A Founding Document · April 2026
The CRM was built for humans. Something else needs to be built for agents. This document names the category, defines the standard, and asks who is building it.
"The future is already here. It just is not evenly distributed."
William Gibson · He was not talking about automotive retail. But he could have been.
Any buyer can now access a tool that transacts without fatigue, without emotion, and with complete data transparency. For less than the cost of a dinner they can hire one. In the time it takes to prepare a dinner they can build one.
When an agent lead enters the CRM it looks like any other lead. A salesperson picks it up. They introduce themselves. They try to build a connection. The agent is not there to be known. It has parameters and it is working them.
There is no system that identifies the agent lead at the point of entry, separates it from the human lead, and handles it differently. That system does not exist. The buyer has a machine. The dealer has a person. That is the current state of the market.
The ARM is not a detection tool. It is not a routing layer. It is not a CRM with agent-awareness bolted on.
The ARM is the system that contains, configures, and deploys the dealer's own transacting agent. When a buyer's agent enters the pipeline, the ARM identifies it, verifies it, and deploys the dealer's agent to meet it. The dealer's agent operates within parameters the dealer set in advance. Pricing floors. Margin guardrails. F&I terms. Rate floors. It knows exactly what it can offer, what it cannot offer, and when to hand off to a human. The transaction happens at machine speed. By the time a human sees the result, the deal structure is already formed or lost.
The ARM also reports on everything that happened. Closing percentage by agent type. Gross by parameter set. Which guardrails triggered most often. Where the dealer's agent lost the deal and why. The dealer can read that data, adjust the knowledge base, tighten or loosen the parameters, and send a better-configured agent into the next transaction.
The human closes the human. The agent handles the agent. Your team spends their time where it actually matters.
I am a GM at a Subaru dealership in Salt Lake City. I am not a vendor, a founder, or a builder. I am the person who watched this become necessary in real time and decided that someone needed to write the first definition before the market wrote it for us.
This document is that definition.
This is the canonical reference for what qualifies as an ARM. Use it to evaluate a product. Use it to build one.
The ARM identifies agent-originated leads at the point of entry into the dealer's lead surface. Not at the network edge. Not after the fact. At the moment the lead arrives. Detection works from two directions simultaneously. Passive surveillance reads user agent strings, session behavior, IP origin, and form submission patterns to distinguish machine traffic from human traffic. Active verification adds a lightweight proof of human layer at the lead surface. Not a CAPTCHA. Not a wall. Something conversational that a real buyer answers naturally and an agent either skips or answers in a pattern that reads as machine-generated. Passive surveillance tries to catch the agent. Active verification confirms the human. Together they make detection complete.
Scoring note: Detection must occur before lead routing, not after. A product that identifies agent leads only in retrospect scores zero. A product using only passive surveillance without any active proof of human mechanism scores partial credit.
The ARM contains and deploys the dealer's own transacting agent in response to an incoming buyer's agent. It does not route the lead to a human. It meets the agent with an agent. Detection without deployment is surveillance. Deployment is the product.
Scoring note: A system that detects agent leads but hands them to a human does not qualify as a full ARM on this criterion.
The ARM holds dealer-defined pricing floors, margin guardrails, and F&I parameters during the transaction. It does not make concessions outside those parameters without human authorization. It does not hallucinate offers. It does not capitulate under algorithmic pressure. The dealer sets the terms. The ARM enforces them. Every time.
Scoring note: Enforcement must be parameter-based, not judgment-based. A product that uses AI to decide what to offer scores lower than one that enforces explicit dealer-defined rules.
The ARM applies different logic to agent leads than to human leads. The qualification questions are different. The handoff triggers are different. The urgency thresholds are different. Human leads go to the human team. Agent leads go to the ARM.
Scoring note: A product that routes all leads through the same workflow regardless of origin scores zero on this criterion.
