Andreessen Horowitz (a16z) — one of the most influential venture capital firms in technology, known for identifying structural shifts in software markets before they reach consensus — recently published an essay called “From ‘System of Record’ to ‘System of Intelligence’” [1]. The argument: in go-to-market (GTM) software, the database layer — Salesforce, HubSpot — has been the gravity well for twenty years. But the value is migrating upward, into the reasoning layer that sits above the database. The system of intelligence that orchestrates across data sources, synthesizes context, and tells the user where to focus is becoming the new hub. The database doesn’t go away. It becomes infrastructure — one of many inputs consumed at the API layer.
Their analysis is about sales software. But the structural shift they describe is not unique to sales. The same forces are at work in financial advisory — and the gap between what exists and what should exist is, if anything, wider.
This essay maps that parallel. It describes what the financial advisory landscape looks like through the lens a16z has drawn, where the analogy holds, where it breaks, and what a system of intelligence looks like when the domain is markets rather than sales pipelines.
My co-founder and I have spent 30 months building in this space [5]. We didn’t read the a16z essay and then find the parallel. We read it and recognized what we’d been building.
The advisor’s morning
Here is one way to think about how financial advisors work today.
A typical advisor managing client capital across a book of 100+ households starts her morning by checking her calendar in her CRM — Redtail or Wealthbox if she’s independent, Salesforce if she’s at a firm like Edward Jones — to see who she’s meeting today and what they need to hear. Then she opens a market data source — Morningstar, YCharts, Koyfin, or her custodian’s platform; Bloomberg [4] if she’s at a larger firm — to scan overnight moves and the economic calendar. A second system for portfolio performance — Orion, Black Diamond, or Tamarac — to see how client accounts are positioned. Maybe a news aggregator. Maybe a macro commentary she subscribes to. If she’s more tactically oriented, an order flow tool — though most advisors don’t use one.
Each system is good at its narrow job. Bloomberg shows her data. Orion shows her portfolios. Redtail shows her relationships. But none of them talks to the others. None of them connects the macroeconomic shift she just read about on Bloomberg to the microstructure behavior she might have noticed on Bookmap to the specific client portfolios in Orion that are exposed to the sector in question. That synthesis — the most important cognitive work of her day — happens entirely in her head, across five browser tabs and a cup of coffee.
She is the one connecting all of it — because nothing else does.
a16z describes the same dynamic in sales: before AI agents arrived, the account executive toggled between the CRM, email, Slack, call recordings, and enrichment tools, synthesizing context that no single system provided. The a16z essay then describes what happens when AI agents begin pulling from all of those sources simultaneously — “the CRM, the calendar, the shared inbox, the call recording, Slack, the enrichment API, the billing system, and the product telemetry” [1] — and the human is freed to do the work that actually matters.
In sales, that transition is underway. CRM usage has actually risen since AI tools arrived [3], because agents enriching the data gave reps fresh reason to consult it. But the intelligence layer — the thing that decides what the rep sees when she opens her laptop — is migrating out of Salesforce and into the reasoning layer above it.
In financial advisory, the transition hasn’t started. The advisor is still the integration layer. And the stakes of getting the daily synthesis wrong are not a stalled deal — they are client capital.
Why no incumbent will build the intelligence layer
a16z observes that Salesforce and HubSpot are responding to the intelligence-layer threat by adding AI features within their own walls — the defensive play every dominant platform owner makes [1]. Bloomberg is doing the same. So is Orion. So will Redtail and Wealthbox.
But there is a structural reason this won’t produce the intelligence layer financial advisory needs. Bloomberg’s AI will reason about Bloomberg’s data. Orion’s AI will reason about Orion’s data. Each incumbent improves its own system of record without incentive to orchestrate across the others. Bloomberg doesn’t gain by making FactSet’s data more useful. Orion doesn’t gain by incorporating order flow signals. The layer that connects macroeconomic conditions to market microstructure to regulatory filings to client portfolio exposure — the layer the advisor is currently building in her head every morning — is a structural gap that no incumbent has an incentive to fill.
