Why Capturing Real-World Decisions with Decision Traces and Context Graphs Is Harder Than It Looks
A real-world look at decision traces and context graphs, based on observing how pricing, promotions, and executive decisions actually happen inside organizations.
After reading decision traces and context graphs by Jaya Gupta, I started watching how decisions are actually made and wonder if we can really capture decision traces in organizations.
As a technologist, their argument makes sense. We have systems of record. Amazons and AirBnB already algorithimized trust. The next logical step is to capture how decisions are made so agents can reason over past context, exceptions, and precedent, and eventually act with autonomy.
So instead of debating it abstractly, I started observing decisions around me.
In a B2B commerce system we built, there’s a field called “reason for discount.” On paper, this is exactly the kind of artifact a future context graph would ingest. But when I looked at real entries, they mostly read like this:
“Customer asked.”
“Strategic account.”
“CEO approved.”
None of these are false. But they don’t explain why the decision was made. They’re shorthand. Socially acceptable summaries. Not reasoning.
Then I looked at a different kind of decision. We’re hosting a sales event soon and debated what offers to present. We came up with 10–15 options. Some were AI-heavy, some leveraged existing competencies, some were consulting-led offers meant more to get us noticed than to generate immediate revenue.
The discussions were long, nuanced, and thoughtful. We even transcribed them with native Google Meet feature. But when we finally converged on two offers, I couldn’t reconstruct a clean logical trail that justified them over the others. What actually happened was more subtle: Someone senior lead the discussion and others followed that direction. Very human.
I’ve seen the same pattern with promotions. With hiring. With pricing.
Having sat in both junior rooms and executive rooms, I’ve seen this truth played out: many high-impact decisions are not driven primarily by logic. They’re shaped by incentives, relationships, emotion, fatigue, fear, trust, and power. Logic often enters later, as justification.
Which leaves me wondering.
If the most important decisions are social acts first and rational artifacts second, what does it really mean to “capture decision traces”? Are we capturing the decision, or the story we tell ourselves about the decision after it’s already been made? Can context graphs encode authority, emotion, and politics in any meaningful way? Or is there a hard ceiling to how much of this can be formalized?
I don’t have answers yet.
My thinking here will likely evolve as I observe more and discuss more.
This isn’t a pitch.
It’s me thinking in public.
You can help me think through.
This became popular on LinkedIn. You can read the replies there.
You should read their original article and tons of others that have come after. Listed below are some of them.
Under: #aieconomy , #decisionmaking