Sandbox VR Alpharetta sells corporate events, school trips and group bookings, which means a local B2B motion. Rather than renting a database, we built the list: companies by real radius, each row gated on a reachable decision-maker, headcount from public federal filings, chamber directories taken whole, and every address verified before a single send. Then we built the thing that works it.
The problem
Corporate bookings, school trips and group events do not walk in. Somebody has to go and get them, and the venue was doing that from a hand-typed spreadsheet with the outcomes living in whoever made the call.
37 rows
Hand-typed, and that was the list
The corporate lane was a few dozen companies someone had thought of, with no way to know which had a decision-maker anyone could actually reach. Bought databases solve this at a subscription price that makes no sense for one venue working one radius.
In someone's head
Call outcomes went nowhere
Who was called, who never answered, who asked to be rung back next quarter: all of it depended on the person who made the call remembering to write it down, in a sheet nobody else was looking at.
One cold domain
One bad send list to burn it
Outreach ran from the store's real mailbox. An invented address bounces, and bounces on a cold domain are how a sending reputation dies, taking every future email with it. Guessing addresses was not a small risk, it was the whole channel.
The solution
Two halves. The first assembles a local prospect list from public sources and refuses to include a row nobody can act on. The second puts it in front of the person whose lane it is, dials it, and writes down what happened without anyone having to.
A list builder that assembles local companies by genuine driving distance rather than by a crude postcode filter, and refuses any row without a reachable human in a decision-making seat. It replaced a hand-typed few dozen with hundreds of usable corporate records, and it can be re-run rather than re-typed.
Stack: Radius-banded company assembly with geocoding, decision-maker gate, generated data file read by the outreach board.
Company headcount is the field that decides whether a venue can host a team, and it is the field data vendors charge most for. It is pulled here from the Small Business Administration's public PPP release instead, free, for local companies. Chamber of commerce directories are taken whole rather than a page at a time, and nearby restaurants and venues are banded by real distance for partnership outreach.
Stack: SBA PPP FOIA dataset for headcount, whole-directory chamber ingest, venue and partner list generation.
Every address is checked for whether it can actually receive mail before it enters a send. This exists because outreach runs from the venue's real mailbox, and a run of bounces on a cold domain does not cost you one campaign, it costs you the channel. Verification is the cheapest insurance in the system.
Stack: Address verification pass ahead of every send, run against generated and hand-curated lanes alike.
Corporate, schools, hotels, local business, residential, social and personal contacts are separate lanes, each assigned to a specific person, and the board serves only the lanes that person may see rather than hiding the rest in the browser. Everyone opens the same board and sees their own beat.
Stack: Role-gated API responses, per-user lane assignment, hand-curated and generated lanes side by side.
Continuous dialing straight from the board with a safety buffer between calls, controls to dial, pause or stop, a single button to end a voicemail and move on, and inbound call registration so a returned call lands against the right record. If the caller hangs up their own phone the session terminates immediately, which is what stopped phantom dials.
Stack: Twilio Serverless dialing with conference bridging, live status polling on the caller's own line, auto-advance with an interruptible buffer.
When a dialed lead goes to voicemail they get a text, once, without anyone deciding to send it. It takes two independent facts: answering-machine detection must have returned a machine verdict, and the call must have run under a minute. The duration is a veto and never a reason to send, because a screened handset puts a real machine on the line before a human takes the call, which produced a machine verdict on a call the owner answered in person. One automatic text per lead, ever, event leads only, with a kill switch.
Stack: Twilio AMD plus completion callback, verdict held against the call, duration veto, permanent per-lead send guard.
Every outcome writes itself to the contact record from the phone system rather than from someone's memory: reached, no answer, voicemail, called back, owed a follow-up. Voicemails arrive transcribed against the contact. The board answers who has been contacted, by whom, when, and what happened, and it answers it the same way for everyone looking at it.
Stack: Ledger built from call and voicemail records, shared team-editable state, follow-up and win-back status per contact.
Paid acquisition gets the same treatment: the Google Ads account graded and ranked by dollars, the Meta account audited read-only with a configuration fingerprint and a change log so drift is visible, daily spend checked directly because it is the only honest confirmation that a pause actually took, campaigns staged paused rather than launched blind, and account changes verified afterwards to confirm they registered. Customer Match audiences are built from the venue's own booking and payment records.
Stack: Read-only account audits with change logs, day-by-day spend verification, paused campaign staging, first-party audience export.
Part of the same build: the operations platform and the AI booking agent.
The results
Figures describing displaced manual work are the operator's estimate rather than instrumented measurement, itemized so they can be argued with, and counted only once across the three case studies in this build.
Built, not bought
Their own prospect database
Hundreds of local corporate records assembled by real driving distance, each gated on a reachable human in a decision-making seat, with headcount taken free from public federal filings rather than rented from a data vendor. It replaced a hand-typed few dozen, and it can be re-run rather than re-typed.
~7.5 hrs
Recovered every week
Roughly 4 hours on dialing by hand, logging what each call did and working out who never answered, 2 on researching who to approach in the first place, and 1.5 on checking what the ad accounts actually did rather than what they reported. Nothing enters a send without a deliverability check first, because the sending domain is a single shared asset one bad batch could burn permanently.
1 per lead
Automatic texts, ever
The voicemail follow-up fires once per lead for the lifetime of the record, on two independent signals, only to someone who asked to be contacted and only after a human dialed them. A second automated “we just tried you” adds nothing and costs trust.
Built for one venue. Replicable for yours.
Most local businesses do not need a national prospect database. They need every company worth calling inside a radius, with a real person to ask for, and a system that remembers what happened. We scope, build and hand off custom software for operators. Book a free 30-minute audit call.
Book a Free Audit Call