AI in Business, Use cases, and Understanding the Purpose
Like every business with access to the correct resources, an interest in exploring AI is inevitable. Every business is jumping the gun to get on the bandwagon of AI related tools leading to a saturation in the market of such tools, and it is becoming increasingly more difficult to differentiate yourself from the competition. This is a problem that TribalScale faced, where we have an interest in exploring "AI" (I put quotations around this because of how easily the word is thrown around) but unsure how or where to start. The boom of this technology can be seen quantitatively where the AI market will be worth 1,811.75 billion USD by 2030, indicating a compound annual growth rate (CAGR) of 37.3% (2023-2030).
These thoughts and conversations began a long process of discovery and due diligence to understand where our problems lie at TribalScale and how we can selfishly solve them. This was done before we go to the market and start promoting how AI can be used to help digitally transform businesses ensuring we have the correct foundation.
AI generated image created with Bing's Image Creator
User Research
User Interviews
As the interest of AI grew it was important to speak to the different departments in the organization to gauge where problems lie. These conversations included the HR, Sales, Product, Engineering and C-Suite teams. It was important as a Product Manager to move these conversations away from ideation and focus on where problems lie.
This involved curating a list of questions to help guide the conversation to focus on identifying gaps/problems and away from an impromptu brainstorming session. These questions were influenced based on conversations with Sheetal, CEO of TribalScale; where the general goals were emphasized, notably “What can AI do to make your life easier?" and "How can we leverage this technology to be on the top of our game?”. As well as conversations with Tim, CTO of TribalScale where it was noted that “AI can be leveraged to better identify patterns in large datasets that humans find challenging; we’re using AI to improve processes at TribalScale while also transforming processes for our clients”. All which helped guide my later conversations with other TS team members.
And depending on the conversation with the various departments I was able to get some super specific insights or some led to dead ends. It was important to realize where the potential of developing something was and researching if tools exist. For example with Sales, the team already has automations with Hubspot built-in that make their processes very seamless. So prioritizing our efforts elsewhere was done where we could have a greater impact.
When collaborating with the HR team, the following questions and pain points were identified:
•
What is the current process of recruitment from crafting a job post to collecting resumes to selecting candidates?
•
How much of the work is done manually?
•
Time for completion of a typical recruitment cycle?
•
What do you not enjoy about your process/job?
Pain Points
•
Manual reading of all the resumes; each resume takes 20 seconds to skim by an experienced recruiter
•
Human error occurs when skimming resumes
•
Lack of context of where people worked on their resume
•
Often takes 21+ days to complete a whole recruitment cycle
User Personas
For our potential use case, a user persona of an HR Executive can be drawn up. This outlines the typical HR Executive who is extremely busy with meetings, job postings, interviews and more. The need for an integration to assist in some of these manual processes is prevalent as listed in the pain points and show in the persona below:
User Persona of an HR Executive
Solution Outline
ResumeAI, a tool that will allow its users to automate the reading of resumes and gain insights that were not available beforehand. This application was developed as a low level introduction to solving a relevant problem with accessible tools. The steps to use the tool are as follows:
•
Folder of resumes or individual resumes are uploaded into the dropbox
•
Integration of Greenhouse is in place (and being worked on) to allow users to upload resumes directly from job postings on the platform
•
Description of job and tags are also inputted
•
OpenAI integration allows for resumes to be reordered based on relevance and attaches summaries for the resumes for the recruiter to read
Preliminary UI
Technology
A more refined understanding of the technology used is as follows. The team took measures using AWS and Open AI to create this tool in a cost effective manner. This was important as the tool is being used as a stepping stone to something larger so to make it worth it, it was important we were not depleting our resources. Below is a diagram of the architecture that was used for this tool. It is serverless and leverages APIs from AWS and OpenAI.
Architecture Diagram
How?
The resumes are sent to an s3 bucket directly rather than using an API gateway, along with the tags and descriptions inputted which is stored as Metadata. This invokes a lambda function where the resumes are then passed through an AWS API named Textract, which will extract the text from the PDFs leaving us with the raw text to analyze. What's cool about these lambda functions is that they run simultaneously, rather than triggering for a resume one by one, it does it together, saving costs as you have to pay per use for these lambda functions! Once we have the raw text it is then sent over to be analyzed by the OpenAI API. The API gets given a few orders to do for each resume including generating titles, summaries and analyses, and calculating how many of the tags are present in each resume. This information is then stored in an AWS database under the key that was assigned to the session in use.
Sample Resume Analysis From Tool
This process like others went through different rounds of iterations such as utilizing API gateways for s3 uploads and using different APIs such as Rekognition. Prior to implementing the current architecture, one response was taking 20-50 seconds per file which is approximately the same time it takes an HR Executive to skim a resume.
This led to more research and experimentation to finally land where we are today. The system was able to get to a point where now individual resumes can be read in up to 6-8 seconds and in experimentation found that after clicking upload with the pdf and description/tags to receiving the analysis took 6.4 seconds. This is approximately a 75% time reduction it takes for an experienced HR executive to skim a resume and now instead of skimming, they are able to get a detailed analysis of a resume to make a better informed decision on a candidate. Overall, due to the lambda functions being invoked simultaneously rather than per resume, the overall time to complete the request will take as long as the longest resume to parse. Delays in reading the resume can happen through incorrect files being uploaded, naming conventions and lack of organization in the files; all to be addressed in future iterations.
Outcome/Findings
I found that working with AI is very tricky. It is the new craze where businesses are jumping on the bandwagon as fast as they can to get to market before others beat them to it. While I applaud the efforts, I found it very important to take a step back and reflect. The value of setting a foundation for a project before beginning and ensuring it has value to your company or your consumers is very important. While this project was the beginning of something very cool, it's just the beginning.
As a product manager working in AI, I've noted some key steps in navigating this field of AI development:
•
Research: Understanding the market and what is already available. Sometimes the most impactful way is not to create something revolutionary but make something that exists better!
•
Experimentation: It's difficult to fathom the limits of AI and often times it's easy to get overwhelmed with the capabilities of it. Therefore it's a good idea to just experiment with development and see what you can do (like outlined in this article).
•
Reflection: Upon experimentation of potential development ideas its crucial to take some time to reflect on what was done and gather insights. This will help inform future iterations and/or projects you chose to work on.
•
Project Definition: If this is something you chose to move forward with as it has value, the next steps would be to follow your procedures to set up the project as you would any other project. This ensures the foundation of the project is solid and ensures proper time and resource allocation.
Next Steps
TribalScale as an organization has been making strides in growing their competency with AI, recently forming a strategic partnership with Senso AI— a SaaS based solution that allows companies to customize niche chatbots for their internal usage such as reading through wikis, documentation, websites etc. Potential use cases include integrating this chatbot on top of our resumeAI, assisting with team allocations, scoping, answering questions about the TribalScale wiki and more. For more information about our partnership with Senso AI, click here to read our full announcement.
The sky's the limit when it comes to AI, it's all a matter of how strong the foundation is for it.
