---
title: "How sellers and CX at Profound use Prophet"
url: https://stacklist.com/card/4177de7e-9660-4660-8c2b-e382a7f459ef
source_url: "https://thegtmengineer.substack.com/p/how-sellers-and-cx-at-profound-use"
stack: https://stacklist.com/stack/ebe7c82a-3c5d-43a2-a3e9-ce20328a0dfc
summary: "Profound's GTM Engineering team built Prophet, an in-house revenue intelligence platform that automates deal scoring, call mapping, and account health analysis to empower sellers and customer success teams. The platform leverages AI extraction and proprietary analytics to provide predictive insights that traditional semantic search tools cannot deliver."
tags: "gtm-strategy, revenue-intelligence, sales-enablement, ai-extraction, customer-success, internal-tools"
key_entities: "Profound (organization), Prophet (technology), Edgar Sze (person), Marco (person), Matt Suri (person), Ayush Sharma (person), Salesforce (technology), MEDDPICC (concept), semantic-search (concept), revenue-cycle-management (concept)"
classification: "analysis"
content_hash: "sha256:bab9262166e52dfcc7dd30d4070170476b74dcf1b2b288912974fb3f463240f2"
acp_version: "0.2"
token_counts_approximate: 2708
visibility: public
agent_accessible: true
status: "final"
---

# How sellers and CX at Profound use Prophet

How sellers and CX at Profound use Prophet to see every deal, account, call, next step, and more in one place How Profound built its revenue intelligence platform in-house in under two months, why search was the wrong thing to build, and the always-on agent layer coming next. Edgar Sze Jul 02, 2026 1 Share I’m Edgar, and I lead GTM Engineering at Profound . Over the last two months my team built Prophet, the platform that now runs our entire revenue cycle, sales and post-sales, off our own data and puts it directly into the daily workflow of our sellers and CX team. Most things a GTM Engineer is tempted to build already exist as products, made and maintained by a company whose entire job is to build that one thing well. Buying almost always beats building. Prophet was the exception. If all we needed was semantic search on top of all of our sales and marketing data for our sellers, we would have bought a tool. But what the team needed was not search. It was a system that does the work - writing the brief, scoring the deal, mapping every call and email to the right account, recomputing health across the whole book. That cleared the bar for building on every axis. It is unique to Profound. It needs deep integration with proprietary systems, including our own analytics, which is data no vendor has and no off-the-shelf tool understands. It is a core differentiator to have reps and customer success walking into every meeting already completely prepared. We had the team to own it long-term, too, so we built it. In under two months, with a lean team. Our Staff Engineer, Marco , came up with the idea and laid down the v0 version in less than a week. Matt Suri dug into the data and built out post-sales features. Ayush Sharma built the application and pre-sales layer. We named it Prophet (fka Profit) because the point is to tell a rep or CSM what is coming before they walk in: the expansion signal, the churn risk, the question the customer is about to ask. None of this was buildable a few years ago. You could index all your text and search it by meaning, but that part is old. What is new is that a model can now read a messy call transcript and reliably pull the action items, the MEDDPICC signals (MEDDPICC is the qualification framework our reps sell with), and the commitments out of it. Plus, you can run that extraction across thousands of records cheaply enough to do it every night. Extraction and judgment were historically too hard, and that line has moved forward as AI has gotten better. Why we built it instead of buying the layer on top The obvious alternative is to leave everything in Salesforce and put a chat tool on top, the kind of semantic search layer plenty of vendors sell. We use tools like that and they are good at what they do. But that gets you a system that answers questions about your data. It does not get you a system that does the work. Prophet writes the brief, scores the deal, maps every call and email to the right account, recomputes portfolio health, and soon it will run agents against the whole book overnight. A chat layer cannot own any of that, because the part that makes it work is specific to us: our own analytics, which no vendor has, and our own definitions of what a qualified deal or an at-risk account actually means. Those definitions are the product. And the asset is not really the data, it is our living understanding of how we sell, who buys from us, and what they care about. That only exists inside of Profound. A bought tool freezes a vendor’s view of our business. A built one moves when we move. Renting that understanding inside someone else’s tool means renting our own roadmap too, so we owned it end to end instead. What it does today Start with a real morning. One of our Engagement Managers opened an account brief and Prophet had already flagged it: strong, growing usage, but no clear champion. Instead of a routine check-in, she walked in with a plan: loop in the account manager, open a line to the economic buyer, and use the usage data to make the case for expanded third-party placement. The work of knowing was done before she showed up. Here are some of our current features. The Home tab. Your calendar is wired to its Gong calls. Click a meeting and the pre-call brief is already written: prep materials, a summary of your last call, and every open action item on the deal. After the call, the post-call brief writes itself, with the wins, the friction, and what the customer committed to versus what we committed to. Every action item across every deal collects on one page, drawn from Slack, calls, and email and kept current without anyone tracking it by hand, so nothing slips. The Accounts tab. Your whole book in one place: company context, every email and