We’ve all been there:
“Thank you for calling, our regular business hours are 9 to 5 Monday through Wednesday, please listen carefully as our options have changed. If you are a current customer, press 1….”
Many of you have probably memorized key sequences to speed through IVRs that you call frequently. More than a few have spent minutes shouting “AGENT” or “REPRESENTATIVE” into the void while the machine isn’t listening. Calling a business and hearing a computer answer the phone can be a frustrating experience.
But does it have to be? Not at all.
You see, your small business should implement automated inbound call handling this year. Let me tell you why.
The first automated business phone system that could answer an incoming call was created by Bell Lab's in the 1930s. it was called the "Voder" machine. Since then, we've seen many changes in the landscape of call automation. Let's review.
The earliest IVR (Interactive Voice Response) systems were based on Dual Tone Multi Frequency (DTMF) tones emitted by the keypad of home phones. You’ll remember these as “the sound the phone keys make when I press them”.
In hindsight, these systems worked well for the time. They never failed to understand your input if you typed on your keypad fast enough. The biggest downside was that you had to listen to every option and decide on your own whether it applied to you.
“I want to cancel, am I looking for Billing or Account Management?”
If you made the wrong decision, you had to wait for the human to finish their spiel about how the call was recorded and there was going to be a survey, before they told you, you’d reached the wrong person, and they’d need to transfer you. One moment was always a little longer than that.
Eventually, technology came to a place where we could convert speech to text and start to recognize the words that callers were saying on the phone. This was generally limited to just one word, or a series of numbers.
“Please briefly tell us why you are calling. You can say Billing, Customer Service or Orders.”
One thing we all know about our customers is that they never listen, so they generally respond to this with a sentence describing what they want. The system responds:
“Sorry, I didn’t get that Please briefly tell us why you are calling. You can say Billing, Customer Service or Orders.”
Before it finished speaking though, the callers spoke over it and before they knew it, they were stuck in an inbound call death spiral. Many people hung up and resolved never to call again.
Natural Language Processing (NLP) by computers was first considered by programmers in 1957, but it wasn’t until 2011 when it hit the big-time with the release of Siri by Apple. Through machine learning, it because possible to understand not just the words someone was saying, but the intent behind them.
The ability to understand that the questions “Where can I get help with my website”, “Where can I find an expert in web design” and “I need help with my site” are all the same question is easy for a human but difficult for a machine. NLP brought us tools for intent classification.
The problem with these classifiers was that the quality of the model was based on the training data. You still had to predict the intents that your users would call with, and you had to know roughly how they would structure their questions. Small changes in the training data set had big impacts on the outcomes and it was difficult to get things just right, especially if you weren’t a large company.
That brings us to today, and the promise of large language models for call automation. A Generative Pretrained Transformer (GPT) is a type of Large Language Model (LLM) based on a neural network that is trained to predict the next token (or word) based on all the previous ones. The surprising thing about a GPT model is that, when you interact with one, they give a strong illusion of humanity. This changes the whole game for call automation.
By their nature, GPT models can respond to virtually any input, continue the conversation, and appear responsive. This is a double-edged sword. A GPT model is just as happy to discuss nonsense, or earnestly invent things out of thin air, while pretending to be your honest and trustworthy companion. It wouldn’t do anyone any good to just automate a nonsense conversation from a business standpoint.
But there are two more things that the GPT model seems to be good at: classification, and structured data extraction.
Here, I Specifically mean zero-shot or few-shot classifications where you provide no, or very little, training data. What would have been a 3-monthproject a few years ago can now be done in a single prompt. The GPT model understands context, essentially by default.
Classically in a voice system if you wanted to ask for any complex object from a caller, you had to break it down:
This drags out the conversation and wastes the caller's time. There were NLP methods for handling this like entity extraction, but they were anything but fool proof. Consider the following input, which is how humans really talk on the phone:
“I’ll have a, uh, oh, pepperoni pizza and extra cheese. No, no cheese, extra olives, what’s that honey? Ok, ok and pineapple and do you want to pick it up? Yes. Ok for pickup”
Just like you understand what the caller wants intuitively and discard the side conversations and speech patterns, so too does the GPT model. It can return a perfectly formatted “PizzaOrder” object from this text, where all prior implementations would fail.
We’re closer to the holy grail.
Business phone automation is generally targeting one or both of: Operational Efficiency and Always-On Support.
Operational efficiency is the ability for an organization to reduce waste in time, effort, and materials as much as possible while still producing a high-quality service or product. Automated business call answering reduces interruptions for employees who have productive work to do. By automating responses to routine inquiries, humans are freed up to deal with the more unique and complex needs of the business.
