Using GenAI to find case study snippets (a RAG example)

This is a use case that a lot of us in sales / GTM can relate to. You have a meeting with an executive at a company or you want to send an email to an executive at a company. You want to show that you did your homework and you know that they are always interested in how you have helped similar companies / peers.

So you research the web for the executive's priorities and then you review the role / title of the executive and you do research on what is top of mind for the persona. Using this research you build a point of view on what kind of use cases and value statements would resonate with them. You want to make sure that you are showing up with proof of value. You also want to make sure you have quotes from other customers.

The process looks something like -

Few things to add here - you could research their priorities based on calls you have had in the past with other stakeholders at the company or from other internal sources. This would improve the quality of the output, but often we are stuck with only public sources. In this example I am focusing on public sources but it is easy to replace these with private sources.

This whole process can take about an hour and let's say there are 10 executives you want to do this for in a month. That works out to 10 hrs.

So the question is - what does this process look like with GenAI and does it add value. What if I were to say that you could do this in 1 minute using GenAI (after the initial setup). That is a 60X savings. There is another reality that this is such a manual and time consuming process that most reps don't do this. Even if they do it, there is tremendous variance between reps. Some do it well, most do it ok. GenAI can help take the best practices from your best reps and standardize it.

Here is a visual of the process using GenAI

Imagine, if you could just enter the name of the executive, their title and company and you got an output with their initiatives and relevant snippets? It can be done, you just need the right tools for this.

If I put this through my trusted 2x2, here are your options.

Firstly this is not a feature in any of the existing Sales tech toolset today. I can't do this in Gong or Salesforce (at least not in a way that I know of that is easy).

As far as building this using LLMs (CustomGPTs or Artifacts), it would require an enterprising person, but you would still get stuck when it comes to searching against your repository of case studies.

So, what is needed in a product to make this happen -

1. The ability to search the web to do public research on the person and their role.
2. The ability to look across a repository of case studies, find the relevant pieces from those case studies and then pull out relevant quotes, value statements etc.

So firstly, this is not as simple as just putting all the case studies and research into a prompt and have it run. There are 4-5 different prompts that need to be orchestrated to generate this output and each of them needs a different input.

Secondly to be able to search across all use cases and find the relevant pieces needs an approach called RAG (Retrieval Augmented Generation). It is a way to feed your organizations' knowledge to the LLMs to get output based on your information.

• The challenge is that to do this technically is a lot of work and requires developers. I am not a developer and most GTM folks I know are not developers.
• Secondly this is not the kind of use case where a sales leader is going to get access to a highly in-demand GenAI developer to build something like out using Langchain and Pinecone etc.
• Thirdly, you want to build it like a reusable agent so a salesperson who has no idea about prompting / GenAI etc. can just give this information and get their research brief.

There are purpose built solutions like Tribyl.com and Letter.ai that can definitely help with this use case. While their products do a lot more, this is a use case they can handle. They are both examples of GenAI first applications that support RAG natively etc. Tribyl can look through your calls and do this, Letter.ai can work on documents you upload.

The other option is to build a solution using a product like AnyQuest.ai and for the purposes of this article that is what I did. There are a few big learnings for me as I tried to do this -

1. You need to be really good at prompt engineering. To get the web search to return what you want, to get the right snippets from the case studies, to tie the output to executive priorities etc.
2. While there is a lot of literature on the challenges of RAG with big data (what if I have a million documents and I want to make this available to my customers), I decided to channel my Tableau experience to develop RAG on small data. I loved Tableau because I could use it against my excel data or CSV file and I could do it myself. That is the same with AnyQuest. I can use it to do RAG against just a folder of case studies.
3. Knowledge management is key. Most people just want to dump a gazillion docs and then do ChatGPT against it and expect it to work. The technology is not there yet to do this (there are companies working on this). You have to spend time thinking about the content you are uploading and how you organize it.
4. Web research has its challenges, it is only as good as your Google search results are. You can get old information or no information. If you have more reliable research it should be easy to plug that in as needed.
5. A lot of people are concerned about accuracy and that is totally valid. This is why I believe we need humans in the loop to review the output. The other way I look at it is - if I can give you 60X time savings and I can do 100X the work and it is as good as what an above average person can do, what is that worth to you? Would you rather, not use this and do nothing? Let's say 7 out of the 10 research briefs are good, Let's say 7 out of 10 snippets are accurate. Is that good enough? Isn't it better than current state where you are not personalizing at all?

So to put this to the test here is the scenario I chose. I am a rep from Pigment (a bunch of my Tableau friends are there) and I am selling to the new CFO at Sigma Computing (a bunch of my Tableau friends are there). Btw.. both are private companies, so there is limited public information available. Here is what I did -

1. I went to Pigment's website and downloaded 11 case studies. Their public facing case studies. This is the internal repository from which I want to pull snippets and value statements from.
2. I then built an agent that does the web research and used RAG to pull the snippets.
3. I give the agent this information - Christina Liu, Chief Financial Officer (CFO), Sigma Computing

This is the output that was generated -

Here is a screenshot..

Here is a link to a video on how I built this in AnyQuest as a non technical business savvy person without any coding. It shows that RAG at small scale in an easy to use way is accessible to people today. Start experimenting and finding use cases today. There are tons out there.