Gong call, and a MEDDPICC score that fills itself in on each deal. It lines up your next meeting and surfaces intent, expansion, and risk signals per account, so you decide what to push and what to nurture across the book. The Portfolio tab. The post-sales view. Click into your account and the page recomputes instantly. Health distribution splits the accounts by churn risk, upsell score surfaces expansion, and the ARR (annual recurring revenue) timeline shows where revenue sits month to month. Every control is a toggle, so you stack them into the exact question you are asking and get a ranked, prioritized list in a few clicks. Open any account for the full read: summary, recommended next steps, key contacts, open action items, escalations and wins, the expansion and churn signals Prophet is detecting, a bear, base, and bull forecast, and a milestone track showing where each customer sits on their journey. Academy. Underneath all of it is the enablement layer. It builds its own question bank from our product docs, blog, changelog, and battle cards, generating personalized exams from each person’s weakest topics, and clustering the questions reps miss each week back into the bank. Ramping new reps gets faster, and product knowledge becomes something we can actually measure. Everything above is live. Here is what is next. The hard part of pointing an agent at thousands of accounts is consistency. Ask an agent to review a deal or qualify an account and it has to re-derive what ‘at-risk’, ‘qualified’, or ‘upsell-ready’ even means, every single time, against raw database tables. The answers drift and the cost balloons. So, we are building a semantic layer where the business concepts are defined once, in language every agent and every part of the company shares: what an account is, what a deal is, how we measure health, what an expansion signal looks like, how MEDDPICC maps to our pipeline, etc. Agents ask the layer instead of re-deriving the answer from raw tables, so they all speak the same language. Paired with vector search over our GTM corpus (our library of product docs, calls, and battle cards), it is the foundation that makes always-on agents possible. Once that exists, the question stops being whether I can effectively run an agent against one account and it becomes what I want running against all of them, every night: Deal review agents that run across every open opportunity overnight, score it, and flag the ones that slipped. Account qualification agents that requalify the whole book continuously as new signals land, instead of waiting for a human to revisit a record. Always-on enrichment that keeps the underlying data fresh, surfaces risk before it turns into churn, and flags upsell the moment the signal appears. This will batched overnight, always on, so that reps wake up to a book that requalified itself while they slept. The roadmap the foundation unlocks Forecasting mapped to how we actually sell. Not generic stage-and-close-date math, but a forecast tuned to Profound’s sales process, weighted by the signals Prophet already extracts: MEDDPICC, engagement, call sentiment, and momentum. The bear, base, and bull view becomes process-aware instead of a guess. Signal-based outbound and sequencing in each rep’s own voice. When Prophet detects intent, expansion, or risk, it drafts the outbound the moment the signal lands and writes it in the individual rep’s voice, not a generic template. Sequencing included, so the follow-through is automatic. Automated multi-threading once team mapping is in place. Multi-threading just means working more than one stakeholder in an account instead of leaning on a single contact. As Prophet maps the buying committee and the org structure inside an account, it can do this on its own, reaching the right additional people and pushing into new business units inside our enterprise accounts without a rep manually chasing the chart. Process and journey aware deliverables. Prophet knows where each deal or customer sits and drafts the right artifact for the call - discovery decks, close plans, onboarding plans, pushed across Slack and Gmail. And much much more. What we learned The cleaning layer matters more than the model. We spent more time normalizing Gong transcripts in Snowflake than we did choosing or prompting models. It was the right trade. A frontier model on dirty data loses to a smaller one on clean, structured data every time. The unglamorous pipeline work is what makes the rest trustworthy. Logic before UI. The product people see the application layer, but the thing that actually makes Prophet work is the logic that resolves every activity to the right account, and that is harder than it sounds. One company shows up under several domains and a dozen spellings, accounts have parents and subsidiaries, a Slack thread or an email has to attach to the right deal and not just the right company, and CRM data is full of duplicates and stale fields. On top of that, the pipelines have to run observably, so when a mapping silently breaks you find out before a rep does. Get any of it wrong and every brief, score, and chart downstream is wrong with it. We’re thankful to have a great RevOps and data team here. Build the foundation, not the agent. An agent is easy to stand up and easy to throw away. The leverage is in the layer underneath it, the shared definitions every agent reads from. Build that once and the agents get cheap. Skip it and you rewrite the same assumptions into every workflow until they quietly disagree with each other. Where this goes from here We are still early, and to be honest things are breaking and being fixed all the time. The agents are being built now, and we are growing the team that owns the foundation they run on: the pipeline orchestration, the GTM corpus, and the data quality underneath all of it. If you are working on this same problem, or you want to, I would love to compare notes. Also, we’re hiring. Feel free to shoot me a note! Contact: edgar.sze@tryprofound.com 1 Share Previous Next