The ARM breaks out agent-originated activity as a separate reporting line with the same depth a CRM provides for human leads. Agent volume. Close rate. Average gross. Which parameters triggered most often. You cannot manage what you cannot measure. Agent volume is a new metric. The ARM makes it visible and actionable.
Scoring note: A product that reports agent activity only as a subset of total lead volume without independent breakout scores partial credit.
The ARM allows the dealer to configure the response logic of their own agent. Not to control the buyer's agent, but to prepare their own. The dealer cannot dictate how the buyer's agent works. What they can control is how their own agent responds. The knowledge base is updated based on what worked and what did not. The dealer's agent gets smarter with every transaction.
Scoring note: A product with fixed response logic that cannot be updated based on outcomes scores partial credit. The dealer must be able to refine their own agent over time.
The ARM connects to a trusted registry of known agent identities. It distinguishes between legitimate consumer advocacy agents, scraper bots, and malicious agents probing for exploitable pricing inconsistencies. Each category receives a different response. Not all agents are equal. The ARM knows the difference.
Scoring note: This criterion requires active maintenance of a registry or integration with a third-party verification service. Static bot detection scores partial credit.
The ARM lives inside the dealer's existing technology stack. It integrates with the DMS, the CRM, and the lead management layer. It does not require a parallel workflow outside the dealer's existing systems. Infrastructure has to live where the work happens.
Scoring note: A product that requires dealers to adopt a separate workflow outside their existing stack scores lower than one that integrates natively.
The ARM maintains a persistent record of every agent interaction, agent identity, transaction history, and outcome. When a buyer's agent returns, the ARM knows it. When a new agent arrives with a similar profile, the ARM recognizes the pattern. The CRM remembered every human interaction. The ARM remembers every agent interaction. Without this criterion the ARM starts from zero every time. That is not infrastructure.
Scoring note: A product that does not maintain persistent agent interaction history scores zero on this criterion. Memory is not optional in a relationship management system.
The ARM uses historical agent interaction data to project forward. A GM should be able to open the ARM and see not just what happened last week but what is coming next quarter. Agent traffic volume trends. Expected close rate trajectory. Gross margin impact. The CRM gave leadership visibility into the human pipeline. The ARM gives leadership visibility into the agent pipeline.
Scoring note: A product that reports historical data but provides no forward-looking projection scores partial credit. Forecasting must be derived from actual ARM interaction data.
We cannot yet detect agent traffic on our own vehicle detail pages in the wild. We know agents are visiting. We cannot see them.
We hired CarEdge to shop Mark Miller Subaru. Our store. Our inventory. Our people on the other end of the line. We pointed an AI buyer-agent at ourselves, paid for the seat, and watched the machine transact with our own sales floor. That was the one agent we invited in on purpose. Every other agent that visited our store last quarter came and went without a trace.
That experience produced this document. A dealer admitting what they cannot see has more credibility naming the solution than a vendor claiming they can already see everything. That is why I am writing this instead of waiting for someone else to write it for us.
This is a map, not a competition. It shows where the market currently stands relative to the ten criteria. It will be updated as the category forms.
Updated as the category develops · Last update April 2026This list grows as the category forms. To submit a product for evaluation, email chudson@mmsubaru.com.
I want ten dealer conversations. Thirty minutes each. What are you seeing. What are you building. Your store's name or anonymous, your call.
chudson@mmsubaru.comI already named the category. I will evaluate your product against the rubric honestly and update this site accordingly. I am not a competitor. I am the person who wants this built.
chudson@mmsubaru.comGo read Issue 06 of The Starting Line newsletter. Then hire a robot to shop your own store. The receipts will change how you see every lead that comes in next week.
The Starting Line →"The CRM was the infrastructure of the human lead era. The ARM will be the infrastructure of the agent era. The category exists. The standard is written. Now someone needs to build it."
Chris Hudson · General Manager, Mark Miller Subaru · chudson@mmsubaru.com