How a US-Based VA Marketplace Launched After a Year Stuck at 80% — With Airtable, Zapier, and a Scoring Algorithm That Actually Works
A bootstrapped VA matching platform stuck at 80% for over a year launched in 11 weeks — with Airtable, Zapier, and a scoring algorithm built by Connex.
TL;DR: A bootstrapped VA-entrepreneur matching platform that had been 80% complete for over a year finally launched publicly in 11 weeks — after Connex rebuilt the matching algorithm, added AI-powered resume screening, and wired the full subscription and payment stack.
Photo by krakenimages on Unsplash
The Situation
A small business owner with a background in executive assistance and project management spent more than a year building a marketplace to connect US-based virtual assistants with entrepreneurs. The platform — built on Webflow, Airtable, Zapier, MemberStack, and Stripe — had a clear vision: entrepreneurs fill out a match form describing what they need; vetted VAs fill out a profile form; the system automatically scores and surfaces the best matches by subscription tier.
By early 2026, the platform was 80% there. The other 20% was the hard part.
The Problem
"The matching logic was stuck," as the founder described it on the initial call. The platform collected data fine — Typeform questionnaires fed Airtable, Stripe handled subscriptions, Webflow showed the front end. But the algorithm that took an entrepreneur's requirements and ranked VAs against them — considering must-have criteria, nice-to-haves, availability, experience tier, time zone, and subscription level — wasn't working reliably enough to go live.
The contractor who'd built the foundation was a capable Zapier expert. But the scoring and ranking logic required a different level of systems design: multi-step conditional logic across a large pool of VA profiles, scoring by weighted criteria, surfacing matches by tier without crossing subscription boundaries, and handling edge cases like mismatched time zones or mid-session plan changes.
The platform couldn't launch without it. And it had been stuck for months.
What They'd Already Tried
The founder and her contractor had spent roughly a year building the platform from scratch — which got them 80% there. The pieces were real: Airtable as the VA profile database, Typeform for both the VA application and entrepreneur match request, Webflow for the front end, MemberStack for member authentication and tiered access, Stripe for subscriptions. The automation scaffolding existed.
What they hadn't been able to complete: the multi-criteria scoring algorithm in Zapier that combined must-have hard filters with weighted nice-to-have scoring, applied it across potentially hundreds of VA profiles, returned results sorted by match percentage, and respected subscription tier boundaries — all within Zapier's 30-second timeout constraint.
What Was Built
A Connex consultant took over the project in early February 2026. Working in closely-spaced sessions over 11 weeks, the team built and shipped:
The matching algorithm. A multi-step Zapier workflow that filters VAs on hard must-haves (availability, experience level, subscription status, US-based location), then scores them on weighted nice-to-have criteria, sorts by match percentage, and returns results separated by tier (Growth, Executive, White Glove). Edge cases handled: overlapping answers, no-preference time zone selection, entrepreneurs whose tier constrains which VA profiles unlock.
AI-powered resume screening. VA applicants upload a PDF resume. A Zapier step sends it to OpenAI for analysis — checking job titles, experience levels, and US-location indicators before the resume ever reaches Airtable. Non-PDF uploads trigger a re-upload email automatically. This gate replaced a manual review step the founder had been handling herself.
Full email automation. VA onboarding sequences (welcome, pre-screen pass/fail, assessment, NDA), entrepreneur match notification emails, and payment failure recovery flows — wired through ConvertKit and triggered by Zapier.
Subscription and payment stack. MemberStack plan tiers synced to Stripe subscriptions and Airtable membership records. A sync bug (51 members showing in Stripe but not reflecting in Airtable) was found and fixed before launch. Coupon and promotional-code flows built and tested.
Entrepreneur match dashboard. Webflow front end updated to show match percentages, hide unmatched tiers, display VA previews (first initial only, for privacy), and trigger the correct unlock flow on subscription purchase. A threshold was set: if no matches clear 70%, the entrepreneur sees explanatory copy rather than a blank result.
The Results
The platform launched publicly in the third week of April 2026 — approximately 11 weeks after Connex took on the project.
At the point of launch, the system had 51 active subscribers in Stripe, a full three-tier VA roster (Growth, Executive, White Glove), and automated flows running end-to-end for both VA onboarding and entrepreneur matching. The founder went from manually reviewing VA applications and trying to coordinate matches by hand to a system where new VA sign-ups are screened by AI, scored, slotted into the right tier, and surfaced to matching entrepreneurs automatically.
The key milestone isn't a time-saved number — it's that the platform launched at all. A product that had been 80% complete for over a year, unable to go live, crossed the finish line in under three months.
Post-launch, the Connex team continued in an operational support role — fixing edge cases, adding features, and handling the kind of ongoing automation issues every live platform surfaces (a coupon expiry, a webhook delivery bug, a Zapier connection drop after an API key rotation). As of July 2026, the platform remains live and actively matching VAs with entrepreneurs.
Client Voice
We have the pieces — Webflow, Airtable, Zapier, MemberStack, Stripe — but the matching algorithm isn't working well enough to go live. We've been stuck at 80% for a long time.
— The platform's founder, at the initial consultation. (Paraphrased from the sales call; direct quote withheld per anonymization policy.)
Frequently Asked Questions
How do you handle matching logic that has to run across a large pool of records within Zapier's 30-second timeout?
The answer is in how you structure the filter and score steps. Hard must-haves (availability, subscription tier, US location) run first and eliminate the bulk of non-matches before any scoring math runs. The scoring step only processes VAs that cleared the hard filters — dramatically reducing the record set. Edge cases like overlapping preference fields are handled with conditional branches rather than trying to score them the same way as standard fields.
Did the AI resume screening step add significant cost?
It was designed to minimize OpenAI calls. The Zapier workflow checks whether a file is a PDF and whether basic location indicators are present before ever sending to OpenAI. Non-PDFs are bounced back to the applicant immediately. This means the AI step only fires on screened-in applicants — keeping token costs proportional to real application volume rather than running on every submission.
How long does it take to go from 80% done to launched?
In this case, 11 weeks — but that number is downstream of how the project was structured. Work happened in closely-spaced sessions (sometimes daily during the final pre-launch crunch), with the consultant owning execution and the client providing context and sign-off. The fast pace was possible because the foundation (Airtable schema, Typeform, Webflow front end, Stripe account) was already in place and documented.
What keeps this kind of platform running reliably after launch?
Subscription platforms break in predictable ways: payment failures, API key expirations, coupon logic edge cases, member sync drift between Stripe and your database. Building a clear path for each of those — payment recovery flows, sync reconciliation views in Airtable, automated alerts — is part of what makes the difference between a launch that holds and one that needs constant intervention.
Ready to Finish What You Started?
Ready to unstick a project that's been 80% done too long? Book a consultation →
How a regional mortgage lender automated offline conversion tracking across Google, Bing, and Meta using Zapier and the Velocify API — no weekly manual exports.
How a mid-sized countertop manufacturer replaced their inbound status-inquiry phone queue with an AI voice agent connected to Moraware. Read the case study.