There is an important distinction from the a16z analogy worth noting. Salesforce’s moat is data ownership — every call note, contact, and deal record is created inside Salesforce and lives nowhere else. Bloomberg’s moat is different. It doesn’t own market data; it licenses exchange feeds, aggregates government releases, and pulls regulatory filings. Its lock-in comes from workflow, breadth of coverage, and the messaging network that every professional in finance uses. The underlying data is largely public.
This makes the financial advisory opportunity structurally different — and in some ways stronger. In sales, the system of intelligence must read from Salesforce because the data lives nowhere else. In markets, the data that feeds the intelligence layer is accessible without the incumbent’s permission. What doesn’t exist is the synthesis — the transformation of public data into proprietary intelligence. Bloomberg aggregates and displays. The intelligence layer transforms and reasons. That distinction is the entire gap.
Intelligence that doesn’t exist yet
Here is where the analogy to sales software breaks — and where, we think, the opportunity is larger.
In the GTM context, the system of intelligence connects data that already lives in Salesforce, Gmail, Slack, and billing systems. The value comes from orchestration — pulling signals from many places and synthesizing them. The inputs exist; no one had connected them.
In markets, the most important signals are not sitting in a database waiting to be connected. Consider what it would mean to detect, in real time, statistically significant behavioral anomalies in market microstructure — patterns in order flow and related dynamics that are consistent with informed positioning occurring before public disclosure. That intelligence doesn’t live in Bloomberg. It doesn’t live in FactSet. It doesn’t exist until someone builds a system that monitors market microstructure continuously, catalogs the behavioral signatures that have historically preceded major events, and evaluates new patterns against a longitudinal record — a dataset that grows with every market cycle, preserving not just what happened but what the system detected before it happened.
Two decades of peer-reviewed research from Georgia Tech’s Scheller College of Business have documented that these patterns are real [7][8]. Dr. Suzanne S. Lee’s work, published in the Review of Financial Studies, demonstrated that information enters market microstructure before public disclosure — that the probability of sudden price discontinuities increases measurably in the window before scheduled events like earnings announcements and FOMC decisions [8]. The academic literature established that these patterns exist, that they are distinguishable from noise, and that they carry measurable significance. What the academic literature did not do is operationalize the detection in real time across a broad universe of securities and connect it to macroeconomic context.
That is what we built. We call the detection methodology Ghost Pattern Detection, and the product-facing alerts that result from it Advance Indicator Notices. When the system identifies behavioral anomalies that meet statistical thresholds, it surfaces them — not as trade recommendations, but as intelligence. The detection uses only public data.
But detection alone is half the picture. A behavioral anomaly in a single security means one thing during a period of broad economic expansion and something very different during a period of monetary tightening. The context matters as much as the signal. So the system also ingests government releases, Fed communications, Treasury publications, regulatory filings, earnings data, academic research, open-source intelligence from its own user base, and reported and non-reported news — eight data layers in total — and runs each through a proprietary economic framing layer before passing it to a correlation model that determines how the pieces connect. The output is a forward-looking market assessment: is the environment a Tailwind, a Headwind, or a Crosswind? Which way is the wind blowing?
None of this existed as a product before we built it. a16z observes that the most valuable AI-native companies are “inventing new jobs entirely — doing things that nobody was quite doing before” [1]. In financial advisory, the entire intelligence layer is a new category. No advisor has ever had a system that does what we just described. The competitive void is not a gap between incumbents. It is the absence of an entire layer.
Structured inputs, measurable outputs
a16z notes that successful AI-native companies tend to cluster around workflows where inputs are structured and outputs are measurable [1]. This is a useful filter, and it describes our system precisely.
The inputs — government data, market microstructure, regulatory filings — are all machine-readable and structured. Some are freely available through public APIs; others, particularly market microstructure data, require novel approaches to acquisition and ingestion.
