uses recommendation algorithm to connect its clients to tailored items that fit their individual style preferences. And that's really important, especially when you're in fashion.
Mo Jalil:
Netflix, again, Netflix recommendation, personalized suggestions for you. So everyone here is familiar with go through your Netflix queue and seeing all these different shows that are designed, particularly for you based on what you've watched and what you've liked. And lastly, touch on banner ads and etailers and third parties have basically looked at your cookies in your browser history and figured out what products you're most likely to buy and then show you those products. Next up computer vision, fantastic for productivity gains and creating a very smooth user experience. Think about Uber. When you take a picture of your card, it adds your payment details to the app. Normally typing your card details can be a pain. There's a really long number, and it can be a really bad user experience, but that use case of being able to take a picture and then having that filled out, reduces the friction to adding an additional payment method.
Mo Jalil:
Another example is Intuit Snaptax, a product that uses optical character recognition. So it recognizes which characters you have from a picture to allow users to use their mobile app, to take photos of tax documents, and then submit different processing. Something again, people generally don't like doing, and that's made that user experience so much more better. On the subject of OCR. We're in our final bucket, which is text analysis. Now text analysis and natural language processing is one of the most funded and researched areas in AI. And it is fundamental in the way that actual humans communicate with machines. So technology in this space is fantastic for customer care extracting information. Think about the amount of data that's encoded in text, we're talking legal documents, contracts, just so much information is there.
Mo Jalil:
When you think about examples of text analysis in products that work really well, there's just a ton, and I suppose the default and go-to here is always Chatbots. Think about Alexa for business, Drift, which is used for sales or Survey Sparrow, which is again, used for marketing and sales. And on the consumer side, we have developed Xiaoice, which is used for marketing and sales, which is used for hundreds of millions of people every day. It's based on an emotion engine, essentially an empathetic chat bot. So we've just looked at how ML and AI can be used to create value for products, or what about value for your team? And internally for yourself.
Mo Jalil:
You can do a lot with data with the correct dataset, you can hack your growth. You can make predictions of competitive data, marks campaigns, your customer personas, and predict user behavior, all ground level stuff that contributes to growth. So why don't we take a look at that?
Mo Jalil:
First up is product market fit. And this is probably one of the most valuable things you can do right off the bat, and you don't even need a personal data set to do it. With resources like Kaggle and Google datasets, You can get public data to answer a lot of your questions on how you should position your product. Now here's the example of using store data to determine the factors that go into like a five-star rating. You can predict which factors have led to that rating, and then the likelihood of receiving a rating yourself, if you have that in your own product. This is a great for figuring out your product market fit and starting the thought process around how you're going to position your product. But the data set like this, you can identify gaps in the competition and where your product could fit to have the most amount of success right out the gate.
Mo Jalil:
Next up pricing. One of the hardest problems businesses face is pricing, and you can use publicly available data to find the current pricing equilibrium. I'm going to go through an example here of Airbnb. This is Airbnb's public dataset, and you look at the homes and use this information to figure out where you should price based on factors like location, what features you have in your home. And you can actually apply this same methodology to a SAS product. There's this really neat dataset that you can get from Google data. Google will search data, I'll add the link at them at the bottom of this, where it shows how enterprises have been spending on SAS products in the course of 2020. So you can actually look at how much they've been spending. At what point did they stop subscribing and use that to set a very competitive price right out of the gate.
Mo Jalil:
Another really useful use case is building customer personas, but machine learning. It's probably the most, one of the most exciting things that you can do to be honest. Generally, businesses want to find and retain customers that grow their business. And for example, Stewart Butterfield, the CEO of Slack talks about this, where they use data to predict the customer's most likely add teammates to their workspaces because when you add teammates, the audience grows, and so does the customer base. If you own another SAS product, you might want to predict which customers are likely to make a purchase and talk with them accordingly. If you were a healthcare company, you might want to build personas around, who's likely to show up for appointments.
Mo Jalil:
Now, the easiest way to identify high value users is to predict basically three things. LTV, so lifetime value, churn, and purchases, or the likelihood of them making a purchase. So you can build your personas around different customers and see which ones have the lowest and highest likelihood of performing any of these actions. So given a particular persona, you can see the LTV, the likely to churn and how likely they are to make a purchase. That's great for growth hacking. It's also really good for comparing different customers side-by-side.
Mo Jalil:
And finally, content optimization. If you have a content strategy to generate inbound leads, then this can really useful for you. It can be great for identifying the type of content and especially the duration that you should be targeting. It's also one of the easiest areas to collect data. So if you're using Google analytics, I mean who isn't, or an SEO to like Ahres. Take a look at this with Udemy dataset for example. Udemy for those that don't know is a online education platform. And with this dataset, you can see the different subject areas that users on the platform tend to like. You can quickly see the kind of courses that they want to take as well as the number of predicted subscribers. You can also see, and this will be the key, the optimal content duration. So at what point do students and users start dropping off? Now this example isn't specific just to your Udemy.
Mo Jalil:
Content metrics are bonded online, and you can use tools again, like Ahres and Hotjar. And the fantastic thing about them is that they allow you to export the data as a CSV, which is perfect for most platforms that use machine learning or most models that you want to train on. It's already kind of in a format that you can use. So we just looked at how AI and ML can really help with growth. And really, there's a lot use cases out there. If you do this, there are a few big gotchas to keep in mind. Number one, data, to do any kind of AI and ML. You need a dataset. Now, luckily, most PLG teams already has some they do work with. And it's actually not as difficult to source information as you may think. I've already mentioned a few places you can get data, as well as you can source it from your product.
Mo Jalil:
Google analytics, if you have that, Hotjar, Ahrefs. Another thing you should do is consider how you store and clean your datasets. It's likely that you'll need to do some kind of pre processing on the data to make it useful. And if not done right, it can be a huge time sink. Lastly, you also want to think about how you store and keep your data relevant. As new information comes in, you're going to have to update your models. You have to rerun them. Do you do it in real time? Do you not do it in batches? So when you think about how that's done. Number two, humans in the loop. With AI and ML, there is some point in need to have human support. Whether that's building datasets, that we just talked about, whether it's handling edge cases or verifying the accuracy of a model, you will need to design how humans fit into the role of the product. At least from an ML and AI perspective.
Mo Jalil:
There are a few things that you can do where you can actually make it actively beneficial for yourself. A perfect example is feedback loops. If you have feedback loop in your product, you can use that to passively collect data that then is used indirectly to train your models and make them even better. Point number three, implementation how you implement your AI and ML solution can make or break whether it's successful. And there;s an example of something that we did. We did analysis and ran a number of surveys, and it was clear that our users really wanted a recommendation as a feature. That was their most requested feature. So we spent a significant amount of time building a deep learning recommendation engine. After we released the new feature, we found that actually the usage was incredibly low. Did some investigation that it turned out that the deep learning recommendations while really accurate were slow.
Mo Jalil:
It took anywhere between 5 to 10 seconds to generate prediction. That was just beyond the tolerance of our users. To test this theory out, we switched up with a basic hyper filtering approach. So for people that don't know, it's a different algorithm. And this algorithm essentially took sub-second to generate a recommendation. And what we saw was usage skyrocketed. Now the lesson to take away here is that not only do you need AI and ML literacy in your team, but the way that you implement it really matters. Now, luckily there are many, many platforms or solutions that would do a lot of the heavy lifting for you. But it's something that you need to be thoughtful about.
Mo Jalil:
So in summary, we've looked at why, where, and when ML and PLG go hand in hand and how to use machine learning to help grow your product. And really regardless of the type of business that you work in, or you operate, whether that's an enterprise, B2B, B2C, if you're not implementing some of these principles, you really should, and now it's much easier than ever. So if you enjoyed this presentation, you have any questions about what we've talked about, and you want to go into a bit more depth. You can reach out to me on any of these channels and be happy to talk.
Mo Jalil:
Hope you've enjoyed that. Take care.
