Automation of client acquisition with AI

How we automated Upwork job qualification

For a digital agency like Kodebusters, staying ahead of the competition isn't just about being quick—it's about being smart. Our latest project, dubbed "Sandra," shows how integrating AI and automation can transform the search for new clients on Upwork into an efficient, streamlined process. Sandra isn't just a system; she's become a vital member of our team, navigating through heaps of job postings to spotlight the ones that are just right for us, all while ensuring that we're among the first to respond.

This case study explains how we turned the challenge of manual job searching into an opportunity for innovation. By leveraging tools like the Upwork RSS Feed, Notion, and Make.com, alongside the analytical power of ChatGPT, we crafted a solution that not only finds job postings but also intelligently qualifies them with minimal human intervention.

Goal

Our goal was to create an efficient system to help find new clients on Upwork. The system needed to:

• Identify relevant Upwork job postings.

• Enable swift responses to job postings.

• Minimize the effort required to filter job postings.

• Simplify the process of replying to job posts.

Solution

We developed a solution incorporating the following tools and processes:

1. Upwork RSS Feed: To fetch job postings.

2. Notion Database: To track job posts and their statuses.

3. OpenAI: To auto-qualify job posts.

4. Telegram Bot API: To interact with the system through a bot.

5. Make.com: To automate the process.

Step 1: Search & Auto-Qualification

• Fetch and Parse: The system fetches the feed and parses the job description.

• Check for Duplicates: It checks Notion for duplicate entries.

• Qualification Decision: It calls ChatGPT to make a qualification decision.

• Save Results: The results are saved to the Notion database.

• Notify Team: Our team is notified about new matching job posts through a Telegram bot named Sandra, which sends messages to a private group.

Step 2: Manual Qualification

• Manual Review: A teammate manually qualifies the message, and a webhook is fired.

• Update Status: The system updates the Notion database entry to mark the job post as fully qualified after passing both automatic and manual stages.

• Notify via Telegram: Sandra marks the message in Telegram with a thumbs-up or thumbs-down emoji to indicate that the post has been processed.

• Forward to VA: Sandra forwards the message to our Virtual Assistant (VA), Lisa, for manual applications on Upwork.

• Cover Letter Suggestion: Sandra generates a suggestion for the cover letter using a custom prompt and template that works well for our agency.

Results

Since the launch one month ago, the system has processed approximately 1,074 job posts. Key metrics include:

• Auto-Qualified Posts: 251 posts (23.37%) were auto-qualified by the GPT agent.

• Applications: We picked and applied to 107 posts (9.96% of the total posts).

• Time Saved: The system saved us at least 18 hours of manual qualification work.

Cost Efficiency

• Make.com: We optimized the automation, spending just $10.59 per month on Make.com, keeping it under 10,000 operations per month.

• OpenAI API: We spend less than $5 per month on OpenAI's API, as the majority of job qualifications are done using GPT-3.5. We use GPT-4 only for the top 9% of fully qualified posts to create a cover letter template.

Ongoing Improvements

We continuously improve the system to make our AI assistant, Sandra, smarter and more efficient. Sandra now also helps set the proper tone for our day.

Now your turn

If you want to cut operational cost by automating repetitive tasks, book a discovery call below.