The outputs are measurable in a way that is unusual for intelligence products. Twenty-eight public market assessments posted to X (formerly Twitter) before the corresponding market events — timestamped, auditable, pre-event [5]. Live proprietary trading across two leveraged funds returned 75% average ROI from January through September 2025 [5]. Past performance does not guarantee future results. But the structural point is that the system’s assessments can be scored against reality. The feedback loop is closed. This is not a black box that asks you to trust it. The record is public.
What compliance actually means here
The a16z essay mentions “handling permissions and compliance” as part of the domain-specific work between the foundation model and the customer [1]. In sales software, compliance is operational friction — SOC 2, GDPR, data handling policies. Important, but not a barrier to the category.
In financial advisory, compliance is a different animal. Registered Investment Advisors are fiduciaries regulated by the SEC. Chief Compliance Officers review technology before it touches client-facing workflows. The SEC Marketing Rule (206(4)-1) [6] governs how advisors can characterize performance. A system that produces forward-looking assessments without a verifiable reasoning trail faces a difficult path through compliance review — most CCOs will not approve a tool whose outputs cannot be audited.
General-purpose AI — ChatGPT, Perplexity, and similar tools — lacks the audit infrastructure that compliance departments require: no reasoning trail, no source attribution, no exportable documentation. Some firms use these tools internally for research, but the gap between informal internal use and compliance-approved, client-facing deployment is where adoption stalls.
This matters for the intelligence layer because it means the compliance architecture is not a feature you bolt on after the fact. It is structural. It has to be designed into the system from the beginning.
So we built a full reasoning trail that accompanies every forward-looking market assessment — showing which data sources contributed, what the correlation model connected, what historical base rates support the classification, and what Advance Indicator Notices are active. We call it our Chain of Logic. For the advisor, it answers “why should I trust this?” For the CCO, it is the audit trail that makes adoption possible. It is exportable, timestamped, and designed for the compliance file.
Any AI company can build a sales intelligence agent. Building an intelligence system that a fiduciary can legally put in front of clients requires compliance infrastructure that generic AI companies are unlikely to retrofit. That barrier does not exist in the GTM analogy. In financial advisory, it is the moat within the moat.
When the signal disappears
There is one more thing that has no parallel in sales intelligence, and it is worth describing because it illustrates what a mature intelligence system looks like.
Under normal conditions, Advance Indicator Notices appear every day across the monitored securities [5]. Most are short-lived and do not develop into sustained signals. Their presence is the baseline; their absence is not. A day where no detections appear across any monitored security is statistically anomalous. We call it a Zero-Detection Day. Historically, Zero-Detection Days have preceded high-uncertainty events within 24–72 hours [5].
This capability — recognizing when the absence of signal is the signal — is possible only because the system has built a deep enough behavioral baseline over years of continuous monitoring to know what “normal” looks like. It is not a feature someone designed. It is an emergent property of a longitudinal dataset that has accumulated over enough market cycles for absence to become informative. a16z writes about “institutional memory” becoming “something a company can actually ship” [1]. In markets, institutional memory means the system has seen enough cycles to know when the silence is louder than the noise.
What compounds
a16z argues that switching costs in the system-of-intelligence era shift from data accumulation — “all our customer data is in Salesforce” — to intelligence accumulation — “all our workflows, our reasoning, our accumulated institutional context live in our AI layer” [1].
In our system, the compounding happens along four dimensions at once.
The longitudinal dataset — what we call the Market Substrate — grows with every cycle. Every detection, every assessment, every subsequent market outcome is stored and evaluated against the full historical record. A competitor starting today starts with an empty dataset. The Substrate cannot be replicated by competitors and cannot be reverse-engineered from our outputs.
The detection methodology improves with history. Ghost Pattern Detection depends on base rates — the statistical record of what has preceded major market events. The longer the system runs, the richer the base rate, the more precise the detection.
