haggl.ai Blog
Your Segment Blurb Is a Prompt. Write It for the Agent.
When you define an ICP segment in haggl, you give it a short blurb — a sentence or two describing who the segment is for. In the dashboard it sits in the segment’s intel panel and reads like a label. It’s easy to treat it as one: type something plausible during onboarding, move on to the fields that look like they matter — the margin ceiling, the priority, the checkout URL.
That’s a mistake, and a specific kind of mistake. The blurb is not a label. It is the single piece of free-form natural language in the entire protocol that you fully control, and it is read by a machine that is about to decide whether to recommend you. The blurb is a prompt. The audience is the buyer’s agent. Almost everyone writes it for the wrong reader.
Who actually reads the blurb
Trace the path. A buyer’s agent discovers your negotiate endpoint, issues a GET, and gets back your ICP segments. Your blurb is right there in the payload as each segment’s description:
"icp_segments": [
{
"id": "multi-site-smb",
"name": "Multi-site Businesses",
"description": "<-- your blurb lands here -->",
"priority": 3,
"max_margin_percent": 25
}
]No human in the negotiation loop ever sees it. The buyer doesn’t read it — their agent does, and the agent uses it for two decisions that happen before a single proof is submitted:
- Self-selection. “Does the person I represent fit this segment, or a different one?” The agent matches your blurb against what it knows about its principal and picks a
target_segment_id. - Evidence planning. “If this is the segment, what would prove it?” The agent reads the blurb, infers what evidence would score well, and goes looking for it — usually in the user’s inbox.
Both decisions are driven by the words you chose. A blurb the agent can act on produces a correctly-routed buyer carrying the right proofs. A blurb it can’t act on produces a mis-routed buyer with a thin package — or no negotiation at all.
The three jobs a blurb has to do
A blurb that works does three things at once.
- It draws the boundary. The right buyer self-selects in; the wrong buyer self-selects out. A blurb that could describe anyone routes everyone into the same segment and quietly breaks your priority logic (see Segment Priority Is a Bet).
- It names the evidence. The agent will bring whatever proof it thinks fits. If your blurb points at evidence the agent can actually fetch — ideally DKIM-verifiable receipts — you get strong packages. If it points at things the agent can’t get, you get self-reported hand-waving and low fit scores.
- It encodes the LTV thesis. The blurb should make clear why this segment earns the discount it earns — which of retention, expansion, cost efficiency, or network value it’s really about. That’s what tells the agent which dimensions to argue. (The ICP measures lifetime value, not purchase intent — here’s why that distinction matters.)
Before and after
The fastest way to see it is to rewrite a blurb from human-voice to agent-voice. Here’s a multi-site energy segment.
Written for a human:
The perfect plan for ambitious businesses ready to take control of their energy costs. Join thousands of happy customers who’ve already made the switch.
An agent can do nothing with this. “Ambitious” and “ready to take control” are conversion intent — exactly the thing the ICP is not about. There’s no boundary (who is this not for?), no evidence to gather, and no reason stated for the discount. A buyer’s agent reading this learns nothing about whether to self-select in or what to submit.
Written for the agent:
Small businesses operating 3+ metered sites under one owner. We authorize our deepest business discount here because multi-site accounts retain longest and expand as the business opens new locations. Strong fit: DKIM-verified electricity receipts across multiple sites, consistent payment history, plans to add locations. Not this segment: single-site businesses (route to Business Standard) or households (Residential).
Now the agent has a job description. It can check the site count, knows to pull receipts for each meter, understands the discount is paid back through retention and expansion, and knows when to route its principal somewhere else. Same field, same character budget — one is inert, the other steers behavior.
The pattern holds outside energy. A developer-tools SaaS segment:
Written for a human:
For high-growth engineering teams who demand the best tooling.
Written for the agent:
Engineering orgs of 50–300 seats with strong net revenue retention and multi-year budget authority. Deepest discount here because seat expansion historically outpaces the initial contract. Strong fit: DKIM-verified billing receipts from existing developer tools (proves current spend and tool maturity), multi-year procurement. Lower fit: under 20 seats or month-to-month buyers — route to the Scale-up segment.
Notice the evidence is something an agent can genuinely produce: it can’t verify “high-growth,” but it can find a year of signed receipts from the tools the company already pays for, sitting in the billing inbox.
The formula
Every blurb worth shipping answers four questions, in roughly this order:
- Who is this? The observable profile — size, structure, role. Things an agent can check, not adjectives.
- Why do we want them? The LTV reason for the discount — name the dimension (retention, expansion, cost efficiency, or network value).
- What proves it? The specific evidence the agent should bring. Favor things that can be DKIM-verified over things only the buyer can assert.
- Who is this not? The disqualifier, with a pointer to the right segment. This is what keeps the boundary sharp.
The mistakes
Four patterns we see over and over when reviewing vendor blurbs.
- Writing marketing copy. The blurb reads like a headline from the pricing page. It might be great copy — for a human. The agent extracts zero actionable signal from “the smart choice for modern teams.”
- Listing conversion signals. “Actively shopping,” “frustrated with their current provider,” “price-sensitive.” These describe intent, which the negotiation already handles — and the agent can’t prove them anyway. Worse, price-sensitivity is a churn signal, a reason to discount less, not a fit criterion (more in Your ICP Is Not About Who’s Ready to Buy).
- Being so broad everyone fits. A blurb with no disqualifier is a blurb every agent self-selects into. Now your segments overlap, your priority tiebreaker is doing all the routing, and the blurb is contributing nothing.
- Naming evidence the agent can’t get. “Must demonstrate €1M+ annual revenue” sounds rigorous, but if there’s no document an agent can fetch to show it, you’ve asked for proof that can only ever be self-reported — which scores low by design. Ask for the receipt that implies the revenue, not the revenue. (And if a segment genuinely shouldn’t need proof at all, say so — that’s no-proof mode.)
How to test a blurb
There’s a five-second test, and it’s literal: paste your blurb into any LLM and ask two questions — “Do I fit this segment?” and “What evidence would you bring to prove it?” That model is a faithful stand-in for the buyer agent that’s about to read it for real. If it can’t answer both crisply, neither can the agent negotiating against you, and you should rewrite before you ship.
You can watch the whole thing play out in the annotated transcript — the agent reads the segment, decides it fits, and assembles its proof package entirely from the words in the blurb. A sharper blurb there would have meant a stronger first package.
The point
The margin ceiling decides how much you’ll give. Priority decides who you’d rather close. The blurb decides whether the right buyer ever shows up holding the right proof — and it does that by being a prompt, read by a machine, that you happened to write during onboarding and probably haven’t looked at since.
Go read your segments’ blurbs as if you were the agent on the other side. If you can’t tell who fits and what to bring, rewrite them. It’s the cheapest change in the dashboard with the most direct line to the next negotiation.
Related Reading: