ICP Engineer: AI Agent for Win-Loss Analysis & Ideal Customer Profile

Install this AI agent for win-loss analysis from the ASCN marketplace and get your first ideal-customer profile in the very first session. Refer a colleague through the referral program and you both earn credits. Pay for 6 or 12 months and the subscription costs less. See current pricing at ascn.ai/pricing.

Overview

You usually know your ideal customer by gut feel. Deals close on facts. Outbound sprays a wide list, marketing brings the wrong leads, reps burn weeks on segments that convert at 12%. A consultant win/loss study runs $2,000–5,000 and takes weeks.

ICP Engineer is a ready-made AI agent for the ASCN platform and an AI agent for win-loss analysis that builds your customer profile from real deal history:

• 📊 Computes win rate by segment. Industry, company size, region, source, decision-maker role, deal size, sales cycle — where you win more than average.
• 🎯 Builds scored personas. For each core segment: pains in the customer's own words, buying triggers, objections with answers, and a 100-point scoring checklist.
• 🚫 Finds your anti-ICP. Segments where you lose consistently, with the price tag — how much team time goes to deals you were never going to win.
• 🧹 Checks data before analyzing. Gaps, duplicates, inconsistent values — it shows what it cleaned and confirms edits instead of quietly changing numbers.
• 📄 Maintains a living ICP document. Dated versions and a "what changed" delta on every update, in Google Docs, ready to hand to your team.

You run it through a normal chat in ASCN or Telegram. Nothing to assemble by hand — the agent creates the spreadsheet with the right columns and connects services when you ask.

How the win-loss analysis AI agent works

No coding. You run it by chat, and you get your first ICP the same day.

1. Install from the marketplace. Open the agent's page, click "Use template" — it lands in your account in under 5 minutes.
2. Connect services by chat. Type "Connect Google Sheets," the agent hands you an OAuth link, one click to authorize. Docs is optional, for the team document.
3. Load your deals. Send a CRM export into the spreadsheet or drop a CSV in chat — 20–30 closed deals or more. The agent flags any missing columns.
4. Clean and analyze. It normalizes the data, shows the mapping, computes win rate by segment, and marks thin slices as "low data."
5. Personas and document. You get an ICP document plus the headline in chat, e.g. "73% of wins are 10–50-person companies coming through referrals."
6. Scheduled updates. Optionally, on the 1st at 11:00 the agent checks for new deals and sends the delta to Telegram.

Why this beats hiring an analyst or consultant

• One run instead of a $2,000–5,000 invoice. An outside win/loss study costs months of subscription and takes weeks. The agent delivers the first read in one session.
• Decisions on numbers, not the founder's opinion. The agent shows the math even when the data argues with your "we sell to enterprise."
• Sharper outbound across the board. A tighter ICP lifts email and ad conversion: reps write to the people who close, not to everyone.
• Honesty over pretty conclusions. Small samples are flagged "low data," hypotheses stay separate from facts — you don't build strategy on noise.

Results users see

Across rollouts, the first run surfaces 2–3 segments where win rate is 2–3x the average, plus one anti-segment eating up to 40% of team time at under 15% conversion.

Results depend on data volume and quality: on 20–50 deals these are directional signals, not scientific significance. The agent owns the honest math; accuracy grows as you add deals.

Rollout statistics

B2B SaaS startup — logistics software

• Setup: 42 deals exported to Google Sheets, ICP document in Docs
• Result: found that 73% of wins were 10–50-person companies via referrals; enterprise ate 40% of time at 12% conversion
• Savings: dropped enterprise outbound, freed roughly 10 SDR hours a week

Performance marketing agency — B2B services

• Setup: 30 deals in Sheets, notifications to Telegram
• Result: 3 scored personas for cold email, a "hot lead" threshold on a 100-point scale
• Savings: about $3,000 on a win/loss study they no longer needed

Case figures are illustrative — an expected effect, not verified results. Your outcome depends on your niche, data quality, and setup.

What you get

• A ready-configured agent: role, analysis rules, and win/loss skills already written, live in under 5 minutes.
• A control guide: how to connect Sheets and Docs, load deals, and request a recompute by chat.
• Skills and a data schema: winloss-analysis and persona-builder plus a column schema with export steps for amoCRM, Bitrix24, HubSpot, and Pipedrive.
• A personal manager: helps tune the agent to your niche — on the Personal plan and up.
• A private community: help from other users and fast support answers.

Stop guessing who your customer is. Pull the answer from deals you've already closed. 🔥🔥🔥

Install ICP Engineer, load your deal history, and see today who to sell to first — and who to walk away from.

FAQ

Do I need coding skills?

No. The agent installs from the marketplace and runs by chat: you say what you need, and it creates the deals spreadsheet and connects services itself.

How many deals do I need?

20–30 closed deals or more, both won and lost. Fewer still works, but the agent marks the findings as preliminary and suggests filling in more history.

Does the agent connect to my CRM directly?

There are no direct CRM connectors. Data comes in through a Google Sheets export or a CSV in chat; exporting from amoCRM, Bitrix24, HubSpot, or Pipedrive takes a couple of minutes and the agent walks you through it. You can wire a CRM in via a custom MCP server if you want.

How is my data kept safe?

Connections use official OAuth, tokens live in encrypted storage and never reach chats or logs. Deal data stays in your spreadsheets — amounts and client names aren't saved to the agent's memory. You can revoke access to any service anytime.

How is an ASCN agent different from a chatbot?

A chatbot answers questions. This agent does the work: it cleans the spreadsheet, computes segments, maintains the ICP document, and checks for new deals on schedule.

What if a segment has too few deals?

The agent won't report statistics on samples under 5 deals. Those slices go into a "low data" block and the conclusion is flagged as a hypothesis to test in interviews.

Can I hand the result to another agent?

Yes. The ICP document can go to an SDR agent for outbound, which then prospects against that profile. Agents pass data through an API bridge or a shared channel.

Can I get notifications in Telegram?

Yes. Connect Telegram as a channel and the monthly data check with its change delta arrives there. The agent's logic stays the same.