It is difficult and expensive to staff a phone line 24x7x365.But most businesses also can’t afford to miss their customers when they reach out. Automation provides a way for a business to always answer the phone and provide some level of support, day, night, weekends, and holidays. Answering machines back in the day, and voicemail today are other examples of tools for automating always-on support. You might not be able to get through right now, but you can at least let someone know why you called.
While the benefits of phone automation are easy to see, customer service is critical to every business. So how do customers feel when they call a company and end up talking to a computer? Historically, not great. There are two main factors that drive this poor sentiment:
89% of customers who called a business in the last 6 months with an automated system reported frustration that they had to repeat information to a human, after giving it to an automated system. The most common example of this is at the bank. You enter the account number when prompted, but when the human picks up it’s the first thing they ask you again.
Many business phone automation systems are still based on the old DTMF and antiquated speech recognition technologies. While 89% of companies believe their business phone automation system understands the intent of their customers, only 54% of callers report feeling understood. These misunderstandings lead to poor customer service experiences and can cost you business.
Small businesses have special concerns around automation in general, and even more so with automation of inbound calling and IVR. Let's take a look at a few highlights.
It is expensive to employ a group of people whose only role is to accept inbound phone calls. It is also expensive to train the people inside of a call center, including a remote contact center, to communicate with your customers in your brand voice, and with your desired scripts and information. Automating inbound call handling in your small business creates efficiency. The automation follows your scripts exactly. Callers who need to speak to a human can be routed intelligently to the right person, quickly.
In most small businesses the employees and owners wear multiple hats, and this often includes fielding routine inquiries over the telephone. By automating responses to routine inquiries, your team is freed up to spend their time on growth and revenue-oriented activities instead. Even just automating informing callers of your business hours, or how to find the front door, can result in massive time savings over the span of a year.
We live in the era of immediate gratification, for better or worse. Your customers expect quick and efficient responses from your business at all hours. Automating your small business inbound phone calls helps you meet those expectations without requiring 24/7 staffing.
Further, have you called American Airlines or USAA recently? Their phone systems are amazing achievements in automated customer service. Your customers have probably spoken with a well-implemented automated voice system recently. They notice when you don’t have one, and it sets you apart when you do.
In 2010, one of the first software projects the founders of BackslashDev ever led was an effort to make a high-volume call center more efficient. We realized quickly that when a call was coming to a human, they needed to have all the information about the caller, their relationship with the company, and their last few interactions ready at their fingertips. We have built an excellent solution for this integrated directly with the business telephone system.
In 2014, we were approached by a startup in the Law Enforcement space where we first dipped our toes into inbound call automation. We worked out way up from a simple system that only understood “Yes” and “Any words that weren’t ‘yes’” to a platform that allowed subject matter experts to train complex NLP models and handle inbound calls. These systems weren’t just giving business hours. We were automating more complex tasks like taking non-emergency police reports or providing critical information in a safe, protected, and confidential manner to victims of sexual assault.
The tools we have built for automating telephone conversations have significantly improved the perceived level of service in local governments and non-profits. We’re taking everything we’ve learned and combining it with GPT models to bring automated inbound call handling to small businesses as a service. We call it FirstContact by BackslashDev.
We’ve taken everything we’ve learned and distilled it into a small business phone automation service with:
First Contact by BackslashDev is a bespoke service offering that results in software that you own. Pay once, we’ll construct your project from our toolkit, and then you own it forever. We offer three tiers of service:
Most implementations can be completed in as little as 3weeks once we accept you as a client. We do limit our implementation slots to no more than 10 projects at a time, which may decrease based on the complexity of the individual projects. We apologize if you must wait, but we want to make sure we are delivering top quality to every customer.
We’ve discussed the history of call automation, and why it matters so much, especially for small businesses. You’ve learned a little about our history working in this space and how our newest service offering could dramatically improve customer satisfaction in your business. What’s next?
If you’d like to learn more about FirstContact by BackslashDev we invite you to pre-qualify for the service by taking a short 9 question pre-qualification quiz. This will also get you onto our waiting list, and you will be notified as soon as a slot opens for your implementation. We truly cannot wait to work with you.
If you’re not quite ready to commit to a waiting list, we invite you to discover your Project Readiness Index score. In under 10 minutes, you'll receive a personalized PDF via email with insights into how your organizations positions across 7 pillars of technical project success. You’ll receive actionable insights that you can start implementing tomorrow, based on your specific needs. The Project Readiness Index is completely free, with no obligations. Try it today!
To learn more about FirstContact by BackslashDev please visit our service landing page.
3104 E Camelback Rd #2209
Phoenix, AZ 85016