I was about to hire my first SDR after years of running OutcomeCatalyst’s sales myself. Before they started, I sat down to write the one document they needed most, a definition of our ideal customer I actually trusted, and I couldn’t do it.
I opened HubSpot and found stage names, deal amounts, and close dates. Good data, but nothing about the essence of our customers: their pain points and how they were using our AI workflows to solve them.
How could I hand him an ideal customer profile if I couldn’t pinpoint which workflows were working best for which industries?
The answer existed. It was already in HubSpot, spread across the transcripts and emails of 130+ buying conversations attached to old deals, along with the texts and voice notes that never made it in at all. None of it lived in a field I could hand to a new hire.
So I gathered it all in one place, and asked. What came back was 13 distinct AI workflows ranked by industry, with two of them making up more than a third of everything my buyers were asking for. The insight had been sitting in my own data the entire time.
Here are the steps I took to find where the data was hiding, gather it into my system of record, and ask it the right questions to truly understand my ICP. Plus, which AI workflows I discovered were working best.
The framework
Uncovering hidden data that sharpens your ICP takes four steps:
Gather: Pull every record of a buying conversation out of the places it actually lives.
Structure: Attach each one to the right contact and deal in your system of record, so the CRM finally holds what your buyers said.
Ask: Ask the right questions that your existing CRM cannot answer.
Update: Turn the answers into permanent CRM fields and backfill them.
Let’s take a look at exactly how I did each step below:
The walkthrough
Step 1: Gather
Every interaction you have with a buyer leaves a trace somewhere. Most of those interactions never make it into your CRM, but they need to if you want complete context on your customers.
For this exercise, you want to collect all of the raw data across 20 to 50 real deals, won and lost. Mix them on purpose. The difference between the two is where you’ll get your deepest insights.
Make sure you collect all of the following for each deal:
Discovery and demo call recordings and transcripts: All call notes should be included.
Every email thread across the life of the deal: Not just yours, anyone who interacted with the customer.
PDFs, proposals, and scanned documents: Both to and from the buyer, even versioned proposals so you can see the iterations.
Text messages and WhatsApp threads: Take screenshots and upload them to the contact record.
In-person meetings, coffees, and dinners: Record them with Granola or any transcription tool. If you didn’t, type notes within an hour of walking out.
Voice notes: This includes notes you recorded for yourself as well as ones you sent.
What lives only in your team’s heads: Ask them directly and record the answer.
I speed up the collection process by adding it all to the OutcomeCatalyst platform, the context layer that knows everything about my business. I’ll toss in my customer records unstructured and have it save and catalog them for me so I don’t have to waste time uploading them to specific fields in my CRM.
33%
After surveying our customers, we found that CRM data only accounts for a third of all sales data. Nearly two-thirds of it is hidden, waiting to be collected.
Step 2: Structure
Everything you gathered now needs to live on the right record in your CRM, attached to the contact and the deal it came from. Once it’s on the record, the AI can reason over all of it at once and actually make sense of the data.
Done by hand, this takes weeks, which is why it never gets done and why most CRMs stay thin. Do it for your 20 to 50 deals and stop. That is enough for the next step.
Once it’s in, connect Claude or ChatGPT to your CRM so you can ask it questions directly. Whatever AI tool touches this data should run under a business agreement with training on your data turned off, and if your conversations contain patient or client-confidential information, de-identify first or stop here.
I skip the manual work here. OutcomeCatalyst’s context layer reads everything I tossed at it and structures it into HubSpot against the right contact and deal, and I connect Claude to HubSpot directly through the platform.
Step 3: Ask
Now ask the questions your CRM could never answer on its own.
Not “what stage is this deal in,” which is a field lookup, but the ones that only resolve when the AI reads across every conversation at once and finds common threads such as simply, “How can I increase my conversion rate?”
The mistake most people make is trying to write one perfect question, one perfect answer. But it’s a conversation.
Here’s how mine went, in Claude, connected to HubSpot:
Start with the outcome: "I want to increase my conversion rate and find what's repeatable enough to package and sell across clients. What would you need to track to answer that?" That told me what to look for instead of me guessing at fields.
Then ask what’s missing: "To tell me what my buyers care about in their own words, what's missing?" It flagged deals with no transcripts, so I went back to Step 1 for those.
Then find the commonality: "Across every deal, what pain points did buyers describe and what did they ask us for, with a count and one quote each?"
Then aggregate it up: "Group those common threads into repeatable workflows I could sell as modules, and rank them." That's what came back as 13 workflows.
Then follow the thread: "Break that out by industry." Then, "Which of these do I have no field for?" Then, "Where does that information live today?"
Claude connects directly to my company brain through our platform, so I don’t have to connect individual systems myself. From there, all I need to do is fall down the rabbit hole asking questions about my ICP in the Claude interface.
Step 4: Update
The insights you get back are great, but they only become truly useful when you structure those insights into your CRM.
Before you change anything though, spot-check the output. Pick five deals you know cold, ask Claude which deals each quote came from, open those records, and confirm the quotes say what the answer claims. The AI will find patterns in noise if you let it. Whatever survives the check is a real signal.
I added a new primary workflow field in HubSpot with the 13 options and backfilled every existing client and lead. New conversations populate them automatically now from the information collected by OutcomeCatalyst. The question I couldn’t answer before my SDR started is a two-second lookup.
The outcome
13 distinct AI workflows, ranked
Each industry mapped to workflow
Mid-market operators ($20M - 200M revenue) showed up to the first call with a named workflow 72% of the time.
■ The self-audit
Write down the five questions you would want answered before your next sales hire. Now count how many your CRM can answer today.
If any of the answers would have to come from memory instead of a field, your CRM is missing the data that matters most.
Where it breaks
I see it all the time first-hand: the hardest part is actually gathering and structuring the data before you even ask it the right questions about your ICP.
This is where most companies stall. Connecting and structuring all that sales data across systems and individual conversations is a burden that halts operations.
That’s why I use OutcomeCatalyst. I toss my buyer data into the platform, and it automatically structures and saves it in my CRM. All I need to do is focus on asking my data the right questions.
Zach
Founder & CEO, OutcomeCatalyst
Book 30 minutes with me →
