Intents vs Entities vs Routing: How AI Agents Really Understand Customers | L17 AI
Intents vs. Entities vs. Routing: How AI Agents Actually Understand Customers
The three-layer system that makes AI agents function — and why most “AI failures” are really design failures in these layers.
Introduction: Why This Matters
Most people interact with AI and assume it's “thinking.”
It’s not.
Not in the way humans think.
AI agents work by breaking every message into three layers of meaning:
Intent — What the user wants
Entities — The details needed to fulfill that want
Routing — What the agent should actually do
If any one of those three layers fails, the whole agent breaks.
In this article, we’ll walk through:
what each layer actually means
how they work together
real examples from real estate, development, CRE, hospitality, and service businesses
the most common errors businesses make
how L17 AI structures these layers to avoid misfires
This is one of the most important “under the hood” concepts in AI automation.
Section 1 — Intents: The Goal Behind the Message
Definition:
The intent is the user’s purpose.
It answers the question:
What is the person trying to do?
Example messages:
“Do you have any units available?”
“Can I book a tour?”
“What’s your pricing?”
“Did last quarter’s distribution go out?”
“Maintenance still hasn’t shown up.”
Each message has a totally different intent.
Why Intents Matter:
Intents determine:
the workflow
the script
the next question
the action the agent must take
the entire conversation flow
If the intent is wrong, the agent cannot do the right thing.
Intents are the foundation.
Everything else builds on it.
Section 2 — Entities: The Specific Pieces of Information Inside the Request
If an intent is the goal, then:
Entities are the ingredients.
Entities are the details the agent needs to fulfill an intent.
Examples:
User: “I want to book a tour for Tuesday at 4pm.”
Intent = Book a tour
Entities =
Day: Tuesday
Time: 4pm
User: “Can you send me the Phase 2 updated floor plans?”
Intent = Request documents
Entities =
Phase: Phase 2
Document type: Floor plans
Entities allow the agent to complete the request without asking for info the user already gave.
Section 3 — Routing: The Action Path
Routing is:
the workflow
the logic
the decision tree
the next move
It answers the question:
What does the agent do after it understands the intent and entities?
Examples:
Intent: Book a tour
Routing:
Check availability
Offer time slots
Confirm booking
Log into CRM
Send confirmation message
Another:
Intent: Report maintenance issue
Routing:
Ask for unit or room number
Ask for details
Create ticket
Escalate if urgent
Follow up
Routing is where real automation happens.
Section 4 — Why You Need All Three Layers
Imagine someone says:
“Can I take a tour tomorrow morning?”
If the agent only understands the intent…
It knows they want a tour, but not when.
If the agent only extracts the entities…
It may know “tomorrow morning,” but not whether the user wants:
a tour
pricing
availability
meeting
support
If the agent lacks routing…
It may understand everything but not know:
which calendar to check
whether to ask questions
whether to confirm
where to log the booking
Great AI requires all three working together.
Section 5 — Real examples across industries
Here’s how it works in practice.
A. Development / CRE Leasing
User: “What’s the pricing for the 2-bedroom units?”
Intent → Request pricing
Entities → Unit type = 2-bedroom
Routing → Send pricing sheet + ask if they want to book a tour + tag lead
B. Hospitality / Guest Experience
User: “Can I check in early on Thursday?”
Intent → Early check-in request
Entities → Day = Thursday
Routing → Check policy → Check availability → Confirm or decline
C. Investor Relations
User: “Send me the updated PPM for Fund II.”
Intent → Request documents
Entities → Document type = PPM, Fund = Fund II
Routing → Verify email → Send secure link → Log request
D. Service Business
User: “I need to cancel my appointment tomorrow.”
Intent → Cancel appointment
Entities → Date = tomorrow
Routing → Find appointment → Confirm cancellation → Open slot
Section 6 — Most Common Mistakes Businesses Make (And Why Agents Break)
Mistake 1 — Too many intents
If you create 80–100 intents, the agent gets confused.
The best agents usually use:
12–25 core intents.
Mistake 2 — Not combining similar intents
Example:
“Book a tour”
“Schedule a showing”
“Set a time to come by”
These must be ONE intent.
Mistake 3 — Unclear routing
If the routing is vague, the agent:
loops
makes assumptions
escalates unnecessarily
gives partial answers
Mistake 4 — Missing entities
Agents struggle when you don’t define:
dates
times
unit types
document names
property names
You must train entities purposefully.
Mistake 5 — No priority hierarchy
High-priority intents include:
emergencies
scheduling
cancellations
human escalation
billing
They must override everything else.
Mistake 6 — Poor fallback intent
Fallback determines:
“What does the agent do when it doesn’t know?”
A weak fallback → confused, repetitive agent.
Section 7 — The L17 AI Framework (How We Build Correct Intent Layers)
Here’s the actual system L17 AI uses.
1. Audit All Communication Channels
We gather:
calls
emails
DMs
website forms
past tickets
investor questions
leasing patterns
guest messages
sales inquiries
We map recurring patterns → initial intent list.
2. Consolidate and compress
We combine overlapping intents into a tight core set.
What’s left becomes:
Your master intent architecture.
3. Build the Entity Dictionary
We define entities such as:
unit types
dates
times
investor terms
documents
property/phases
locations
packages
service tiers
policies
This allows the agent to extract meaning with precision.
4. Create Routing Maps
Each intent gets:
steps
branching logic
escalation rules
fallback rules
CRM actions
confirmations
This is where the agent becomes fully operational.
5. Testing and “breaking sessions”
We intentionally try to confuse the agent with:
slang
shorthand
poor grammar
indirect questions
emotional language
long paragraphs
emojis
partial info
This forces refinements and produces a durable, high-quality system.
Section 8 — Why This Three-Layer System Is the Key to Multi-Agent Orchestration
If you want your business to run:
a voice agent
a chat agent
a follow-up agent
a workflow agent
a CRM update agent
an escalation agent
Each one must use the same underlying architecture.
Otherwise they conflict.
Example:
Voice agent: “Tour booked.”
Chat agent: “Following up with forms.”
Workflow agent: “Create task in CRM.”
They all depend on:
correct intent
correct entities
correct routing
This is why larger, more complex businesses need orchestrated systems, not “one-off bots.”
This is also the real meaning of AI orchestration (which you have a separate blog for).
Section 9 — Final Take: If You Understand Intents, Entities, and Routing, You Understand AI Behavior
This is the real foundation behind every modern AI system.
Intents define what the person wants
Entities define details needed to fulfill the request
Routing defines what the agent actually does next
Together, these three layers determine:
accuracy
reliability
user satisfaction
conversion
operational efficiency
And they form the base for:
multi-agent design
workflow automation
sales funnels
IR systems
leasing automation
guest communication
local SMB operations
If you understand these three concepts, you understand the mechanics of real AI — not hype, not buzzwords, but the actual levers that move your business.
