An interesting opportunity in the age of AI for a SaaS company to grab.
Last week, when I wrote about the idea of building telepathy as the new form of communication, for the next decade, some people reached out to me and said, “Please write about something practical.”
I will write towards the end, on how to quest for the impossible ideas. For now, we will talk of a practical idea this week. This week’s idea is to build a service that provides suggestions via an API.
What is Suggestions as a Service?
Every internet business needs to provide some sort of suggestions/recommendations to their users. For example, a social network needs to show “people you may know”, a SaaS product needs to prompt their users, “here’s what you can do next”, a consumer product needs to upsell using, “you might also like”.
Every product needs them and every product needs to build such a thing in-house. One might argue that because the contexts in which suggestions are made within a product are so diverse across the industry, it becomes very difficult to make a general-purpose solution. But broadly speaking, there are only a limited set of contexts in total, but the data that needs to be suggested might be varied to a larger degree.
For instance, in an e-commerce website, the social network’s section of “people you might know” could be translated to “people who bought this also bought”. The fundamental equation is same but the data on which the equation is calculated is different. And that makes it a problem worth solving.
Search’s Better Half
Have you heard of a company called Algolia? It is backed by Y Combinator and provides Search as a Service. If you ever need to put a search box in your product, they take away the pain of bringing that search box to life. Setting up your own Lucene or ElasticSearch instance, creating the index, seeding it and then writing the back-end and the front-end code to make it all work together will easily eat at least a few weeks of your schedule. Not worth it when the core part of your product is not the search (if majority of the users won’t use it every day).
Algolia fixes that for you and it takes about half an hour to get up and running.
Why doesn’t something like this exists for the suggestions, I wonder.
How to pursue an impossible quest?
“It’s kind of fun to do the impossible.” — Walt Disney
I started this edition questioning whether we should be reluctant towards impossible problems? If not, how should we pursue them.
This week’s idea isn’t on the same scale of building telepathy but don’t take it for an easy problem to solve. Let’s see how we can go about solving an impossible problem.
In the first look, it makes absolute sense to only follow the practical problems. Why should we even talk of things straight out of science fiction? Why don’t we put our efforts into fixing things that can easily be fixed by any ordinary human being? It makes me think.
To be entirely honest, I believe, all trees (small and big) are only a tiny seed in the beginning. The art of pursuing big and audacious ideas lies in identifying the shape and scale of the seed (also known as, the v0.1 of the product).
Besides identifying the seed, it will ask great efforts from you. Pursuing unpractical ideas require you to read hundreds of books, talk to hundreds of experts and speed-learn everything on the subject. You’d be continuously learning new things up to a point where you’d start feeling, “I’ve read tens of books and there are still hundreds more to go.” You’d be on the right path.
Crowd quests for easy. The lone explorer quests for the fruit on the highest branch.
v0.1 of Suggestions as a Service
I believe the version 0.1 of such a service will be to build a graph that contains entities present on the internet (everything from movies to books to TV Shows) and the relations between them. An easy way to get started is to base your graph on the Wikidata.
Once that happens, you’ve got a data structure that can answer really interesting questions that you can throw at it. The final step would be to derive a query that you’d make on this graph to get the suggestions.
And that method of making the query will be different for every context. For example, suppose you’re service Quora and give a question, you need to find more questions that the user might be interested in. The steps would involve tokenizing the question, identifying the nouns in it. Build up a query that would find the connected nouns. Return the questions that contains those nouns and adjectives.
For example, a question about the author J K Rowling might return questions about the author George R R Martin.
This won’t be an exhaustive service at first. More things will get built into this. This is the v0.1 — enough to test out our hypothesis and get a few paying customers.
Start by selling a Wordpress plugin for $10 per month subscription that suggests what posts to read next to the users. If customers see that your suggestions are better than the default functionality, and increases the time their visitors spend on their websites, they will spread the word about you.
The art lies in developing a minimal viable product that sells. Everything else will take care of itself.
About the author
Mohit Mamoria is the curator of a weekly newsletter, Unmade, which delivers one idea from the future (just like this one) to your inboxes.
Originally published on Unmade.