Introduction

In order to ensure that the productized services I plan to offer on my site have real potential users, I’ve built a system for “pre-validating” potential services. Pre-validation is the process by which I try to find instances on the internet where people working in the specific industry have specified an interest in someone performing a service, or have posed a question about the specified task.

The Stack:

Make.com
Airtable
OpenAI
RSS Feeds (Reddit, Remotive)

The Plan:

I have identified 3 client niches that I want to investigate as a part of this trial run of my productizes service business. Each of these niches presumably have online communities in which they discuss relevant topics with one another, send out job requests (either full-time or freelance), and ask for feedback or assistance on things related to their job. My plan is to:

(1) Find a selection of 2-3 different forums, job boards, or other online communities (i.e. Reddit, Facebook Groups) relating to each niche/relating to remote jobs currently hiring.

(2) Aggregate the data from these platforms into an Airtable spreadsheet, either manually or using an RSS feed*.

(3) Feed this data into the OpenAI API using Make.com and ask for a brief 1-2 sentence summary of what the post or comment is discussing + some tool categories.

(4) Manually go through each generated summary and try to find groupings of topics which might suggest a possible productized service.

(5) Create the service description for publication on the website (format TBD).

‍

* What is RSS? 
RSS is a “web feed” or “news feed” that converts website data into a computer-readable format. They can be accessed through a regular URL link. Many websites have RSS versions, usually indicated by the orange and white symbol.
‍
On Reddit, you  can turn any feed into an RSS feed by adding .rss to the end of the URL.

Walkthrough:

(1) Niche Selection

Some niches I’ve decided to tackle in this first iteration are:

1. Small-to-medium freelancers
2. No-code platforms
3. Small think tanks
4. Software companies based in São Paulo, Brazil
5. Construction/urban planning companies

I chose these niches because they offer a sufficient degree of specificity, while still being large enough to garner a significant number of paying customers.

I’m going to start with the freelancer, no-code, and construction niches first because they have the strongest Reddit presence (which is easier for this trial run).

Small-to-medium freelancers:

https://www.reddit.com/r/freelance/
[RSS: https://www.reddit.com/r/freelance/new/.rss?sort=new]

https://www.reddit.com/r/Freelancers/
[RSS: https://www.reddit.com/r/Freelancers/top/.rss?t=month]

No-code platforms:

https://www.reddit.com/r/nocode/
[RSS: https://www.reddit.com/r/nocode/top/.rss?t=month]

Construction/urban planning:

https://www.reddit.com/r/ConstructionManagers/
[RSS: https://www.reddit.com/r/ConstructionManagers/top/.rss?t=month]

Remotive

Additionally, I found some general RSS feeds for live remote job postings in a couple of plausible industries right now. These include:

https://remotive.com/remote-jobs/feed/writing
‍https://remotive.com/remote-jobs/feed/marketing
‍https://remotive.com/remote-jobs/feed/project-management

The idea here is that if companies are hiring remotely for these position, they (or companies like them) might be interested in purchasing the job posting as a service as well.

(2) Airtable

I then created a spreadsheet in Airtable which aggregates all of these RSS feeds into a single location. I like this format because it allows me to return to this sheet in the future and add more feeds as I am looking to validate more ideas.

Because I plan to integrate this with Make.com, I also added a trigger field and a “Last Modified” (called “MAKE TRIGGER”) field, which allows me to tell Make.com that I want that particular RSS URL to be sent through the system. Note: for the Reddit RSS feeds, I’ve used both the newest posts and the top performing posts of the month (because I feel this gives me a better representation of what people are talking about).

(3) Make.com

My scenario in Make.com does the following:

‍

‍

(a) Watches the records in Airtable for the MAKE? column to be modified (here, that means for the checkbox to be selected).

(b) It goes to that RSS feed and extracts the data from the most recent 25 posts made there (the 25 limit is the hard limit from Reddit). This outputs the HTML from the text on the page, as well as the Title of the post or job listing, and its URL.

‍

‍

(c) The HTML text (”Description:” in the output) is sent through a text parser module and converted into plain text.

(d) Those variables are sent back into a new Airtable sheet (I call PREVAL), where it creates a new record.

‍

‍

(e) The parsed description text is sent to the OpenAI API where it generates a completion based on the prompt below (I developed this prompt using some trial and error):

SYSTEM:
You are a helpful idea pre-validator bot. You should take in text, which is either a job description or a forum post, and come up with a one to three sentence description of a productized service tool which could be created around the text. The descriptions should be very general, and should describe exactly what the tool does. Do not create a platform, create a tool with a very specific function. You should use the format:

[BRIEF DESCRIPTION OF THE SERVICE]|[SERVICE CATEGORY]

[SERVICE CATEGORY] should be a one to two word description of the type of tool it is.

I also have an ASSISTANT/USER few-shot prompt below, to demonstrate to the LLM the kinds of responses I’m looking for. This looks like:

‍

‍

I used [SAMPLE TEXT] as a placeholder, because the real text is too long for the screenshot. I just took the output from a test run and manually wrote what a good output would look like. This is generally a best practice when using LLMs on Make.com.

(f) The completion from ChatGPT is sent back into two columns called “service-idea” and “service-category” on the recently created record in the PREVAL sheet. I split the output from OpenAI using some array functions (see image below).

‍

‍

Et voila! I have created a list of 275 “pre-validated” ideas for tools based on the things real people are talking about on the internet. Realistically, less than 10% of these ideas are any good — but as I sort through them, I should be able to get a better idea of what tools I’d like to start with.

‍

‍

Keep following along on Memberstack to see what happens next! If you have any questions (or want me to make something like this for you) reach out at me@robertjett.com.

‍