---
title: "From one analyst to a fleet of agents - Atlan Frontier Labs"
url: "https://atlan.com/frontier/essays/from-one-analyst-to-a-fleet-of-agents/"
description: "Atlan's Intel Engine wasn't built for scale. Here's how we turned it into a network of AI agents wired into real revenue workflows."
---

> Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/

An essay from Atlan Frontier Labs (Operating Notes, essay 004), "We turned our GTM intelligence from one analyst into a fleet of agents", by Mohammed Fadil, GTM Strategy, Atlan, dated 2026.07.29, 2,552 words. Atlan's Intel Engine wasn't built for scale; this is how it became a network of AI agents wired into real revenue workflows: a Deal Health Agent, a Closed Lost Agent and a Partner Win-Wire Agent. Part of [Becoming Frontier](https://atlan.com/frontier/).

## The Intel Engine: built by human, not for scale

In 2021-2022, before Atlan had the scale for quantitative analysis to mean anything, every qualitative signal mattered. The thesis: whoever learned the most from every signal would build the fastest-learning organization.

The Intel Engine was a commitment to listen to every call with the first 100 customers, run by two people including the author. Each customer averaged 20 to 30 calls of 30 to 60 minutes, so the author sat through nearly 3,000 conversations and turned that unstructured signal into insights: buying triggers, titles and keywords to target, use cases, product improvements. It produced voice-of-customer insight pages, loss-reason tags, a stakeholder database, buying triggers ranked by frequency, and a buyer-title matrix that shaped targeting. "In effect, I was the human version of what an LLM does today."

The output built the foundation of the GTM playbook:

- A 200-slide master sales playbook replacing each rep's individual narrative with one shared language.
- An 11-use-case framework that crystallized how the company created value and gave Customer Success its implementation foundation.
- The Command of the Message selling methodology, matching field communication to the language of C-suite buyers.
- A competitive program that turned scattered institutional knowledge into a shared motion, delivering a competitive win rate above 75%.

After the first 100 customers, the Intel Engine was sunsetted in its original form, intentionally.

## Rethinking revenue intelligence in an AI-native world

As Atlan scaled, leadership kept asking "which forecasted deals are truly at risk, and why?" and "why did we really lose this opportunity?" A shift in win rate set off causal debates (pipeline quality, unramped reps, channel mix, competitors catching up), each answered with an ad hoc static dashboard.

CRM fields tell you what happened (stage, amount, close date), rarely why: the pain on a call, the champion who went quiet in Slack, the objection buried in a support ticket. Most GTM truth lives in the unstructured layer (call transcripts, Slack threads, customer emails, support tickets, escalation threads), but alone it is as incomplete as CRM fields. The two had to be fused, which neither dashboards nor a shared spreadsheet could do. A figure contrasts old BI answers (dashboards and tribal knowledge) with AI-native BI (structured synthesis across structured and unstructured data) for what is happening, why, and how to improve.

Atlan began treating unstructured GTM data as a first-class asset and rebuilding the Intel Engine as a system of agents. Reasoning models like OpenAI O3 made reliable synthesis possible. The approach: start with one recurring revenue question the team struggled to answer and build the first agent around it.

## Deal Health Agent: data-driven deal forecasts

The question, from Atlan CEO Prukalpa Sankar: "What's the most data-driven way to understand the health of our active, forecasted deals?"

- Problem: deal risk surfaced reactively in 1:1s and forecast calls, when Sales was already catching up. Issues weren't documented, so deal health was narrative-driven. No tool, including Gong's native deal health monitor, stitched calls, emails, tickets and POV documents into one cross-system snapshot. Teams spent days reviewing a single forecasted opportunity.
- Solution: a weekly batch of all active forecasted opportunities, each scored by a rubric across five categories (qualification, stakeholders, commercial, competitive/product, sales execution), each rated red, yellow or green with rationale. First a shared sheet for adoption velocity, now a dashboard built in Claude Code, so any rep or leader can self-serve full deal context.
- Impact: forecast calls moved from narrative-heavy to evidence-led. In every forecast call Prukalpa asks for the link to the AI analysis and says she can't imagine a forecast call without it (a Slack message from October 2025 shows her asking for the doc). Coaching is grounded in specific dimensions like champion depth or POV execution, and account strategy is more specific. In Q4 2025, a quarter after launch, Atlan had its biggest quarter ever: beat its aggressive Q4 forecast, recorded its highest-ever competitive win rate (91%), and doubled its overall win rate from Q2 to Q4.

## Closed Lost Agent: deal loss post-mortems

Lost deals are the clearest mirror of GTM reality (unmet needs, competitor moves, execution gaps), but post-mortems are usually slow and unscalable.

