Using AI in building a SaaS product

The internet is littered with either ‘AI is going to take away jobs’ scare mongering opinions or ‘how AI will make you superhuman’ type self-help stuff. I feel anyone who has used these AI tools in their work will realize both of these opinions are not entirely true. In our experience, tools like ChatGPT, GitHub co-pilot serve as very useful digital assistants.

When AI turned to be not so ‘intelligent’

The other day, we were trying to create realistic demo data using ChatGPT. We gave the table structures and the relationships. The prompt included the kind of data we needed.

While there were more than a dozen tables, ChatGPT generated data only for 3 or 4 tables. When asked to generate the data for the remaining tables, there were one or more of these errors:

• The data had many columns with placeholder text like ‘Col1’. ‘Col 2’ instead of realistic data
• Some foreign key relations were missed
• Some tables were still missed

When pointed out, it generated data for tables that didn’t exist as well. It had added additional tables that didn’t exist in the schema. 😄

There were multiple iterations and a few hours spent before we decided to stop and have the rest manually done.

So are we saying it was not useful to us at all? Definitely not. Actually it has been helpful in certain situations. Here’s how it helped us in improving marketing copy.

A digital assistant to collaborate with for improving marketing copy

For the website - scalepeak.in, we first wrote up the marketing copy on Notion. The tone, the style and substance of the message was clear in our mind. For example, we wanted the tone to be authentic and genuine. We wanted to avoid jargon, cliched sales-speak and marketing mumbo-jumbo (isn’t that a cliche? :) ). So now it was time to let ChatGPT help us as a collaborator to suggest edits to the draft keeping the tone, style and other requirements of the message in mind. In each iteration, we picked what we liked from the suggestions and made other adjustments and repeated the process.

While the final product of the process was good, I have to say that on its own ChatGPT couldn’t come up with the required copy magically without us having written the draft and us having a sense and taste for what we want.

This process has worked well for improving cold email copy as well.

Other scenarios

There are scenarios where AI tools helped in programming - like automating simpler stuff, getting unstuck with not-so-complex problems etc. We will talk about it in another post.

In conclusion…

If you are good at something, AI can help you in bettering the output. If you write decently and have a sense of what’s good, you can use AI to write better. Similarly with programming. If you know programming, you can leverage AI to do things faster, learn new things faster.