Praveen Sellaperumal Kalaiyarasi

Interviewed consultants and industry experts to derive real pain points and the language MSOs actually use, then drafted the core pitch from that research. Mined keywords from a handful of known ICP accounts, then built a SERP-based discovery agent to scour the web for additional MSO accounts — standard ABM account-search tools didn’t surface this segment well. Everything the agent returned was verified by hand before it earned a place on the list, which is why the final target set was 25 accounts rather than a few hundred: at this deal size a wrong account isn’t just wasted send volume, it burns the one credible approach you get to an operator. Built an AI agent to personalize messaging at both the individual and account level, then automated email sending plus LinkedIn connection requests and messaging on top of it.

Six of the 25 accounts became live leads and the first positive reply converted into a $2.5M deal — roughly $100K of closed revenue for every account on the list. The win became social proof for a second outreach wave into adjacent operator networks, opening an estimated $15M in expansion potential over three years. The lesson that transferred to every campaign after it: at enterprise deal sizes, list accuracy outperforms list size by a wide margin.

No single source came close to covering this market, so the build ran several in parallel and reconciled them against each other. Sourced and enriched accounts through ZoomInfo, Apollo, a custom-built Python scraper, and SERP-based discovery agents for accounts not surfaced by traditional databases, plus direct Google Maps scraping for local and independent clinics — the segment commercial providers cover worst. Enrichment went beyond standard firmographics — current EMR, insurance provider network, number of locations, org type, specialty, insurance details, Google reviews, and recent job changes, all captured per account.

Grew the account base 5× and the contact base roughly 4.7× from the inherited starting point, reaching an estimated 66% of TAM — turning a thin inherited list into the primary account universe the rest of the GTM stack (targeting, routing, campaigns) runs on.

Deliberately not a static calculator — each pricing tier is a real decision the prospect makes, so by the end they’ve configured their own deal and the sales team knows exactly which one. Selecting one of the major competitor EMRs auto-estimates current spend from an internal pricing-intelligence bank — totals only, rate cards never exposed. Unlisted EMRs capture the vendor name plus self-reported costs, growing the pricing data bank with every visit, with an industry benchmark as fallback.

Software compared against software; full cost of ownership against the bundle — every number has to survive a live deal conversation. The bundle’s percentage fee applies only to insurance collections, never cash or patient responsibility, so the page shows the true blended rate on total collections: a materially better headline, told honestly. Clinics still on paper get no false comparison at all — for them the page prices going digital and leads with the upside.

Verdicts, stage summaries, and the final pitch adapt to the prospect’s stated goals — including a 5–8× EBITDA enterprise-value framing for owners planning a PE exit. Twenty verbatim five-star review quotes are selected against those same goals, feeding a “talk to a live clinic owner” reference capture instead of a generic contact form.

Every journey streams to a dedicated sales channel in real time — five event types stitched by a session tag, so a rep can replay a single lead’s entire path: the full reveal snapshot (goals, horizon, EMR, size, today’s costs with sources, every number shown), each tier decision, the final configured quote, and reference requests flagged as hot leads with email attached. Aggregate analytics run alongside so the funnel is readable without touching any single lead’s data. No alert can ever block or slow the prospect’s page.

Used in 120+ opportunities every month, with detailed structured intelligence flowing back to AEs and BDRs on every one. Because cost is a major predictor of wins in this market, putting a credible personalized comparison at the top of the cycle is expected to lift win rate by at least 5%. Beyond the headline: every unlisted-EMR write-in grows the competitor pricing data bank, every journey produces an intent score, and every reference request arrives pre-qualified carrying its own configured quote.

A single dark, glassmorphic page — a deliberate design sibling to the CEO Keychain built earlier for the chief executive, inheriting its exact design tokens (deep warm ground, glass cards, live IST/EST/CST clocks) so the two read as a family of internal products. Tools sit in two labelled groups rather than a flat list — “Get data & fixes” and “Answers & AI” — and every card is written in rep-facing voice (“Need a prospect’s email…”) instead of process jargon. Five tools on the keychain: the quote configurator, prospect and account enrichment requests routed into Salesforce, the Salesforce admin ticketing form, one-click install of the team’s Salesforce skill for Claude, and the searchable FAQ synthesised from hundreds of real sales calls.

