Building an AI-powered CX Account Insights tool
In 2024, I released an AI-powered account insights tool that generates a strategic and insightful customer experience report for any major brand or company you provide. I couldn’t have built this without the help of several colleagues, and I’m blown away by rapid adoption and usage statistics of Sitecore colleagues.
It’s a simple report. The tool works like this: you submit a form with a brand or company name, wait for 5 minutes, and a PDF report arrives in your Outlook inbox packed with account-level insights.
Understanding the Need for CX Insights
When positioning SaaS martech software enterprise sellers focus on product features while the buyers are looking for business value and solutions to their brand’s problems. Some of my favorite Account Executives over the years worked earlier in their careers as billable consultants. They approach the sales cycle in a consultative mode that differentiates their approach. These strategic sellers fundamentally understand the importance of thoughtful discovery. They also help guide prospects through a buying journey that can be overwhelming and difficult to navigate.
In order to facilitate a productive focus interview, or discovery session, with a prospective buyer it’s essential to do your homework and spend time preparing for the meeting. I normally spend double the amount of time preparing for Discovery than the actual call, (I’ll do 2 hours of customer research for a 1hr interview).
As I research the brand and explore their website, there are a few things I try to map out in order to ask informed questions:
- Where does the brand’s revenue come from (this helps if they’re publicly traded)?
- What are the key account segments?
- What does a typical buyer journey look like (customer journey phases)?
- What are the top 2-3 key conversion points on their main website?
- How are digital marketing trends impacting this industry?
- Are their competitors innovating in any interesting ways?
- What are a few examples of engaging personalized experiences that could be deployed in days or weeks?
Answering the questions above, even in my own handwritten notebook, helps me to have a well researched and informed conversation with key stakeholders on projects or accounts. Discovery calls sometimes lead to onsite workshops where I lead marketing and IT teams through a series of exercises to develop transformational experience strategies. For most of my career, I’ve done all of this preparation and critical thinking manually but sudden advances in machine learning created opportunities to automate some of the more repetitive tasks.
The AI-powered solution I use today generates account-level insights in minutes.
View a Sample Report
Sudden Proliferation of Generative AI
OpenAI launched its ChatGPT interface and in just 2 months it reached 100 Million active users, shattering records as the fastest-adopted consumer application ever. To put this in perspective, it took Instagram over 2 years to reach the same milestone. Your colleagues and family members who aren’t exactly early adopters of technology started experimenting with a powerful Large Language Model to generate content. My neighbor is a commercial roofer and he started using it to build proposals. Microsoft backed the OpenAI tech and poured billions of dollars of funding through both cash and Azure computing provisions.
The chart below is staggering – ChatGPT outpaced all modern tech in the race to 100MM users.
I started paying serious attention to this space when Stanford Law Review shared that GPT-4 beat the bar exam. With a $20 per month subscription to OpenAI, I gained early access to the GPT-4 model and began experimenting.
Most of my marketing consulting experience has been aimed at helping industry-leading brands define their personalization roadmap, so it was familiar territory. It’s where I focused my energy on applying this emerging access to LLMs. I spent a few late nights and weekend days reverse-engineering prompts to produce similar outputs that workshop planning resulted in. Models could predictably generate strategic messaging for customer segments at each phase of the journey, and it was surprisingly accurate. The concept worked.
Working with Gene and the Ideal Customer Profile
I knew these insights could be packaged into a product and scaled to impact customers across a variety of industries, but I needed some advice on how to get started. Last year I published a YouTube video sharing my initial processes which required manual effort. The video only attracted about a hundred views. That’s when I picked up the phone and called my friend and former colleague Gene DeLibero, a former Sitecore SBOS consultant and frequent contributor to Martech.org.
Gene and I met weekly for a couple of months, and he challenged me to focus on my ideal customer profile. Several sketches later we arrived on an expanded set of individual prompts which key decision makers at either internal marketing leadership or external consulting firms would find valuable.
