5 ChatGPT CRM Integration Takeaways
CMOs, CCOs, and Sales Managers Should Tread Lightly
Hubspot and Salesforce announced ChatGPT integrations last week, capturing headlines and interest from sales and marketing organizations worldwide. Like other ChatGPT announcements, both CRM companies promised easier content creation, from campaigns to individual sales emails. But should sales and marketing organizations be cautious?
I think so. But before we dive into the takeaways, let’s differentiate between the two implementations.
Salesforce has a much wider series of integrations coming, and will even offer companies licenses for their private implementations. This is useful, but more on that in a bit… Hubspot is a bit more cautious, unleashing alpha ChatSpot.ai personal assistant for sales team members and beta Content Assistant for more of a macro-level campaign content-building experience.
Based on just press releases alone, I feel better about Hubspot’s more reserved and customized ChatGPT approach than Salesforce’s Einstein GPT. They seem to have a more pragmatic approach to an early stage somewhat flawed technology. This belief is based on my personal hit-or-miss experiences using Chat GPT and other generative AI tools to date. Salesforce is somewhat cautious, too, only offering pilot access to these tools.
I think it’s smart to maintain a healthy skepticism toward these tools. Let’s discuss some obvious concerns.
1) Garbage In, Garbage Out
The first thing a sales or marketing leader needs to know is whether or not the CRM implementation contains accurate information about their particular industry or domain.
It doesn’t matter if you are working with GPT 3.5, GPT 4, or a competing algorithm. If the data used to train the implementation is suspect, then the outputs are going to be equally bad. This is what is known in AI circles as “garbage in, garbage out.”
In the end, GPT algorithms produce text based on source data with answers that have the greatest probability of accuracy. However, GPT algorithms do not know what is factually correct. They are not truly “intelligent.” So you get false positives and mathematical “best guesses.”
ChatGPT’s biggest criticisms and failures to date are a direct result of it sourcing inaccurate data, and not providing those sources. That’s true of any generative AI or any other AI implementation. If you put garbage in, you’re going to get garbage out.
Until sales and marketing organizations can source implementation data, train an implementation with their own customer and industry data, or choose to put into place procedures to verify outputs, they incur a significant risk of disseminating inaccurate sales and marketing content.
Most sales and marketing organizations will not be happy when they see their customers’ reactions to factually incorrect communications. Consider the reputation risk if an egregiously incorrect email or content piece is discussed on public social networks or the media.
2) Don’t Let Untrained Salespersons Use GPT Tools
ChatGPT and other generative AIs are not perfect, but they are not as bad as some writers and journalists have made them out to be. They can be formulaic, loaded with passive and unnecessary text, and tonality can be off.
That may not make a difference to an untrained salesperson who just wants to sound better than their current writing skills permit. However, therein lies the problem.
ChatGPT cut and pasted will produce misses for your sales efforts. If you don’t train salespeople to review and customize the text to ensure it reads well, then you are asking for problems. I would not assume marketing teams are better or more cautious in crafting emails or content, based on my experiences.
Human review is necessary here. If your teams are not disciplined enough to review, you may want to consider holding off on early implementations until you can implement quality controls.
The tools will improve their accuracy, writing capability, and tone with time. In the short term, know ChatGPT will make quality salespeople better, and poor salespeople worse. It’s an accelerant!
3) The More Complex the Sale, the Less Useful Text Generation May Be
Complex sales are relationship driven. Anyone who has worked within significant sales cycles, whether from direct customer interaction to passive marketing communication, knows this. Building trust is an inherent part of the game, and the most talented salespeople are gifted and active listeners who work with their customers to help them resolve challenges.
ChatGPT is not a relationship tool. It is a text generator that responds to queries. As such, understand the more personal and complex your sale is, the less effective this technology will have. Further, the risk of blowing up a relationship with unvetted machine-generated text increases!
Tech helps build relationships. It does not replace human-to-human connectivity, though. Make sure your content helps salespeople to connect rather than creating barriers.
4) If You Can Afford It, Train Your Own Implementation
In its announcement, Salesforce noted it is offering customers a feature set to “choose their own external model and use natural-language prompts directly within their Salesforce CRM to generate content that continuously adapts to changing customer information and needs in real-time.” I thought this was the most interesting aspect of last week’s announcements.
Depending on how much control your enterprise gets, the ability to train your own implementation with the private customer and industry data makes EinsteinGPT much more interesting.
Like other forms of AI, ChatGPT implementations work better when narrow or domain-specific with a particular outcome in mind. Generating text is an obvious use of this algorithm, but making sure it is tailored to produce quality text on your solution, value proposition, and how it fits within the larger context of your industry and its challenges is nirvana. Then ChatGPT becomes very useful to a salesforce that wants to use its tech stack to its advantage.
This is true for a variety of sales models, consumer or B2B. The more complex and high dollar the sale, the more useful a custom implementation would be. Fiat and Kia just announced implementations to offer self-serve sales information to customers. A more nuanced seven-figure sale could still benefit from creating complex solution-based text that is accurate, yet speaks to a specific question.
Consider your own sales cycle, the costs of Salesforce’s unique implementation, and a private data science team creating your own implementation, and weigh your options. If you can afford training in a private domain-specific implementation, there is a high probability that it will deliver a competitive advantage.
5) The Best Uses May Surprise You
I have used ChatSpot.ai as an alpha user, and it’s not bad on a rudimentary level, providing basic information on companies, developing rather mediocre Dall.E images, and summarizing technologies. Some of these tools will be useful for a salesperson that wants to query ChatSpot or EinsteinGPT or another generative tool to accomplish other tasks besides writing emails.
Some obvious uses include performing research about individual companies and industries, summarizing meeting notes, sharing out basic information, identifying keywords, starting articles with research outputs, querying basic information (a la search), and more. These basic “pre-game” activities are often overlooked despite their clear benefit in forging business relationships.
Easier basic research within your CRM’s interface can hasten and improve your sales cycle. The ChatGPT hype cycle is about writing emails and articles, the best uses are determined by users. CRM-specific ChatGPT implementations may best serve your team in unexpected ways.
Should you go full bore into the Chat CRM era? A measured approach seems smart, particularly with generative text for email writing and marketing materials. Only you know your salesforce and marketing team, but there are obvious risks that you should consider before unleashing these tools. At a minimum, quality policies and training should be weighed.
The top question when it comes to generating text is how much control does the enterprise have over a CRM GPT implementation to affect messaging and ensure non-spammy email? Not much I bet, at least on the base-level implementations.
Some CMOs and sales leaders will rush into the ChatGPT CRM era. They may be able to absorb the collateral damage, too. To what end is the question?
The industry should expect real general, formulaic emails to come from this. And I suspect we will see some very public fiascos resulting from these mistakes. Beware of the risks of moving too quickly.