The user base creates network effects through what we call our Open Source Intelligence (OSINT) layer. Every user who contributes an observation becomes a sensor. Anonymous correlation across participants confirms and enriches the intelligence base without revealing anyone’s data. This is the same flywheel a16z describes in GTM — usage makes the system smarter, which attracts more users, which makes the system smarter.
Example: An advisor in Texas notices several clients independently asking about energy sector exposure in the same week — unusual relative to her normal conversations. She logs the observation into Sonar. An advisor in New York, with no connection to the first, flags that his institutional clients are quietly reallocating toward energy. A third advisor notes unusual inbound interest in oil hedging from clients who have never asked about commodities before. None of them know about each other. But the system correlates these anonymous signals and recognizes a pattern: independent advisors across different geographies and client types are reporting convergent interest in energy — which either corroborates or adds context to what the Intelligence Refinery is already detecting in market microstructure and macro data. No individual advisor could see this. They each see only their own book. The system sees across all of them, anonymously. The more advisors on the platform, the more signals, the richer the correlations.
And the compliance integration creates institutional lock-in. Once an advisory firm’s compliance records reference the reasoning trails from our assessments — once the CCO has approved the system, once the audit trail is integrated into the firm’s workflow — switching to a competitor means rebuilding the compliance justification from scratch. This is the regulated-industry equivalent of the data-accumulation moat that made Salesforce sticky for twenty years.
Each dimension compounds independently. Together, they create a system that gets harder to compete with over time — the characteristic a16z identifies as the signature of durable enterprise value.
Where this goes
If a16z’s framework is right — and we think it is — then the financial advisory industry will follow a progression that looks something like this.
Today, advisors use siloed systems of record. They are the integration layer. The cognitive burden of synthesizing across five or six tools is absorbed entirely by the human.
In the near term, incumbents will add AI within their own walls. Bloomberg will add summaries. Orion will add commentary. Each improvement will be siloed within the incumbent’s data domain. This mirrors a16z’s description of Salesforce adding Agentforce [1] — a defensive play that improves the system of record without creating a system of intelligence.
Eventually, an intelligence layer will emerge that orchestrates across all of it — market data, microstructure signals, macroeconomic conditions, regulatory filings, open-source intelligence — and delivers a unified feed with compliance-grade reasoning. The advisor will open this system first. The data terminals will become inputs consumed behind the scenes. The intelligence feed — not the data terminal — will determine what the advisor sees and how she understands the market environment.
And in the longer run, the intelligence layer will create categories of work that don’t exist today. Proactive alerts when behavioral anomalies overlap with securities in a specific advisor’s client portfolios. Automated compliance documentation generated from reasoning trails. Market memory that compounds across cycles and across advisors within a firm. The advisory firm’s competitive advantage will shift from “our advisors have twenty years of experience” to “our intelligence system has evaluated every signal against every prior signal since inception, and our advisors operate on that base.”
a16z’s observation that institutional memory can become something a company ships applies with particular force to markets, where patterns repeat and the ability to recognize them early is the entire value proposition.
The convergence
We did not build our system in response to the a16z framework. We have been building it for 30 months [5], with $500K of our own capital, because we saw the same structural gap from the inside — as practitioners, not as analysts.
But reading the essay, we recognized the architecture they described. The orchestration layer that pulls from multiple systems of record. The intelligence that compounds with time. The institutional memory that persists. The new categories of work that nobody was doing before. The compliance infrastructure that sits between the model and the customer. The switching costs that shift from data to intelligence.
The same forces that are unbundling the CRM from its intelligence layer in sales software are unbundling the data terminal from the intelligence layer in financial advisory. The gap is wider, the stakes are higher, and the compliance barrier creates a moat that doesn’t exist in the GTM analogy. We think this is where a new generation of companies will be built — not above the CRM, but above the data terminal. Not orchestrating sales pipelines, but orchestrating market intelligence. Answering a different question, for a different professional, with higher consequences for getting it wrong.
The question is the same one advisors have always needed answered: which way is the wind blowing? The difference is that now, for the first time, a system can answer it — with the full reasoning trail to back it up.