The interesting engineering was in what the platform refused to allow. The artifact version runs under a strict content-security policy — no external POSTs, no iframes, no outside hosts — which killed both obvious designs: embedding the forms directly, and collecting requests in-page. The answer was a two-surface architecture. The canonical Keychain lives on Vercel, deployed from a private repo: no login needed, tray fully functional, serverless function forwarding to Slack. The artifact mirror keeps the identical design for the Claude-native audience, and its CTA deep-links to the Vercel page with a URL fragment that auto-opens the request tray on arrival.

Live in production, verified end-to-end — including confirming the Slack webhook with a live test POST before launch — and rolled out with a launch email built around the declutter thesis: one bookmark, everything GTM Engineering builds for you. Drove a 40%+ increase in GTM resource utilisation — the same tools, materially more used, purely by making them findable and giving reps a voice in the roadmap. The request tray turned a scattered wishlist into a standing intake channel where the best asks get built and returned to the same page.

Built the classification model by gathering input directly from academicians, clinicians, sales directors, and CXOs to land on a structure that was actually optimal, rather than designing it in isolation. Extended the same logic across inbound and outbound, and into closed-won and closed-lost history. Built an AI agent to gather account-level intelligence, then pushed that intelligence into Salesforce and allocated accounts to AEs and BDRs for effective, ownership-backed outreach.

The classification pass surfaced that roughly half the database had never been properly allocated or owned. After the rebuild and the AI-driven intelligence pass, the database went from ~50% unallocated to near 100% distributable — every account now has a clear owner and routing logic behind it.

Audited the full open pipeline and found 64 unassigned opportunities across a departing AE’s book worth $381K in ARR, then rebalanced ownership across four AEs at a precise 30/30/20/20 ARR split using a weighted allocation algorithm — writing back 64 opportunities, 64 accounts, and 92 contacts to Salesforce with a live audit report. In the same pass it diagnosed the implementation backlog: 27 accounts stuck for three months or more, over $1M in contracted ARR at churn risk, the worst of them stalled past 300 days. That is two to three days of analyst work — pulling data, rebalancing sheets, updating records, drafting follow-up plans — compressed into a single conversation.

Detecting a stall is the easy half. The agent reaches past the CRM into Jira tickets and email threads to reconstruct what actually happened to an implementation — where it stopped, which blocker it hit, who went quiet — and runs a root-cause analysis on that evidence rather than on stage age alone. It then drafts and sends an email to the account manager with internal managers copied, framing the specific revival path for that account. The escalation reaches a human with the diagnosis already attached, instead of a dashboard row waiting for someone to notice it.

$3M in stalled and orphaned revenue uncovered to date. Orphaned pipeline gets re-owned within a session instead of ageing unassigned; implementation accounts at churn risk reach CS with root cause attached; and automated follow-up sequences mean no account goes dark on memory or a forgotten manual task. Live dashboards in Salesforce and as shareable artifacts give leadership standing visibility into pipeline health, AE balance, and implementation risk with no manual export in the loop.

Takes two inputs — a hypothesis (the angle to test) and a target list. From there it generates the initial message and 4 follow-up emails, then sends directly through AWS SES on an inbuilt scheduler, capable of up to 2,000 emails a day. What used to be a multi-step campaign build became a single prompt.

Cut the time from campaign idea to live send from days to about 6 minutes, while still producing a full 5-touch sequence (initial + 4 follow-ups) per campaign, sent at up to 2,000 emails a day.

For the push into the 6–20 therapist mid-market, a champion board was the highest-leverage motion available — but the brief was explicit that this is not a referral scheme. It’s a tribe of high-signal operators whose public identity gets tied to the product: roughly 70% social proof that gives mid-market buyers permission to choose us, 30% predictable referral pipeline. Referrals are a downstream effect, not the goal. The founding class was capped at ten deliberately — scarcity is the point, which is exactly why selection couldn’t be political.

