We’re at the tipping point for ChatGPT and other AI-powered chatbots.
Right now most attention is focused on Microsoft and Google’s use of them. The discussion so far has been mostly about the impact on marketing and SEO. But what about the impact on martech?
Christopher Penn, co-founder and chief data scientist at TrustInsights.ai, says it is already creating big changes and opportunities.
Q: There’s been a lot of talk about how these AI chat bots will impact marketing, but what about martech? What do you think it will do there?
A: Martech vendors should be looking at it as a progress accelerator. It is a speedup tool for your dev teams. I’ve got the ability to ask a large language model for a piece of code that does X to slot into the existing code I’ve already got. This has got me through a month’s worth of development in a couple of days. I still need to be in there to fix all the weird little oddities that the engine spits out, but boy has it made me faster.
Dig deeper: How AI lets marketers create human-centric CX at scale
I actually use it more for coding than I do anything else. [ChatGPT] as it stands today generates code that is good enough for an experienced coder to look at and say, ‘OK, this gives me some good ideas’ or speeds up the development of a chunk of code. In no way can it write complete code that is ready to run out-of-the-box except for the simplest things.
I don’t see it in its current incarnation as being a replacement tool, but I do think that there will be companies where you’ll have fine-tuned models that can generate code for very specific purposes. Like, this is a library trained specifically on Python, this is a library trained specifically on etcetera.
Q: That’s great news for marketing ops. What can it do for marketers using martech?
I think it has other uses within martech, particularly more classical machine learning tasks like regression analysis for doing like lead scoring as an example. Again, these tools dramatically accelerate your progress. Or one of my favorite tasks is give it my code and have it tell me how you would make it more efficient, right? I would say seven out of 10 times it comes up with useful suggestions that work.
These models are called transformers for a reason. (Editor’s note: The GPT in ChatGPT stands for Generative Pre-trained Transformer.) They’re good at taking something and turning it into something else. They’re OK at generation, but they’re great at refining and that and that’s where I think there’s a lot of untapped power.
Q: What’s another example of that?
A: It’s good for restructuring content. As I’m driving around taking my kids to stuff, I will record voice memos and have it transcribe them. And we all know that what comes out in transcription is not print ready. But we can take that, feed it into the model and say transform this into copy appropriate for an article. Again, it preserves your voice, it preserves your facts, it preserves your point of view. But it does so from the kind of a hot mess that your transcription comes out with. So it’s a very easy way to generate a lot of content for things like newsletters or whatever. And again that’s where these tools really shine.
Martech vendors should be able to say like here’s the last 10 sales emails you sent, let’s rewrite these to be more professional. But it still preserves your voice, your point of view, the factual data that you’re that you’re incorporating. And I think that’s where martech space can see a lot of benefit from these tools.
Q: Speaking of content, when this technology answers a question it does so without linking directly to the source. This has huge implications for content marketing. How concerned should content marketers be?
A: The question of concern is going to be variable based on every individual company. Go into your Google Analytics account or your web analytics account, and look at the percentage of traffic and conversions you get from organic search. If it’s the majority channel, then you need to dig a little bit further to see how much of it is branded versus unbranded search. Branded search is probably going to be mostly OK. If someone searches for Constantine von Hoffman? Even a large language model is probably going to say, hey, this is probably the person you’re looking for. And or they’ll just know to go to your website, your URL, etcetera.
But if unbranded search is the lion’s share of your search traffic, particularly your converting search traffic, you should be very concerned. That’s where the large language models will be intercepting your traffic and not giving anything to you or giving very little to you.
I say this because the previews we’ve seen of Bing’s interface with chat, GPT and Google’s Bard interface. They do cite their sources as tiny little footnotes. I would welcome some actual research on this, but my instinct is people tend not to read footnotes very thoroughly and click (on them).
(Interview edited for length and clarity)
Get MarTech! Daily. Free. In your inbox.
