This post was originally published on LinkedIn.

Attribution is probably the most hated on thing in GTM over the last decade.

The core problem is:

1. No one can validate "influence"
2. It's hard to analyze the data (it's very nuanced, time-consuming to dig in, and complex)

Marketing says "this webinar influenced 12 deals." Sales says "bullshit we sourced those deals". The CFO rolls their eyes and moves on. The marketing team gets a bit unsure and falls back on "it's directional."

"Influence" has always been implied, and until today, there was no way to validate it.

But AI's ability to work with huge amounts of unstructured data changes this meaningfully. You can now "validate" attribution.

Here's the process we're working on building out and testing:

1. Unstructured data > marketing activities/program sentiment data

All sales conversations, emails, calls, and meeting transcripts analyzed by AI to look for positive or negative sentiment on marketing programs, content, and activities.

e.g. you may find a prospect saying "oh I read that article and loved it" or "I really got value out of xyz event I just went to" or the opposite: "I had never heard of you until you cold emailed me"

1. Isolate the Positive Marketing Sentiments. And structure the data in a database of each one captured.

Date/time captured | Category | Positive/Negative | Marketing Activity Mentioned | Source (email, call, SRA)

1. Join the Positive Sentiments to the existing Attribution data (this is important)

e.g. Map the positive sentiment to the attribution data (e.g. Person X's event attendance) so you can get the connection to pipeline and revenue.

1. Mark the attribution data/touchpoint as "Validated" vs. the others "Implied"

Now you can easily report on "Validated" attribution that actually influenced pipeline vs. Implied.

You'll also have a solid database of what marketing resonates with customers.

And your CEO and CFO may actually trust that marketing is doing more than spending money.

Pre-AI there would be no way of operationalizing this at scale due to how hard it would be to parse that volume of unstructured data.

We've been handed this amazing tool (AI), let's get it to fix all the things we've been struggling with for years.

Anyone doing this or something like this already?