DEV Community 👩‍💻👨‍💻


Posted on

3 Crucial Components for a Killer Neural Network Model

What’s the buzz behind artificial intelligence? Tech and non-tech companies alike have invested substantial resources into incorporating AI into their product offerings and business operations. Companies have sought to use customer service bots, interactive voice response (IVR), and email automation to optimize how they run their operations. The idea of automation alone has enormous allure for any company who wants to get ahead of the competition and optimize their use of resources.
My company, Centerfield Media, is no different. As an end-to-end customer acquisition company, we always want to ensure our sales agents are placed in a position to drive revenue and conversion rate. We need systems that will give our agents the tools to increase sales as well as hold them. My manager, the lead data scientist, and I were given the task of helping create systems of success.

After much research, our team decided to build a neural network. For those curious, a neural network is a type of AI that attempts to mimic how our neurons make decisions. Our neural network would utilize speech data to generate informed agent performance scores.
As we sought to build our neural network, we experienced a lot of success and faced several hurdles. I’m going to share with you the three crucial components that contributed to our neural network model’s success. You’re going to walk away with the keys to building a killer neural network model and how you can get started with one yourself. Ready? Let’s get started.

Identify Your Business Goal

There’s a saying in tech: “just because you can, doesn’t mean you should”. Techies and engineers alike use this philosophy when considering new technologies for their respective companies. New software, tools, and technologies are great if they bring legitimate value to your business. But if you’re using these things because it’s a cool thing to do, you’re wasting valuable resources. Neural networks fit this analogy perfectly. Many companies want to build neural networks because they know it will optimize their operations or increase their revenue. But many other companies want to build neural networks without understanding what exactly it will do for their organization.

Our data science team had a simple and concise goal: create a solution that automated agent scoring with accuracy and precision. Our needs were highly nuanced and specific, so we knew a generic solution wasn’t going to help us-we needed a custom and streamlined way to score our agents’ performance.

I recommend you figure out what your business goal and how a neural network model will help you. Feel free to explore and create proof of concepts. Enjoy the process of learning. But you should arrive at a point where you know exactly what you’re trying to accomplish and what you expect the neural network to produce.

Get a Hold of a Large Sample Size of Data

When building a neural network, data will be your best friend. Let’s take a few steps back and consider how a neural network functions. Remember when I mentioned that a neural network attempts to mimic how our neurons make decisions? Our brains try to tap into as many past experiences as possible to be able to make an informed decision. Take stopping at a stop sign for instance. As you consider whether you should stop, your brain runs through all the times you got away with running a stop sign and all the times you didn’t. Neural networks work in a very similar manner except you’re the one feeding it data. A neural network requires you to feed it data or “train” it for it to make as much of an informed decision as possible.

For our neural network model, we needed to feed it as much clean data as possible. We fed it hundreds of speech transcripts that resulted in millions of data inputs. Word count, agitation, call duration, you name it. These data inputs would train our neural network to operate within a clear and defined pattern. In the end, our model graded a test group of agents on their compliance. The model’s grades matched over 90% of the grades that our quality assurance team manually performed.

The more data you can feed your model, the more accurate and holistic it will be with its results. I strongly recommend you find a clean data source with as much variability as possible. Over time, your model will learn to anticipate and recognize patterns. The result? A more informed and accurate model that will only increase efficiency for your company.

Set Aside Ample Training Time

Remember when I said data will be your best friend? Time will be your second best friend. A neural network’s value only increases over time. Why? Time means greater variability in your data. If you feed your neural network a million data inputs from last month, you’re only getting a month’s view of variability. But if you fed your model a year’s worth of data inputs, you have access to a greater breadth of data. A great neural network will have the ability to make second and third order level decisions. The only way a model can perform those types of decisions is if it has the right data. If you want the right data, you need more time.
We learned this lesson the hard way. The Application Programming Interface (API) in which we receive our data limits us to the last three months. In an ideal and perfect world, we would have years and years of data. But we adjusted and kept building. We found ways to access additional months of data and trained our neural network with that data.

Sometimes you’re in a crunch for time to produce results. But if you have a choice between taking more time to train your model or not, opt to take more time. Your neural network will only benefit from a greater breadth and variability of data.

In the End, It’s All About the Data

Neural networks, as well as the greater suite of AI technologies, are incredible to work with. We’ve only scratched the surface for what humans can leverage AI to do. Companies are continuing to find new and innovative ways to integrate AI. Chat bots, IVR, and email automation are becoming increasingly common.

Don’t know where to start with building a neural network? Identify your business goal. Once you’ve found a worthy business goal that will add massive value, locate where your data lives. Make sure the data is clean, usable, and accessible. Finally, determine your threshold for time. Are you on a deadline? Does your model’s operational excellence need to outweigh its accuracy? If all else equal, take more time.

You now have the three crucial components to building a killer neural network model-don’t let it go to waste. You can start building now and possibilities and applications are endless. You just have to find a reason to build.
P.S. If you like this type of content, subscribe to my newsletter here. I want to help people go from taking a coding course to landing a job offer.  

Top comments (1)

seanthegreat profile image
Sean Antony Brunton

Very informative, and practically so. Thank you! Happy Hacking! \"/

🌚 Friends don't let friends browse without dark mode.

Sorry, it's true.