Four qualitative axes scored 1–5 each — advocacy signal (referrals made, NPS 9+, unsolicited testimonials, review-site activity), network reach (following, CEU instructor status, podcast presence, association involvement, speaker circuit), ICP fit (mid-market, ideally multi-location), and usage depth (breadth across modules, engagement, case-study-worthy outcomes) — layered on hard quantitative gates from NPS and CSAT. Maximum 20, realistic cutoff at 15+. Ties break in fixed order: practice-type mix, then geography, then demographics, so the class reads as a representative board rather than a cluster of similar clinics. Hard exclusions remove anyone in contract renegotiation, with active support escalations, NPS under 7, or less than six months tenure.

The algorithm only works if the inputs are real, so most of the engineering went into assembling a candidate table nobody had before. For each of the 232 candidates the pipeline joined NPS and CSAT (pulled from project exports and matched back to accounts), firmographics (therapist and clinic counts, state, disciplines, module mix, ARR, go-live date), owner contact details, and — the expensive part — researched network reach: per-candidate investigation of association membership, podcast appearances, conference speaking, published writing, and teaching roles, each captured as written evidence notes rather than a bare number.

Selection sits inside a governance process built to take politics out of the room: CS pulls every customer at NPS 9+ or with a referral, Marketing / CS / Sales score independently, scores are averaged, top 15 go to a selection meeting, leadership finalises ten. No committee debate, no horse-trading — the rubric decides. Around it, a status-first incentive stack (title and badge, public launch, roadmap access, founder office hours, ghostwritten content, annual summit) with cash deliberately last as a referral-credit floor, a monthly “menu, not mandate” activity cadence, and hard measurement bars — 10–15% of mid-market net-new pipeline sourced by month 12, 15–25% by month 18, with an explicit rule that missing the bar means redesign, not extension.

Delivered as a complete package: the scored candidate universe (232 rows with per-dimension scores), the finalist shortlist (23) with invitation-ready evidence-backed profiles, and the full program plan. All ten seats offered were accepted — the strongest available signal that the rubric was reading the right people.

Tracked how many days each deal sat in each stage to flag stalling risk before it became obvious, and pulled in live signals per account — recent news, job changes, and new leadership hires — so reps had a timely, specific reason to re-engage instead of a generic check-in. Built for the sales team's own day-to-day use, not leadership reporting.

600+ open deals tracked in real time, giving reps a single place to check deal health, catch stalling risk early, and find the right moment and reason to push a deal forward.

Programmed logic to surface only the highlights, risks, and priorities at an executive level — color-coded and skimmable in seconds, with drill-down on click for anyone who needs the underlying detail. Tracks $50M+ in pipeline, 6K+ users, $3M+ in ad spend, 30+ team performance metrics, and 20+ event types across five funnel views: Sales, Churn, Opportunities, BDR productivity, and deal-level activity.

Now the dashboard CXOs use daily for management and board-level meetings, and the source of truth for 1:1 reviews and sales strategy — giving leadership a bird's-eye view of the whole operation instead of a siloed, tool-by-tool picture.

A 5-stage pipeline (Trigger → Enrich → Validate → Voice confirm → Gated write-back) using a no-code automation platform as the orchestration runtime rather than a custom script. Solved the race condition between the CRM's near-instant native sync and enrichment timing with a status-gate field. Pushed junk-filtering upstream to cut wasted enrichment cost, and decoupled voice confirmation from the gate so hot leads aren't held up.

Core enrichment-and-gate pipeline built and operating; an LLM-as-judge validation stage and voice confirmation are planned next phases. The first system in this list that's preventative rather than retroactive cleanup.

Triangulated sales call transcripts (stated need) against engineering tickets (shipped fix) — only entries appearing in both sources counted as confirmed signal, the highest-confidence demand data available. Clustered all 85 entries into 10 themes, then operationalized it: built a tagging system so pain-point discovery runs continuously instead of as a one-time report.

Surfaced a hidden pattern — several sub-specialties kept recurring across unrelated categories, evidence of real product depth that had never been positioned or rolled into any deck, feeding directly into a later specialty-taxonomy gap analysis.

Joined 7+ knowledge-base tables per account (discipline pain points, state regulatory requirements, competitor weaknesses, feature catalog, same-state customer proof) with a problem-to-feature crosswalk so only relevant features reached each brief. Layered persona-level data (priorities, objections, psychology) on top, then generated the actual BDR-ready assets from it — cold emails, cold call scripts, and LinkedIn messages, personalized per contact.

482 account briefs and roughly 964 persona briefs generated, feeding directly into BDR cold outreach — lifting email open rates by 6 points and click-through rates by 50%.

Broke down hundreds of transcripts by year, tracking ICP description, competitor mentions, feature demand, objection types, and closing language across each year, then synthesized the findings into a competitive battle card for reps.

Quantified 47× growth in call volume over the period, pinpointed the single product launch that reshaped every pitch afterward, and captured the exact ICP language and proof point reps were using most recently. That battle card became part of how new reps ramp — a few days with the high-impact material instead of weeks of onboarding and hours of video training, and more confident, consistent answers to objections from day one.

Searches call transcripts and support messages, filters out internal speakers by name-list and email-domain checks, then verifies every candidate quote against a real non-employee attendee before it's usable. The same pipeline that surfaces one campaign's worth of evidence can be re-run against any theme — burnout, pricing objections, feature requests — whenever a new campaign angle is needed.

The “Leave On Time” burnout-relief campaign is one output of it: 240K+ utterances searched, 5 fully-attributed hero quotes and 32 supporting verbatim quotes delivered, plus a documented language gap that directly informed which customers to interview next. The generator itself is reusable for the next campaign angle, not a single-use research pass.

Piped call transcripts, support conversations, CRM leads, competitor intelligence, feature catalogs, and regulatory data into structured warehouse tables — read-only from every source system, with strict write rules and an approval gate on anything customer-facing.

A single, governed source of truth that every downstream system in this portfolio — dashboards, personalization pipelines, campaign tools — depends on.

Built specialty-based routing rules, with exclusions for active relationships, for opportunity reassignment. Ran a full reconciliation of churned accounts between two systems of record, with independent QC sampling and fuzzy re-matching on unresolved records.

129 opportunities re-assigned so no historical pipeline went unworked; 159 churned accounts reconciled, surfacing records present in one system but missing from the other — including priority gaps flagged for immediate follow-up.

Extracted every customer question asked across hundreds of calls via two-pass detection (AI-tagged questions plus missed sentences ending in “?”), then built a searchable app instead of a static document. Trained the AI to reference only authentic sources, and to weight answers from pre-vetted people — the product head, managers, CXOs, and CTOs — over unverified ones, so accuracy held up as the FAQ kept growing.

3,379 Q&A pairs extracted from 501 calls across 9 categories, live as a searchable app used by 500+ customers and 30+ internal AEs and BDRs — one accurate answer source instead of everyone re-asking the same questions on calls or in Slack.

A local execution engine plus gateway running as a background service, with a file-based persistent memory system and a library of packaged, invokable skills — one per capability: outreach drafting, call prep, competitive intelligence, forecasting, and more.

What started as a single assistant became the operating layer for the entire GTM stack — every dashboard, pipeline, and campaign tool in this portfolio was either built with it or runs on it.

Built a 3-tier risk model — contacts an individual directly or writes irreversibly to the system of record is Tier 1, influences CRM routing/prioritization is Tier 2, internal-only output is Tier 3 — and an AI System Register mapping every GTM AI system to an owner, a human checkpoint, and a primary risk. Designed the consent architecture behind automated outbound voice: a unified, append-only contactability object joining consent type, DNC status, and call history across the CRM, marketing, and sales-engagement platforms on a single phone-number key, so no dial happens without a verifiable positive-consent record. Aligned the whole framework to NIST AI RMF and ISO/IEC 42001 as a risk methodology and structural reference — explicitly without claiming certification or legal immunity.

Nine GTM AI systems inventoried and tiered, with the highest-risk one — automated outbound voice — gated behind legal and security sign-off rather than shipped by default. Turned a stack of well-intentioned automations into something a security review or an enterprise buyer's procurement team can actually audit.