I remember Gene’s advice distinctly, don’t rush the technical product part… develop something disruptive and the rest will fall into place.
A Working Prototype
I had a plan but I didn’t have the time to execute.
I started spending late nights watching Zapier tutorials and watching YouTube videos about building webapps on Vercel. Client demands from my day job picked up, and there were some uniquely exciting workshops. In a six-month span, I led customer workshops onsite for brands based in Singapore, Dubai, London, San Francisco, and Sydney. If I didn’t get this out the door, someone else would solve the problem.
Pivot to Open Source
I decided to pivot completely to an open source model, and find colleagues at Sitecore that could solve the technical parts. I would share the set of prompts with Sitecore and use it as a proving ground, investing in the company I’ve been working at for the last seven years. I remember calling Gene from the Minneapolis airport before speaking at an event at Mall of America. We had been playing phone tag and I left him a voicemail explaining my decision to abandon plans to build a paid, subscription product that would be sold to other consultants.
I bet that if I shared my work with all my colleagues at Sitecore then it would build relationships and a larger impact on my career than trying to sell a product on the side. Sitecore could help me prove the benefit of these insights and promote the impact.
Sitecore was (coincidentally) organizing an AI Hackathon, and I recruited a team with Brian Bishop, a stellar Sitecore Solution Engineer-turned-seller, and Zach Escabedo, a Solution Engineering Manager with deep Content Hub expertise. Together we wrote an abstract and entered the Hackathon. I suggested the team name wAIstar Royco, as I had just finished watching the final episodes of Succession air on HBO. It was such a busy work season with workshops and speaking engagements, and we ultimately missed the deadline for submitting a Hackathon project. That didn’t stop our work. We took our time to get it right.
The prompts powering the report are now available to Sitecore colleagues on an internal channel in a Prompt Library. The Loop-powered library allows others to review and refine existing GenAI prompts, add their own, and upvote their favorites.
Configuration and Connecting the Systems
Brian Bishop single-handedly solved multiple technical aspects of this project – I owe him a beer. We named the project Sitecore CX AInsights. Zach added a layer of prompt governance in Content Hub. Together Brian and Zach connected all the dots with some innovative configuration in Sitecore’s drag-and-drop workflow automation solutions called Connect. Within weeks we had a working prototype.
Here’s what the workflow looked like at the time:
- Submit a Google Form
- Store in Google Sheets
- Sitecore Connect Recipe kicks off ~50 steps
- Cycle through set of 5 approved prompts in Content Hub
- Call GPT-4, and store the response in Content Hub
- Email the HTML back to someone form submitter
All of the information was there but we had no way to wrap everything up into a document. Brian found Encodian Flowr which he connected to Content Hub, and it compiled everything into a shareable PDF or Word document.
Encodian’s freemium model allowed for 50 responses each month without moving to a paid tier. The solution repeatedly worked and it produced surprisingly insightful results. I started using it on customer calls, and even presenting the output at industry conferences like DMFS.
Internal Launch and Measuring Success
It’s now been several months since releasing this to the wider field at Sitecore. Check out these usages statistics!
There have been more than 500 reports generated by colleagues across the globe. It’s interesting to watch colleagues from various teams leverage the report. SDRs refer back to different elements than an Account Manager cultivating interest within a longtime customer account, or Account Executives seeking to build credibility with a new logo.
What’s Next and Early Access to Reports
Today the tool is only available to colleagues with a Sitecore email address.
I’d like to introduce the tool soon to Sitecore MVPs, and previewed it at a roundtable during the recent 2024 MVP Summit.
Once we refactor CX AInsights and open it up for external use, they’ll be the first to know. You can be, too.
Please contact me if you’d like to join the waitlist.
In the meantime, if you’re interested in learning more I would be happy to join a quick screenshare call and generate a report for you. Reach out and we’ll set something up.