You Just Got Told, “We Can Build That AI Solution You Need for HR In-House.” Now What?
Let’s face it: times are rough, and the budget is tight. Now more than ever, corporate functions like HR need to show everyone how scrappy we can be. What better way to showcase your ability to save the company a few dollars than by embarking on a collaborative journey with your IT partners, pooling your skills and resources to build an in-house solution instead of buying it from another SaaS vendor?
With all the out-of-the-box AI solutions you can turn on (hello, Copilot) and all the open-source models you can use for AI tool development (looking at you, Llama 2), it should be easy to build your solution, right? Well, kind of. At this point, I can tell you about all the reasons why buying a solution might be easier and cheaper. But I’m guessing you clicked on this article because the Buy option has left the building, and you have to make the Build option work (and RPA of behind-the-scenes processes is probably not good enough of an answer anymore).
So, if you need to figure out how to make the AI Build option work, here is where I would start:
1. Define the Problem (really, really clearly)
I don’t care what you say; “We need to do AI” or “We need a chatbot” is not a problem statement. It is someone somewhere looking to do the minimum to hit their performance goal for the year (there, I said it). Figure out what that high ticket question you are looking to answer is. Is it:
Freeing up your resources’ time to focus on something more high-touch?
Standardizing activities to ensure consistent experience and responses throughout the company?
Predict resignation or low-performance indicators?
Respond to Tier 0 and Tier 1 HR Operations questions?
Walk everyone through Open Enrollment to ensure higher satisfaction?
Regardless of how mundane it may sound, orient yourself to the problem first before proceeding with the Build work. Otherwise, you risk having your developers and product team direct you on potential issues. That could lead to an end solution that doesn’t really do much other than provide additional hands-on AI development practice time to the team.
2. Be Specific on How You’d Like the Solution to Behave
The beauty and curse of Generative AI is that new answers are created based on the model’s architecture and training. So, while GPTs are great at providing solutions in a human-like manner, they also require a lot of training to be tuned for your culture, lingo, and workforce. As you plan for the Build solution, think about the user journey:
What situations would your users need to interact with the solution?
How would you like them to feel before, during, and after their interactions with the solution?
Do you want to have the option of referring them to a human if the AI solution can’t resolve the issue?
Do you want to review the questions the AI isn’t able to answer and provide additional content for an automated response next time?
How would you like to audit for consistency in responses and verify the accuracy of all responses?
Do you need to have records of all AI interactions for traceability purposes?
3. Be Very Clear About the Scope and Extensibility of the Solution
An AI solution that can help you answer questions during Open Enrollment is different from an AI solution that can handle new hire onboarding activities end-to-end. Building an AI solution for a specific activity is easy; by the time you get to activities three, four, and five, you may experience a scaling issue. To save the future you some grief, plan ahead and determine if you are looking to build a simple solution for the current problem or if that solution needs to take on additional tasks and scope in the future.
4. Determine Your Time to Value
Simply put, when do you need to have your problem solved? Sure, you can delay your solution for one or two quarters because your IT Partners are delayed or experiencing issues. But by Q3, you will likely need to indicate results at least. So, be clear with your partners about when you expect the solution to be delivered. Always have a Plan B and be realistic with your partners about the scenarios where the delivery timeline or quality is unmet.
5. Communicate to Your Stakeholders That Build Isn’t Free
It would be best if you accounted for the opportunity costs (or actual costs) associated with:
Resources who will build the solution
Resources required to test the solution
Resources needed to provide or review documentation/data related to the solution
Resources required to train the solution and validate the outputs
Resources required to maintain the solution (technically and functionally)
6. Find a SaaS Solution Provider That Meets Your Needs
..JUUUUUST in case ;)
Always remember two things on your AI journey:
It is okay to go back on your decisions and try another route. Do not lock yourself into a solution (build or buy), and do what is in the best interest of your company and workforce.
Technology is never a Buy vs. Build decision. It is a Buy vs. Rent vs. Wait vs. Build-then-Document-then-Staff-and-then-Maintain decision.