Some more AI tools for CMOs
Scraping, analysis, enterprise prompt management, and outreach
Follow-up from yesterday’s post: Bryan Caplan tested GPT-4 on his new undergrad exam. This was his first time teaching the exam, so no prior data on the topic exists on the internet. GPT-4 not only passed, it got the highest score in the class on blind testing.
Onto today’s update:
On March 24th I wrote my take on how I would be using AI right now if I was a CMO. In the last 10 days I have found a few more tools that I would incorporate into my regular business.
Scraping
There have been many, many times in my career when I have needed to scrape websites. Sometimes this is to know the up to date pricing of competitors. At GA we wanted to know job listing information for our regional SEO pages. At APFM we needed local map information for our senior housing listings (i.e., how far away was the closest park, church , grocery store, etc.). These projects were always doable, but a pain. They either meant taking up precious internal dev resources, or trying to get marketing managers to work with (often sketchy!) external scraping “companies”.
Now there is a better (AI) solution.
Kadoa uses LLM AI’s to “scrape” websites. The LLM are “smart” enough that they can effectively read websites and report back on what they find in a structured format. Effectively it is an API for websites without APIs. Their enterprise solution is $300/month, which is very reasonable if you need something (or many things!) continually updated. Their smaller-scale version for one time uses is free (but not yet available).
Analysis
This guy built a “Warren Buffet” chatbot and had it read 1000 PDF pages Tesla's 2020-2022 10-k annual reports. Clearly most CMOS will not need this particular use case, but a tool that reads massive amounts of new information every day and reports back with summaries of what is important - and can be questioned and grilled when needed, could be very valuable for many uses. Most CMOs don’t have time to read the latest content about their industry, but if it could be collected in PDF form and summarized with alerts for what is really important, it feels like it would be worth doing.
Custom Prompts
Bearly is an enterprise AI tool I am exploring. I believe one of the bigger challenges of getting an organization to use AI is getting people to change the way they work. Bearly has tools that help companies do just that. One big one is the ability to save and share and standardize custom prompts. Getting prompts right is becoming less and less important as the LLM models become more robust, but it is not yet “unimportant”. but once you figure out the right prompt to create the right type of content for your organization, you don’t want every individual and group to have to re-invent the wheel. Bearly, or a tool like it, will be essential in getting larger organizations to use LLMs for content creation.
Outreach
I don’t know if what this guy did should be called sales or market research, but it is pretty ridiculous. I am pretty sure it is exaggerated as GPT-4 is (I think) way too slow to pull this off with the timing he claims. But even if it is a little slow, everyone should be doing this.
The “guy” was a app developer based in Pakistan. He pulled a list of 100,000 Shopify stores. Then he set GPT-4 to work. He had the AI “read” the websites of all the merchants, and then:
Come up with an idea for an app that would generate more money for that specific store
Write a pitch for the (made up) app
Send an email to the merchant with the pitch
If the merchant responds, GPT-4 writes back, and sends calendly links to lock in meetings
The developer claims that his normal emails get <0.01% response rates (which is not surprising). He does not say what response rate he is getting, but he claims that when he sent out 3000 messages he was “slammed” with responses. The link has a number of screen shots of the replies.
Most businesses are not app developers selling to Shopify merchants, but it does not take a huge imagination to see how something like this should be used by any B2B company.
Bulk Processing
The easiest way to access GPT-4 is to pay $20/month for a subscription to ChatGPT and limit yourself to 25 queries every 3 hours (or whatever the current throttle is). That is not going to let you do things like the mass outreach described above. Once you find something that works you are going to want to scale it. At this point you can talk to your development team, or you can use something like this tool. ParallelGPT allows anyone with an OpenAI API code to parallel process mass numbers of GPT queries. It is a step between the chat interface and building a product and tapping into the API.
More to come.
Keep it simple,
Edward