Ad Image

What to Know Before You Invest in AI Text Generation

What to Know Before You Invest in AI Text Generation

What to Know Before You Invest in AI Text Generation

As part of Solutions Review’s Premium Content Series—a collection of contributed columns written by industry experts in maturing software categories—Stephen Marcinuk, the co-founder of Intelligent Relations, outlines some of the things companies should know before they start investing in AI text generation technology.

If you’re in the process of founding a company, there are two problems you must address: a lack of time and a lack of resources. You need content on your site. You need to send emails to current and prospective clients. You must get your name out in the media and build thought leadership in your field. You need to monitor your competitors. Enter one of the sexiest buzzwords to throw around lately: AI. AI has several potential applications, and that number grows every day. One of the most popular at the minute is AI text generation. 

One of the premier AI-powered text generators, GPT-3, is constantly making headlines. People are using AI to generate text about subjects as disparate as the meaning of life and a cockroach’s capacity for love. As a busy founder, you might think, “Maybe I could harness that power to get some work done around here. If AI can write about insects and philosophy, why couldn’t it generate some compelling copy about my company?”

That attitude is what a lot of content marketing platforms are banking on. More and more are adding AI text generation to their repertoire of sellable services. As a busy co-founder myself, I understand the appeal. Like any buzzy technology, artificial intelligence text generation is subject to simultaneous overestimation and underestimation. Proponents will celebrate wins like the cockroach love story.

At the same time, naysayers will point out the problems in controlling and moderating AI and its unfortunate tendency to create insensitive text. Even Meta, a company with virtually limitless resources, has struggled to build an AI-powered chatbot. So, if you’re thinking about investing in technology powered by GPT-3, or making your own AI text generation process, here are some things to consider. 

Be Specific 

As a startup founder, you probably do not have unlimited resources at your disposal. Whether you’re building AI or buying it, carving out the areas where AI text generation can help you requires scalpel-like precision. For example, our company uses AI text generation to write blocks of text for pitches to journalists. However, even writing the entire email in one go produced inconsistent results, so we needed to break the problem down further. We didn’t just unleash our AI text generator and send out whatever it came up with.  

Instead, we built it step by step. First, we got our text generator to generate personalized greetings. Then it was a customized line about the journalist’s latest coverage. Then it was a suggested outline for an article we could write for them or an expert we could connect with the journalist. We set up and rigorously tested each component before we allowed our system to draft a fully fleshed-out email.  

You’re bound to be disappointed if you expect to throw AI at a poorly-defined problem and magically solve it. But if you carefully delineate the area you’d like to automate, AI text generation can be an incredible boon. 

Examine Your Workflow 

Another fatal flaw accompanies most buzzwords: you get excited about having it before identifying where it will fit in your workflow. For example, if you’re having issues drafting website copy that will generate conversions and think buying an AI text generator will help, you will again be disappointed. You need to break tasks into manageable components, identify which parts of your workflow AI could help you optimize, and go from there. 

Look for the inefficiencies: what in your process is there a bottleneck, and what could be causing it? Could AI help? For example, in PR operations, we found an operational obstacle in writing personalized emails to journalists. We knew what we wanted to say, and generally, the pitches followed a particular format. But our writers were too busy drafting articles. Our account managers were otherwise occupied communicating with our clients. A robust PR pipeline depends on great pitches, but our people were too busy maintaining the flow to give them the attention they deserved. 

We streamlined our workflow by creating an AI text-generation process that could generate emails with a few clicks and keystrokes. It’s unlikely that any time soon, AI will be able to generate all of the long-form content and blog posts we regularly craft for clients, and it’s doubtful that AI will ever be able to manage client relationships. But we were able to apply AI to a crucial and focused function that, at this point, it would have no trouble handling.    

Map out your existing workflow. Look for areas of potential productivity enhancement or optimization. Artificial intelligence text generation will fill these gaps more effectively than subscribing to a generic email generator and hoping it will magically produce better emails. 

Know the Risks 

To illustrate this third point, let’s look at a case study of implementation gone wrong: Zillow’s attempt to use artificial intelligence to power its house-flipping division. Zillow complied with rules one and two: it identified a specific use case for AI and examined the company’s workflow to find a proper scope. The company invested heavily to develop a proprietary AI model to produce a “Zestimate,” or the site’s best guess for how much a home is worth. Zillow used this Zestimate to buy and sell houses for a profit.  

But in 2021, the AI forecasts were wildly off base. In Q2, the company sold homes for a 5.8 percent higher price than expected. But in Q3, Zillow sold them for 5 percent to 7 percent lower than expected. In Phoenix, for instance, this whiplash meant that nine out of 10 homes Zillow bought were sold at a loss. Zillow’s AI model couldn’t reliably anticipate the housing market fluctuations, and the company lost hundreds of millions in the process.  

My recommendation is to deploy your AI model where you can monitor it, as it is not a magic wand. In our case, every email draft that our AI process generates is visible to a human user before sending. The text generation model still saves us time and effort, but we’re not letting it run rampant.

AI Text Generation Has Great Potential 

Despite the limitations of AI, we haven’t yet scratched the surface of what it can do. Today, we need these extra engagement rules to ensure that AI behaves rather than runs amok. But getting in near the ground floor now ensures that we can reap the benefits as AI grows in capabilities and learns to behave. Tomorrow, AI text generation might be able to run entire email chains on its own. It might be able to use data to modulate its tone depending on who it’s emailing. As is inherent in the AI model, It could build on previous successes and failures to write ever more persuasive pitches.  

The time to learn the basics of AI text generation is now. Look for places where it can solve a real business problem for you today, and implement it accordingly. Then you’ll be perfectly positioned to take advantage of every new development and improvement in the field. 


Download Link to BPM Vendor Map

Share This

Related Posts