Skip to main content
Framework500 Businesses Scanned

The 6-Step Agent Journey: How AI Agents Interact With Your Business

Every AI agent follows the same 6-step journey when interacting with a business: FIND, UNDERSTAND, SIGN UP, CONNECT, USE, PAY. Most businesses fail at Step 1. Our data from scanning 500 businesses reveals exactly where each step breaks down — and what it takes to complete the full journey.

AH
AgentHermes Research
April 15, 202615 min read

The Framework: 6 Steps from Invisible to Transactional

When a user asks an AI assistant to “find me a dentist who accepts Delta Dental and can see me this week,” the agent does not just search Google. It follows a structured process to find, evaluate, and interact with businesses that match the request. We call this the Agent Journey — and understanding it is the key to understanding your Agent Readiness Score.

Each step maps to specific dimensions in the Agent Readiness Score. Failing at any step means the agent cannot proceed further — and your business loses that potential customer to a competitor who clears the step.

1FIND
2UNDERSTAND
3SIGN UP
4CONNECT
5USE
6PAY
Step 1

FIND: Discover the business exists

The agent searches for businesses that match a user request. It queries MCP registries, checks agent-card.json files, looks for llms.txt, scans OpenAPI specs, and falls back to web search if nothing structured is found.

What the agent looks for

  • agent-card.json at /.well-known/agent-card.json
  • llms.txt at the domain root
  • AGENTS.md file
  • MCP server endpoint in registry
  • OpenAPI/Swagger specification
  • Schema.org structured data on website

What most businesses get wrong

40% of businesses in our database are completely invisible to agents. They have a website, but no structured data, no agent card, no API specification — nothing an agent can discover programmatically. The agent literally does not know they exist.

40% of 500 businesses scored zero on D1 Discoverability

D1 Discoverability (weight: 0.12)
Step 2

UNDERSTAND: Learn what the business offers

Once found, the agent reads the business description, service catalog, product listings, pricing, and policies. It builds an internal model of what this business does, what it charges, and what constraints exist (hours, service area, minimums).

What the agent looks for

  • Structured product/service data via API
  • Machine-readable pricing tables
  • Business description with categories and keywords
  • Operating hours in structured format
  • Service area or delivery zones
  • Authentication requirements documented

What most businesses get wrong

Many businesses have information scattered across HTML pages with no structure. A restaurant might have a PDF menu, a dentist might list services in paragraph form, a SaaS company might say "contact sales for pricing." Agents cannot parse any of this reliably.

D6 Data Quality averages 4.2/10 across all 500 businesses

D6 Data Quality (weight: 0.10)D2 API Quality (weight: 0.15)
Step 3

SIGN UP: Create an account or get access

The agent needs credentials to interact with the business. It looks for self-service API key generation, OAuth flows, or open endpoints. If sign-up requires a human (phone call, manual approval, email verification with CAPTCHA), the agent is blocked.

What the agent looks for

  • Self-service API key generation
  • OAuth authorization endpoint
  • Open endpoints that require no auth
  • Developer portal with automated onboarding
  • Clear documentation of auth requirements
  • Free tier or trial without human approval

What most businesses get wrong

D3 Onboarding is the second-lowest scoring dimension at 0.08 weight but even lower in actual scores. Most businesses require human interaction to create an account — phone calls, email chains, or in-person visits. There is no way for an agent to self-provision access at 3 AM when a user needs help.

D3 Onboarding scored lowest of the 9 dimensions in our 500-business scan

D3 Onboarding (weight: 0.08)
Step 4

CONNECT: Establish a working connection

The agent establishes an authenticated connection to the business service. It calls the MCP server, authenticates with credentials from Step 3, and verifies it can list available tools. This is the "handshake" — confirming both sides can communicate.

What the agent looks for

  • MCP server with SSE or stdio transport
  • REST API with documented endpoints
  • Webhook support for real-time updates
  • Health check endpoint returning 200
  • API versioning for stability
  • Rate limit documentation

What most businesses get wrong

Even businesses with APIs often have unreliable connections — endpoints that timeout, inconsistent error responses, no health checks, and undocumented rate limits. An agent that gets a 500 error with no structured error message cannot diagnose or retry intelligently.

D8 Reliability averages 5.1/13 — most APIs lack health checks and structured errors

D8 Reliability (weight: 0.13)D7 Security (weight: 0.12)
Step 5

USE: Complete the user task

The agent performs the actual work — searching products, checking availability, booking appointments, placing orders. It calls the MCP tools or API endpoints with parameters from the user request and processes the responses to build an answer or complete a transaction.

What the agent looks for

  • MCP tools with clear input/output schemas
  • Search functionality with filters
  • Real-time availability or inventory data
  • Booking or ordering capability
  • Structured response formats (JSON)
  • Error handling with actionable messages

What most businesses get wrong

Most businesses that have APIs only support reading data, not taking actions. An agent can see products but cannot buy them. It can see appointment slots but cannot book one. The gap between "read-only API" and "transactional API" is the gap between a score of 40 and a score of 75.

