slide image

Using Machine Learning to Drive Niche Job Board Revenue

By Alexander Chukovski, CEO

In the past few years, the rise of programmatic job advertising has created a fantastic pool of potential new revenue for niche job board owners.

Get a feed from A, R or J, import the relevant jobs for your audience and start getting paid on CPA/CPC.

Sounds easy, right?

Not really.

The Problem

As a niche job board, you want to ensure that your audience is served only jobs matching the niche.

And sponsored jobs makes it tricky to do that. You need to filter out only the ones that are relevant to your niche.

Most people start by creating heuristics - keyword-based rules that focus on the job title. Unfortunately, this approach can fall short for niche boards since job titles are often not detailed enough.

For example: I operate a crypto job board. By focusing on crypto-specific words like “blockchain,” “solidity,” and “web3,” I will skip plenty of jobs with “non-crypto” titles like “Online Marketing Manager” or “Sales Manager,” which are present in every crypto company.

Over time, job board operators often move to rules that focus on the job description. But they don’t work either – if I use “crypto” as a keyword, I will also get plenty of jobs from companies that include that keyword as part of their description, but are unrelated to jobs in the crypto field. This gets particularly problematic for niche job boards that serve technology employers, as their skills are often transferable.

You can also look into company whitelists, but that approach is not scalable.

So, in the end, most job board operators ignore this revenue opportunity and focus on selling job postings and manually verifying them. You will be surprised how many job boards operate like this, especially in the DACH [Germany, Austria and Switzerland] area.

What can you do?

The Solution

You can build a simple Boolean classifier that uses your historical data of “good jobs” (ones that match your niche) and examples of jobs not in your niche. It is an AI model that looks at a job description and says – yes, this job is right for your job board.

With modern machine learning and AutoML, you need a couple of hundred examples to get a sufficiently good model that can predict whether a job is relevant to your niche. You upload a CSV, press train, and you are done—no need to be a data scientist.

How expensive does it get? Google Vertex AI costs $5 per 1000 predictions (1000 characters in each prediction). Other providers offer cheaper options for higher volumes.

A Caveat

As always, the drawback of AutoML is that it gets expensive at scale – you cannot run millions of jobs daily. You pay for convenience. If you want to reach this type of scale, you have to host the model yourself.

You can make a very simple calculation though to determine the best course of action. Import 1000 jobs daily and measure the clicks and the revenue you generate. Plot this against the cost of prediction and you will quickly know how much you will be able or want to scale this process.

You can also try some work-arounds:

  1. Ask programmatic providers to exclude categories that are not relevant for you (for example, you don’t need gig jobs if you run an AI job board).
  2. Limit the classification task to the first 200-300 words (this approach needs testing).
  3. Use keywords to keep the selection broad and then use the model for the final selection.
  4. Switch to company whitelists, picking one or two jobs per company as an indicator.

And, before someone says “just use OpenAI and ChatGPT”, make sure to calculate this approach thoroughly. A job has an average of 2000 tokens, and adding a prompt to that baseline increases the amount to 2200 characters. At 1000 jobs, you are looking at $60 with GPT4. Of course, GPT-3.5 is cheaper, but I doubt it will provide good results.

There are plenty of ways to optimize the use of machine learning to generate additional revenue for niche job boards. Reach out if you’d like to discuss strategies in more detail. I’m at

Meet Alexander and get his new book ChatGPT Undressed & Unadorned: The Truth About ChatGPT in Talent Acquisition at TAtech Europe & The EMEA Job Board Forum on December 4-6 in London.