How to Give a Sales AI Your Agency's Knowledge Per Funnel

What does "knowledge per funnel" mean in Draco?

Draco, Skyline's AI CRM, doesn't run one generic brain for your whole agency. Each funnel has its own knowledge, its own mission and its own buyer persona, and the AI answers according to the funnel the conversation is in. That means the same business can have an AI that sells adventure tours with one tone and another that sells corporate transfers with a different one, without the two blending into each other.

For an agency this solves a real problem: not every lead wants the same thing, and not every lead should be talked to the same way. A traveler asking about a diving tour in a coastal town doesn't need the same script as a company requesting a quote for event transportation. If you dump everything into a single pool of knowledge, the AI starts mixing tones, prices and promises that don't apply.

Why one funnel per product or audience, and not just one?

A funnel (pipeline) is the sequence of stages a lead goes through from the first message to the purchase, with its own handling rules. When you separate by funnel, you also separate the knowledge, the mission and the buyer persona the AI uses to decide what to say.

Concrete example: an agency in a coastal city sells boat tours to nearby islands, certified diving trips and transfers for corporate groups. If it builds a single funnel for everything, Vega — Draco's agent that handles, pre-qualifies and closes in real time — will try to sell certified diving to someone who just wants a half-day boat ride, or will speak too casually to an event coordinator. With three separate funnels, each one keeps its own knowledge and its own tone.

Splitting by funnel also helps on another front: Vega's intensity and Leo's — the agent in charge of proactive follow-ups — is set per funnel, with a hard cap of messages per reply and a fixed stance of never chasing the customer. A high-value funnel (multi-day packages) may need a different intensity than a low-ticket funnel (a single boat booking).

What goes into a funnel's knowledge?

Knowledge is the information base the AI consults to reply. It's not a full company manual; it's what a human salesperson would need to know to handle that specific funnel.

What to write

1. Exactly what that funnel sells. Names of the experiences, what each one includes, duration, meeting point, what's excluded.
2. Real objections and how to answer them. If customers always ask "what happens if it rains?", write the business's actual answer, not a generic one.
3. Cancellation and change policies, exactly as the agency applies them.
4. Differences between similar experiences. If there are two similar tours, explain when to recommend one over the other.
5. Brand tone: formal, friendly, playful, direct. Draco replies with short, warm messages in the brand's tone, so the clearer you are about that tone, the better it gets replicated.

What not to write

• No prices or availability written by hand. Draco is connected to the real Skyline Partner catalog: the AI only recommends experiences that exist, with real prices and availability, so you don't need — and shouldn't — paste a price list into the knowledge. If a price changes in the catalog and stays duplicated in the knowledge, you end up with two contradicting sources.
• No promises the business can't keep. If the knowledge says there's always availability or that the weather will be perfect, the AI will repeat it as fact, because it has no way of knowing it's an exaggeration. Write what's true, not what sounds good.
• No information from another funnel. If you mix diving knowledge with corporate transfer knowledge, the AI might offer something that doesn't belong in that conversation.
• No generic internet copy. Copying a generic tourism description doesn't help; what the AI needs is the real voice and real details of your operation.

How do you write a funnel's mission?

The mission is the goal of that specific conversation, not a financial goal for the agency. It's not "increase quarterly sales": it's something operational and concrete, like "get the lead to book the diving tour with a deposit" or "qualify whether the company needs transport for more than one vehicle before quoting."

A well-written mission tells Vega what counts as a successful close in that specific funnel. If the mission is vague, the AI doesn't know when to push toward payment and when to simply inform.

Example: in the corporate transfers funnel of a city agency, the mission could be "confirm passenger count and date, and hand Draco the quote with a payment link without insisting more than necessary." That's different from the mission of an adventure tours funnel, where the goal might be "resolve doubts about physical difficulty and close with the package that best fits the group."

How do you define the buyer persona?

The buyer persona is the profile of who that funnel talks to: who they are, what worries them, what language they use, what makes them hesitate. It's not a marketing document with abstract demographics; it's an operational description so the AI adjusts tone and arguments.

For the certified diving funnel the persona might be a somewhat experienced traveler who asks about certifications and depth. For the corporate transfers funnel, the persona is an event coordinator who needs speed, clarity on per-vehicle pricing and formal replies. The same question ("how much does it cost?") deserves a different tone depending on who's asking.

Steps to set up knowledge per funnel in Draco

Step 1: List your real business lines

Before writing anything, identify how many customer types or product types your agency handles. Don't create a funnel for every single experience; group by audience type or sale type. An agency with ten adventure tours may only need one "adventure" funnel and one "culture" funnel, not ten separate funnels.

Step 2: Write the knowledge like a salesperson, not a brochure

Write as if you were explaining to a new salesperson what they need to know on day one. Include the real frequently asked questions from that audience and the answers your team gives today.

Step 3: Define the mission in one operational sentence

Write what the AI must accomplish in that conversation: qualify, quote, close with payment, schedule a human call. Avoid abstract business goals.

Step 4: Describe the persona with their typical doubts

Note the three or four objections or questions that type of customer repeats the most. This helps the AI anticipate instead of just reacting.

Step 5: Turn on copilot mode before letting the AI run alone

Start in copilot mode, where the AI drafts the reply and a person approves or corrects it before sending, and move to autonomous when the team trusts it. This is key for testing a new funnel's knowledge without risking a real conversation.

Step 6: Review corrections and adjust the knowledge, not just the reply

Draco learns from every correction: the Approve/Correct buttons feed the AI. But if you notice you're correcting the same type of mistake repeatedly, the problem is probably in the funnel's knowledge, not in the AI. Go back to the text and clarify what's missing.

How do you test that the AI is actually using the funnel's knowledge?

Writing the knowledge isn't enough: you need to verify the AI is applying it in real conversations before trusting autonomous mode.

If the AI responds off-script in any of these tests, the first place to check is the funnel's text, not the technical setup.

What to do next

Pick the funnel in your agency with the most monthly conversations and write its knowledge, mission and persona first, following the steps above; leave it in copilot mode for a week before switching it to autonomous, and use the corrections from that period to refine the text.

More about Draco: draco.skyln.co