D9 Agent Experience averages 3.8/10 — most APIs lack agent-specific UX patterns

D9 Agent Experience (weight: 0.10)D2 API Quality (weight: 0.15)
Step 6

PAY: Complete payment for the service

The agent handles payment — either through a pre-authorized payment method, a usage-based billing API, or a payment link. For per-call services, the agent may use a wallet or micropayment protocol. For subscriptions, it may provision a plan through the API.

What the agent looks for

  • Programmatic payment API (Stripe, etc.)
  • Usage-based billing endpoint
  • Machine-readable pricing tiers
  • Free tier for initial exploration
  • Payment link generation
  • x402 or micropayment protocol support

What most businesses get wrong

D4 Pricing and D5 Payment are universally weak. Even SaaS companies that charge for API access often hide pricing behind "contact sales" pages. Agents cannot negotiate or fill out forms — they need structured pricing data and programmatic payment. This is the last mile that almost nobody has solved.

D4 Pricing (0.05 weight) and D5 Payment (0.08 weight) are the weakest dimensions across all 500 businesses

D4 Pricing (weight: 0.05)D5 Payment (weight: 0.08)

How ARL Levels Map to the Journey

The 7 Agent Readiness Levels (ARL-0 through ARL-6) directly correspond to how far a business progresses through the agent journey. Each ARL level represents completing one more step.

0-19

ARL-0 Dark

Fails at Step 1 (FIND)

Completely invisible. No structured data, no API, no agent-discoverable presence.

20-34

ARL-1 Visible

Passes Step 1, fails at Step 2

Has some structured data (Schema.org, Google Business Profile) but no machine-readable service details.

35-49

ARL-2 Described

Passes Steps 1-2, fails at Step 3-4

Agent can find and understand the business, but cannot get access or establish a connection.

50-64

ARL-3 Accessible

Passes Steps 1-4 (revenue inflection)

Agent can discover, understand, authenticate, and connect. This is where real agent-driven revenue begins.

65-74

ARL-4 Functional

Passes Steps 1-5

Full read/write capabilities. Agent can complete tasks — book, order, update, cancel.

75-89

ARL-5 Transactional

Passes all 6 steps

End-to-end agent workflow including payment. The complete agent journey is functional.

90-100

ARL-6 Interoperable

All 6 steps + cross-agent protocols

Full MCP, A2A, agent-native bonus. Can interact with agents from any platform seamlessly.

The critical threshold is ARL-3. Businesses below ARL-3 cannot transact with agents. Businesses at ARL-3 and above can. Our data shows that only 52 out of 500 businesses (10.4%) have reached Silver tier (60+), which roughly corresponds to ARL-3. The other 89.6% are losing agent-driven traffic to the businesses that cleared this threshold.

Frequently Asked Questions

What is the agent journey?

The agent journey is a 6-step framework describing how AI agents interact with businesses: FIND (discover the business exists), UNDERSTAND (learn what it offers), SIGN UP (get access credentials), CONNECT (establish a working connection), USE (complete the user task), and PAY (handle payment). Each step maps to specific scoring dimensions in the Agent Readiness Score. Most businesses fail at Step 1 — they are invisible to agents.

Which step do most businesses fail at?

Step 1: FIND. Our data from scanning 500 businesses shows that 40% are completely invisible to AI agents. They have no structured data, no agent card, no API specification, and no MCP server. An agent literally cannot discover they exist. Even among businesses that pass Step 1, most fail at Step 3 (SIGN UP) because they require human interaction to create an account.

How does the agent journey relate to ARL levels?

ARL levels map directly to journey progression. ARL-0 (Dark) businesses fail at Step 1. ARL-1 (Visible) businesses pass Step 1 but fail at Step 2 or 3. ARL-2 (Described) businesses pass through Step 2. ARL-3 (Accessible) businesses complete Step 4 — this is the revenue inflection point where agents can actually connect and start using the service. ARL-4 through ARL-6 represent increasingly sophisticated completion of Steps 5 and 6.

What is the revenue inflection point?

ARL-3 (Accessible) is the revenue inflection point in the agent journey. This is when a business completes Step 4 (CONNECT) — meaning an agent can establish a working connection and start calling tools. Before ARL-3, agents can see the business but cannot transact with it. At ARL-3 and above, agents can complete tasks on behalf of users, which directly generates revenue. The jump from ARL-2 to ARL-3 is the most valuable improvement a business can make.

How can I see which steps my business passes?

Run a free Agent Readiness Scan at agenthermes.ai/audit. The scan probes all 9 dimensions and maps your results to the 6-step agent journey. You will see which steps your business currently passes, which it fails, and specific recommendations for each step. The AgentJourneyScore component shows "X of 6 steps ready" with pass, partial, or fail status for each step.


See where your business fails in the agent journey

Get your free Agent Readiness Score and see which of the 6 steps your business passes. Most businesses fail at Step 1 — find out if yours is one of them.


Share this article: