Building a growth engine from zero: How one COO boosted ARR 38% using Claude Code
Most founders treat growth as a hiring problem. Build the function first, figure out the channels second. Omar Ismail of Ascend did it in the opposite order, and built a better engine for it.
When Omar Ismail joined Ascend (formerly FlyFlat) as COO, the company had everything a startup wants early on: $20M ARR, 650+ clients including top venture firms and technology companies like Ramp, including and a 24/7 travel concierge that people genuinely loved.
What it didn't have was a growth engine. Essentially 0% of revenue came from scalable channels. No paid ads. No email outbound. No programmatic acquisition of any kind. Roughly 95% of growth came from word-of-mouth and a handful of community partnerships.
"The goal was to build a repeatable, measurable growth engine without losing the premium feel that made the product work in the first place," Omar wrote.
Six months later, January was Ascend's best month on record, reaching $27.6M ARR, which was an increase of 38%. Growth channels were returning ~5x on ad spend after just two months, projecting toward 8–10x as pipeline matures.
What makes the story worth studying isn't the result. It's the method. Omar built the entire growth engine without a formal growth team. Nearly all of it was built and operated daily through Claude Code.
Top takeaways for founders building an AI growth engine
- Your best growth insight is already in your data. Ascend had 4,582 bookings. That's more signal than most founders think to use. Analyze your existing customers before building any acquisition motion.
- Brand positioning is a growth lever, not a marketing exercise. The ICP research exposed a gap between how Ascend was positioned and who they actually served. Fixing that alignment unlocked the rest of the engine.
- Meta and LinkedIn require fundamentally different strategies. On Meta, creative does the targeting. On LinkedIn, identity does the work. Running both with the same playbook leaves money on the table.
- Attribution infrastructure is a competitive advantage. Most early-stage teams can't answer "which channel is generating revenue?" Building that answer from day one changes every future decision.
- AI can run what used to require a team. ICP research, paid campaigns, outbound sequences, CRM automation, the entire stack was built and operated through Claude Code without a dedicated growth hire.
- Growth is a sequencing problem, not a headcount problem. The three stages, ICP, brand, execution, only work in order. Teams that skip to execution without the foundation first are building on sand.
Context: incredible product, invisible growth engine
Ascend (formerly FlyFlat) is a premium 24/7 travel concierge serving executives, EAs at private equity and venture capital firms, and high-net-worth frequent travelers. When Omar Ismail joined as COO, the company had $20M ARR, 650+ clients including Google Ventures, Ramp, and Left Lane Capital, and genuine product-market fit. The concierge was beloved. The growth engine didn't exist.
Challenge: 95% word-of-mouth isn't a growth strategy
95% of revenue came from word-of-mouth and a handful of community partnerships. No paid acquisition. No email outbound. No programmatic motion of any kind. The product was premium, however the brand was positioned as a discount flight service, misaligned with the customers actually driving revenue. Without a scalable, repeatable acquisition system, growth had a ceiling. And building a growth engine "the traditional way" would require a dedicated growth team Ascend didn't have and wasn't ready to hire.
Solution and process: know your customer, fix your brand, then build the engine
Omar approached the problem in three deliberate stages, each one unlocking the next.
First, he went back to the data Ascend already had. Analyzing 4,582 bookings revealed that roughly 75% of revenue came from EAs at PE, VC, and hedge funds, with a secondary ICP of HNW executives in crypto, banking, and venture. The top 500 customers were enriched using Firecrawl, producing six targetable segments and the lookalike audiences that would power paid acquisition.
Second, the ICP data exposed a brand problem. Ascend was selling discounts to customers who cared about reliability, status, and provable ROI. Sales call transcripts, run through the Jobs to Be Done framework using Claude, surfaced three distinct personas with fundamentally different motivations. Customer language from those transcripts became the brand voice. A creative angle matrix mapped six psychological hooks across all three segments, ensuring every ad, email, and landing page spoke to the right person in the right way.
Third, with ICP and messaging locked, Omar built the full acquisition stack: paid media on Meta and LinkedIn, three outbound channels running in parallel, and a CRM rebuilt from scratch around a single question: for every paying member, which channel brought them in and what did it cost?
Execution: building the stack without a team
The entire system was built and operated through Claude Code, connected directly to the HubSpot, Meta, and LinkedIn APIs, then packaged into reusable skills that held the full operational playbook so every session picked up where the last one left off.
On paid media, Meta and LinkedIn ran completely different architectures. Meta is creative-led (broad geo-targeting with strong creative, letting the algorithm self-select the audience). LinkedIn is identity-led (precise targeting by job title, seniority, and firm type, reaching EAs at PE firms with a specificity Meta can't match).
On outbound, three channels ran simultaneously: LinkedIn sequences via HeyReach (adapted per persona), cold email via Instantly (structured as Observation → Problem → Proof → Ask), and warm introductions via Draftboard (lower volume, highest meeting-to-close rate of any channel). All three fed into the same attribution system as paid.
On CRM, standard platform integrations left too many gaps. Ascend built a 22-branch attribution rules engine on top of HubSpot that achieved near-100% contact attribution across every channel. Automations fired at every funnel stage: nurture sequences for new leads, five-minute Slack alerts for high-value sign-ups, auto-reschedule for no-shows, renewal sequences 30 days before expiry.
Recurring operations ran as Claude Code slash commands: /daily-ad-review, /weekly-growth-report, /new-campaign, /creative-batch. Growth ops became a continuous, compounding process rather than a series of one-off projects.
Results: from word-of-mouth to measurable, repeatable acquisition
Six months after starting from zero, January was Ascend's best month ever.
| Metric | Result |
| ARR | $27.6M (+38% growth) |
| Q1 ad spend | ~13K |
| Current ROAS | ~5x (projecting 8–10x as pipeline matures) |
| Cost per lead (Meta) | $42–45 |
| MQL → booked call rate | 48.7% |
| Dedicated growth hires | 0 |
The remaining challenge?
Roughly half of qualified leads drop off before booking a call. Ascend is addressing this by calling high-value leads within five minutes of sign-up and moving to a WhatsApp-native onboarding flow, in other words, meeting prospects on the channel they'll use as a member from day one.
“Now that the architecture is in place, it will compound over time,” wrote Omar.
Frequently asked questions about building a growth engine with AI
What is an AI-powered growth engine?
An AI-powered growth engine is a full acquisition system, ICP research, paid media, outbound sequences, and CRM automation, built and operated using AI tools rather than a dedicated growth team. Instead of hiring specialists for each function, a single operator uses AI to research, execute, and optimize across all channels simultaneously. Ascend's version, built through Claude Code, achieved 38% ARR growth in six months with no formal growth hire.
When should a startup start building a programmatic growth engine?
After product-market fit is confirmed, not before. The prerequisite isn't a certain ARR level, rather it's having enough customer data to identify where value actually concentrates. Ascend had 4,582 bookings to analyze. The minimum viable dataset is whatever lets you distinguish your best customers from your average ones. If you can't answer "who are our highest-value customers and what do they have in common?" you're not ready to build the engine.
How many customer segments should an early-stage startup target?
Ascend identified six distinct segments from their ICP analysis: finance executives, tech founders, crypto/Web3, luxury/media, consulting/law, and high-net worth solo operators. The right number is whatever the data produces: enough to surface meaningful differences in motivation and channel behavior, not so many that your messaging becomes unmanageable. Four to six segments is a practical range for most B2B companies at the early-growth stage.
Should startups use Meta or LinkedIn for paid acquisition?
Both, but with completely different architectures. Meta is a creative-led platform: broad audiences work better than narrow targeting, and the algorithm finds your customer if you give it strong creative. LinkedIn is an identity-led platform: precise targeting by job title, seniority, and company type is where it outperforms. Running the same campaign structure on both platforms is one of the most common and costly early-stage paid media mistakes.
How do you build a CRM attribution system that actually works?
Start with one non-negotiable question: for every paying customer, which channel brought them in and what did it cost? Build every other CRM decision around answering that question. Standard platform integrations (Meta's Conversions API, LinkedIn's insight tag) will leave gaps, especially with outbound channels. Ascend built a 22-branch rules engine on top of HubSpot's native tools to achieve near-100% attribution. The channel tag should follow the contact from first touch to paying customer, not just to lead creation.
Can AI really replace a dedicated growth team?
For the execution layer, yes, to a significant degree. Ascend ran ICP research, paid media across two platforms, three outbound channels, and full CRM automation without a single growth hire, using Claude Code to build and operate the system. What AI cannot replace is the strategic judgment that precedes execution: understanding which customer to target, what the brand should stand for, and which channels make sense for your ICP. Those decisions still require human thinking. Once they're made, many elements of the function can be operated by AI.
What is the biggest mistake founders make when building a growth engine?
Skipping the ICP stage and going straight to execution. Paid ads and outbound sequences are only as good as the customer profile they're built on. If your targeting is wrong, more spend makes the problem worse. The highest-leverage investment before launching any acquisition channel is a rigorous analysis of your existing customer data, who your best customers are, what they have in common, and why they chose you.
Omar Ismail is COO of Ascend (formerly FlyFlat). Read his full breakdown of how he built Ascend's growth engine using Claude Code at, including the complete ICP methodology, rebrand process, and paid media architecture.
If you're a founder or operator building with AI, subscribe to Building AI Differently for monthly vault drops on product, GTM, pricing, and engineering, from the teams doing it right.
