I recently stumbled over a stupid and fixable problem.
So far, everyone I talked to confirmed that they are affected too. And no one seems to have a fix.
First, some background:
I went from AI critic to AI believer in the last 5 months.
I have been critical of it for the last two years because I had a hard time seeing the real business impact. All of it seemed “nice,” but nothing would get me excited as CRO of a +10m ARR company.
No real up-lift for pipeline, no real impact on cost.
The only thing I saw was hype and lots of people’s imagination that eventually we will see all the massive improvements.
So, for the last few months, I have talked to a bunch of people about how they are actually using AI. Not some LinkedIn Echo Chamber, but actually taking the time to talk to folks.
Here is what I found:
Lots of RevOps folks are busy building flows with n8n and Zapier or squeezeing every possible use case out of Clay.
Lots of AI is used to write stuff, such as sales emails, marketing content, support conversations, and code.
AI chats used as sounding board, to ask simple questions but also to deeply discuss issues that need solving.
What stood out to me was that bucket number one really isn’t about AI. Everyone puts it in the “AI bucket” because somewhere in some of those flows you have an API call to OpenAI. And that makes it part of the “AI strategy”. Those automations are really used to build instead of buy most of the use-cases in bucket number 2.
Bucket number 2 is really about text. Taking LLMs trained on the internet’s text-data and applying it on text-heavy use-cases. Works extremely well. And while AI SDRs are having a hard time right now, this will eventually be solved.
Bucket 3 is the original ChatGPT use case, and it has strong staying power. However, this use case was trained on text data on the Internet and is mainly clueless about your company context.
That “stupid problem” I talked about?
Why are none of those use-cases including your company’s own GTM Data?
I asked that question to now 47 (and counting) RevOps, SalesOps and VP Sales/CROs.
What I got back is one common problem: our data is a mess, applying AI is just “crap in, crap out”.
And all of them are right.
Here is a version of the process all of them have used so far and failed with:
Get CRM Data to the AI
CSV download & upload
Pipeline into data warehouse with AI
Use OpenAI connector
Ask AI questions
Get absolute non-sensical results.
—> “crap in, crap out”
The simple reason why this does not work is that even the smartest AI won’t be able to guess correctly what the table and column names of your data actually mean. And which ones are important to you in your specific set-up.
And who can blame it?
It’s like taking the smartest person off the street and expecting her to immediately give you insights looking at a CSV export.
They will fail unless they get at least a bit of context or onboarding.
So, what’s the solution?
For us it started with a simple insight:
GTM Data isn’t complex because it has billions of rows of data. It’s complex because it has so many relationships:
And this is just a fraction of what is happening in your company. this Graph looks a lot messier and bigger for all of you.
What is so counterintuitive for us is that all of us understand this Graph easily. Of course, these things are connected in this way. Of course, we know there is a simple relationship.
But if you dump all this data into a CSV or Snowflake, none of these relationships will be clear to the AI. And then the AI can maybe help you construct an SQL query. But it cannot give you any meaningful insights that are not already on your sea of dashboards.
What we all really want from AI is connecting the dots across mountains of data. But with this old approach, it’s just not happening.
So, to fix that, we used a Graph Database (Neo4j) to construct what we call the GTM Graph.
When we then took that Graph and applied AI to it, we were honestly pretty floored. What we got back were true “connecting-the-dots” moments.
And sometimes, the AI went a lot farther than we expected. For example, when generating ICP logic, Churn Risk profiles, and performance management for reps, we were honestly just playing around with the tool, asking it questions that we felt it would not have any clue about.
That then led us to a very simple realization: let’s build a product that helps GTM folks leverage the frontier AI already out there.
“Screw AGI, let’s just help people apply the insane stuff that already exists. And if done right, this will probably solve 99% of AI need in the GTM.“
So far this led to these early use-cases that we were able to solve by simple prompting:
Team Performance incl. Call snippets and objection handling
Forecasting GPT - talk to your Forecast & Pipeline
Churn Risk calculation and risk signals
ICP analysis using data across the full funnel
Create Dashboards to deep-dive into an ad-hoc analysis
Ads analysis on CPC, ROAS and optimisation opportunities
Attribution calculation incl. signals that were mentioned on sales and onboarding calls
Feature request and bug fixing prioritisation by pipeline and ARR at risk
Messaging insights for marketing to learn what actually works and what not
None of these solutions is better than the dedicated and sophisticated SaaS people have been building and selling for the past decade. But all of this was literally just a prompt—by me—and cost 0.1% of the price.
We are not trying to build a unicorn. We are bootstrapped and are trying to do this without VCs. We aren’t trying to “kill” BI or “replace” Data Analysts. We simply want to build a product that people need.
We are offering a Free tier for you to try, and the first paid tier starts at $35/mo. Several people have already told us that that’s ridiculously cheap, and I agree. Check us out on attive.ai and get started without any code.
The Intelligence Layer is here
But, all of this is really just the start. We currently only support Chat and Tasks (scheduled prompts) in the publicly available app.
Prompt-to-Dashboards is in the works behind the scenes and I saw some crazy videos that were recorded internally.
Next, we will allow the app to write back to the source systems. This will be super helpful for pushing intelligence back to the system. Think of a weekly update of churn risk written back to the customer Opp or leads scores and attribution pushed back on the Lead for better routing (the list of use cases goes on, by the way.)
But eventually, we want you to prompt your way to internal GTM apps. Think about a CSM cockpit pulling data from CRM, Ticketing, Product Usage, and Call recordings. Instead of a fixed Dashboard in your BI, it's a simple app from which your CSMs can also update these systems without needing to log in to three tools.
And what used to take 2-3 data engineers in a full-blown BI setup for 3-6 months, I could do myself now with a couple of prompts.
Famously, Klarna did the above and chose to replace their CRM. Currently that doesn’t sound like the smartest way to me. But there is plenty of auxiliary SaaS that you only use 20% of because it doesn’t fit super well into your stack. That stuff probably makes sense for building yourself.
Again, our mission is not to “delete” jobs or legacy products. Our mission is to help companies use the power of AI internally—for more than just writing emails and website copy.
If you want to support the journey, check out attive.ai (and maybe sign up for a paid plan) or forward this newsletter to someone who might be interested.