- Problem: Salesforce's "Closed Lost Reason" was a picklist interpreted differently by each rep, a one-dimensional view. A real root-cause analysis took a week of piecing together emails, support tickets, Slack threads, Google Drive docs and 20-30 Gong calls, feasible for about five marquee cases per quarter.
- Solution: triggers the moment an opportunity is marked Closed Lost in Salesforce. For qualified-stage (early) losses it applies Atlan's hand-built, human-defined loss taxonomy and generates a structured loss reason. For late-stage losses it writes a two-page RCA: executive diagnosis, red-signal clusters, qualification and playbook gaps, competitive and product takeaways, and three to four concrete playbook upgrade recommendations. Loss reasons sync back to Salesforce; RCAs go out as Confluence pages, emails to the deal team and leadership, and Slack alerts within hours of the stage flip.
- Impact: capacity went from 20 deep RCAs per year (each took days or weeks) to an agent that runs end-to-end in under two minutes, more than 25x RCA throughput. Pattern analysis at a previously impossible scale; late-stage post-mortems are standard; learnings feed weekly coaching themes and competitive intelligence.

"These RCAs are good. There's always nuance, but the general learnings are a great reflection tool and foster good conversations." - Andrew E., Regional VP Sales, Atlan

## Partner Win-Wire Agent: the win-side complement

Wins with strategic partners like Snowflake, Databricks, GCP and AWS help joint customers accelerate adoption and carry lessons and people to recognize.

- Problem: partner managers pulled Gong calls, read transcripts, wrote up the use case, drafted a win-wire email and updated a partner sheet by hand, hours across dozens of opportunities. High effort, low coverage: only high-profile wins got documented.
- Solution: when a partner-tagged opportunity is marked closed-won, the agent extracts customer context, the joint use case and the impact narrative from Gong calls and internal notes, fills a structured row in Atlan's partner use case sheet, and drafts a win-wire email for the partner manager to review and send.
- Impact: Charlie Freeman, Snowflake's partner sales manager, estimates it saves him at least an hour per win-wire, reinvested in higher-leverage partner work; it matters most at end of quarter when logos close in waves. It scales within and across partners, and because it mines full Gong transcripts rather than rep memory, narratives are richer (the before/after figure goes from three bullets to the full story). With the Closed Lost Agent, it closes the loop on both closed-won and closed-lost learning.

## What we learned building GTM agents

Top-down designs would not work for a fastest-learning organization; the team built incrementally, one problem, agent and iteration at a time. Principles that emerged:

- GTM data is structured and unstructured, and neither works without the other. Salesforce fields said what stage a deal was in; calls, emails and tickets said why. Structured data alone is a dashboard nobody trusts; unstructured signal alone is noise with good stories. The leverage was agents that read both and land on one answer.
- Centralizing context means the data and its meaning, fused in one place. Gong calls, Zendesk tickets, Salesforce fields and Slack threads lived in separate systems needing live fetches. Moving data into Snowflake solved where it lived, not what it meant; something still has to define each field and how pieces connect. Only with both could agents reliably retrieve the right information.
- Domain expertise plus evals built the framework, not the model. Rubrics, exit criteria and scoring dimensions came from years of doing it manually; that was encoded into a first prompt, then sharpened with evals. Without it, an LLM invents its own generic definition of "good".
- Tie the agent to a business outcome, not an interface. The failure mode is an agent landing only on a dashboard nobody's job depends on. Deal Health became real when it changed forecast calls; Closed Lost when its RCA joined the loss debrief. "Attach agents to meetings, not dashboards: a forecast call, a 1:1." No meeting, no adoption.
- Agents surface evidence; humans make the calls. The system doesn't decide which deals to focus on, which reps need coaching or the competitive strategy; it compresses research so decision-makers work from evidence. Judgment, relationships and strategic calls stay with humans, by design.

## What's next

- Context resolution: conflicting sources need a rule, not a guess. Example: a blank Salesforce field with the answer sitting in a Gong transcript, and the agent guessing wrong. The plan: a trust rule for fields that can be arbitrated, removing those that can't, as a standing field-by-field discipline.
- Context retrieval: more context is not more accuracy. Piling every call and ticket into one window degrades accuracy before it fills (context rot). The fix: just-in-time loading, summary compaction, sub-agent isolation, retrieval by identifier.
- Skills: build the rubric once. Each rubric was a one-off prompt in one pipeline. As skills, others build on them (email filtering and ticket triage became skills, and the deal-scoring skill depends on both), and people use them directly (a BDR runs the qualification check against their own list without a new agent).
- Evals and traces: evals score each new version against known cases before it ships; traces watch production behavior and catch drift before a rep flags it.
- InsightsClaw: insights move from pull to push. Deal Health and Closed Lost signals combine into a weekly brief that reaches a rep automatically, with no meeting or dashboard; it is there when they open their laptop on Monday.

## Author

Mohammed Fadil, GTM Strategy, Atlan. Fadil ran Atlan's original Intel Engine, sitting through nearly 3,000 customer conversations, and now builds the fleet of GTM agents that replaced him as the bottleneck.

More essays: [Becoming Frontier](https://atlan.com/frontier/).