Generative AI in Commercial Services

In the world of innovation, every so often there is a ‘eureka’ moment that transforms the way we approach a certain sector or task. In 2007, that moment was when the first iPhone was unveiled – a groundbreaking device without a keyboard and an integrated internet browser. Yet, it wasn’t until 2008 with the arrival of the App Store that the potential of the iPhone was fully realized, heralding the birth of new services like Uber and Instagram and driving a reinvention of established platforms like Facebook.

Today, we are at the threshold of a similar revolution with generative AI. The initial version of OpenAI’s language model, ChatGPT, launched in November 2022, was groundbreaking, giving users a taste of a revolutionary technology that many argue could completely reshape sales, marketing, data-driven targeting, and data collection. Now, we can create high quality, personalized outreach messages to prospects with incredible ease and speed. But like the iPhone before the App Store, the full potential of generative AI has yet to be realized.

But the technology hasn’t been fully realized–not quite yet.  Which companies will serve which use cases? Will it really work well when rolled out across an entire company?  When the dust settles, how will it affect companies and technology is still to take shape entirely.

But we’re at a point now where we wanted to share how we are seeing generative AI taking shape in commercial services–both for our customers who have been watching this space closely and also for many of our customers who have not had the chance to pay much attention at all yet. In future posts, we will share how Convex is incorporating generative AI alongside Convex’s existing AI applications.

Despite the recent breakthroughs in generative AI technology, AI itself is not a new concept for us at Convex. For years, we have been building and deploying AI solutions in our software, leveraging an extensive dataset of hand-labeled commercial building data–we believe the world’s largest–to train our models.

In Plain Language, What Is Generative AI?

So what exactly is generative AI? It is a specific branch of artificial intelligence that specializes in creating new content. ChatGPT, one of the more recognized examples, can draft texts in response to questions, from academic essays to professional emails. (Don’t worry, this post was written the old-fashioned way).  Generative AI is not just limited to text; it can also generate images, music, and videos.

Why Are People Talking About It Now?

Generative AI has been around since the mid-2010s, and AI in general for far longer. However, recent breakthroughs have made these AIs far more powerful than before. (The “T” in ChatGPT is named for the “transformer” architecture behind this). ChatGPT was the highest profile breakthrough to be released last November, but the broader AI field has been awash in similar advances. Stunning image creation, image editing, and Freddie Mercury covers of Michael Jackson are cropping up seemingly daily.

What Does It Mean For The Commercial Services Industry?

For many industries right now, generative AI is somewhat terrifying. Computers that can draft emails, documents, perform data analysis, and soon, even make presentations are alarming to many white collar workers. Stories of work drying up are already circulating in professions like copywriting in the months since ChatGPT’s launch.

Commercial services, however, is not most industries. Robots can’t yet restore a property after a hurricane, repair a frozen sprinkler system, or eliminate a pest infestation. The trades are expected to be among the least disrupted industries by generative AI. Commercial buildings will still exist and will still need to be serviced, generative AI or not.

How Will It Affect My Job? Where Will I See It?

Even if generative AI will not dissolve the commercial services industry, for our customers who still use a computer for most of their workday, generative AI will become a part of their day in a significant way.

We expect our customers to experience generative AI in both mainstream software and in more specialized applications in different ways.

Mainstream applications will show up in software like Microsoft Office and are already being piloted today for a small number of customers.  These will be generic use cases (“help me rephrase this passage”, “please summarize this document”) but we are excited for their potential to familiarize users with generative AI’s capabilities and pave the way for IT departments’ trust.

However, we believe that the most exciting applications for commercial services will show up in more specialized applications.  This is because the more specialized the use case, the more fine tuned an AI model can be.  Rewriting an email to generically formalize it is amazing, but answering that customer service email in the first place with proprietary industry and company data is where workers really become superhuman.

At Convex, we are especially excited about this potential.  As the largest commercial services intelligence company, being able to build generative AI features that leverage our proprietary knowledge graph of properties, companies, people, and (in a secure, private way) customers’ own data today is a unique opportunity.

Over time, we believe generative AI will show up in more industry-specific applications such as field service software and quoting tools. However, these will show up more slowly because of limitations we discuss below.

What Can’t Generative AI Do?

The sheer magic of a breakthrough like generative AI can create a sense of omnipotence; what can’t it write for me? Definitely not everything. As we apply generative AI to our own software at Convex and speak to others in the industry, we use the following framework for assessing use cases:

1. Low cost of being “wrong”: Because generative AI can never guarantee 100% accuracy, it should not be used when errors can be catastrophic. Making a mistake when suggesting a daily todo list is not a big deal; having it send signable proposals with errors to a customer could be disastrous.
2. Correctable: Because generative AI is not perfect, you also want to deploy it where users can easily jump in and tweak its work. Suggesting a draft email message that can be refined is great; drafting and then sending the email sight unseen would not be the best idea.
3. Context-light: Because of the way early versions of this technology work, generative AI can typically only “look” at a small sample of your company’s work before it suggests text. Asking ChatGPT to recommend the best prospects based on looking at 20 years’ worth of sales records is simply not possible today. However, asking generative AI to consider your most recent sales outreaches and suggest the right messaging for a new prospect is perfect.

Is It Safe To Use? Will It Steal My Company’s Data?

At the time of this writing, we are seeing many of our customers’ IT departments taking a cautious approach to generative AI. Because companies have to “give” data to the AI companies in order for the features to function, IT departments are rightly asking for confidence about how their corporate data will be handled.

Having studied the architecture of these systems, Convex is confident that generative AI systems are enterprise-ready when built properly.  By enterprise ready, we mean meeting industry standards for privacy, regulatory compliance, and reliability SLAs. We believe IT departments will come to the same conclusion as they have time to study the systems as well.  We believe generative AI adoption in the enterprise is a question of when, not if.

How Does Generative AI Even Do This Under The Hood? How Is It So Good?

At its core, text-based generative AI is a prediction engine. It is predicting the most probable words to write, given what it received. It turns out that this technique of predicting the most probable next word is a remarkably effective technique.

It does not have its own critical thinking nor is it a “general artificial intelligence.” However, it can give the remarkable appearance of having so. 

Seeing an example of a model being trained can be fascinating to watch.

How Is Convex Approaching Generative AI In Its Software?

Stay tuned for the next post 😉 

We hope you enjoyed this overview of what we have observed within our customer base. Please share your feedback and comments with us directly.