A prototype for Snaplio, built on 30 real Miami businesses
Who should you follow up with today?
Thirty captures from trucks, signs and voice notes, each checked against a public page (its own site, or a directory listing) and sorted into a short list. What you do today changes tomorrow.
Today
Oldest captures first, so nothing goes cold.
- 1Likely fitCaptured 5 days agoCENTRAL C1VILCONSTRUCTI0NMEDLEY FL305-888-33 44DOT 1 . . .
Truck door, Medley. Colors illustrative. Pink: what the OCR misread. Central Civil Construction
The phone on the capture is on centralcivil.com, the business's own site: identity checked.
Still unknown, ask themDo they own or rent their excavation and paving equipment?Large civil contractor doing excavation, grading, utilities and paving needing heavy equipment fleet
Their site: “structural excavation and backfilling for pile cap construction, underground utilities”
16 more waiting behind these. 9 captures kept off the list: outside this rep's profile.
Rules on the lettering, or the proposed addition?
Same 30 captures, two ways, for the equipment rep. The simple baseline reads only the capture: trade words on the truck, and the phone if it is complete. It costs nothing. The addition finds the business on the web, checks the phone, reads its site and writes the question to ask.
17 of 30 come out the same. Of the 13 that differ, 4 change what the rep actually does. All 4 are captures where the lettering said nothing about the trade or gave no phone. Two voice notes said the number out loud; rules that read the transcript get it too, so those count as the same.
So the addition only needs to run on those: 7 of the 30 here, about 0.32¢ per capture on average instead of 1.36¢, and every change that mattered is kept.
“Engineering Contractors” gives rules nothing to go on. uagconstruction.com says heavy civil, earthwork, paving: likely fit. The other case where it earns its cost is a voice note with no number: Dade Wrecking, found with (786) 355-8065 on dadewreckingcontractors.com.
Found the number the capture did not give
1A voice note with a name and no number. The addition found the business and the number on its own site.
- Dade Wreckingrules Likely fitaddition Likely fit
Found a fit the lettering hid
1The truck said Engineering Contractors. The site says earthwork, drainage, paving.
- UAG Constructionrules Needs qualificationaddition Likely fit
Took two plumbers off the list
2The lettering said nothing an equipment rule could use. The sites show plumbing service.
- Seaway Plumbingrules Needs qualificationaddition Outside profile
- Bradley Plumbing & Gasrules Needs qualificationaddition Outside profile
Changed the label, not the call
8Rules said likely fit, the addition said needs qualification. Either way the rep calls and asks whether they own or rent.
- PAL Demolitionrules Likely fitaddition Needs qualification
- Swimming Pools of Floridarules Likely fitaddition Needs qualification
- USA Plumbing & Septic Incrules Likely fitaddition Needs qualification
- A. Palmer Roofing Incrules Likely fitaddition Needs qualification
- A1 Duran Roofingrules Likely fitaddition Needs qualification
- Extreme Roofing Incrules Likely fitaddition Needs qualification
- Henry's Fence Incrules Likely fitaddition Needs qualification
- P.H.B. Landscaping & Nurseryrules Likely fitaddition Needs qualification
Refused to guess
1Half a name and half a number. Rules read AIR COND and set it aside; the addition found three possible businesses and asks the rep.
- Oscar & Srules Outside profileaddition Identity unresolved
The grouping is our reading of each difference. No salesperson has judged these calls, so this shows what changes, not that the changes sell more. A second run of the addition on the same evidence gave the same label 84 times out of 87.
What each capture costs, against a $149 plan
Incremental matching and recommendation cost per capture: what the addition spends on top of what Snaplio already does. Reading the photo or the voice note is not in it.
Measured
per attempted capture, the refused one included (30 captures). Per usable result: 1.41¢ (29).
- Web search, 5 results with page text
- 0.70¢
- Recommendation, Claude Sonnet 5
- 0.57¢
- Reading the capture text, Claude Haiku 4.5
- 0.09¢
- Each rep who sells two more lines
- +1.10¢
Search: the price Exa returned on each request. Models: tokens logged on every call (about 2300 in, 140 out for a recommendation), priced at Anthropic's public rates. That part is a calculation, not an invoice: two real API calls replaying one capture came in 16% under it, so the figure leans high.
Projected, per rep per month
| Captures a day | Addition on every capture | Only where the lettering is silent | Left of $12.42, every capture or only silent ones | Glide updates past the allowance, every capture or only silent ones |
|---|---|---|---|---|
| 5 (110/mo) | $1.50 | $0.35 | $10.92 or $12.07 | $2.20 or $0.51 |
| 20 (440/mo) | $6.00 | $1.40 | $6.42 or $11.02 | $8.80 or $2.05 |
| 50 (1100/mo) | $15.00 | $3.50 | -$2.58 or $8.92 | $22.00 or $5.13 |
22 working days a month. “Left” is the $149 a year less the addition only, before anything else, so it is not a margin. On Glide Classic, a Call API run uses one update; the Business plan ($199 a month, billed yearly) includes 5,000 and charges $0.02 each beyond. Past the allowance, Glide can cost more than the AI. Whether you pass it depends on your plan and the rest of the app's usage.
Not included
- Reading the photo or the voice note (Snaplio already does it; our inputs were typed text)
- Glide plan, seats and included updates (depends on his current plan)
- Email or in-app delivery of the Today list
- Development, support, payment fees
Whole experiment, search spend: $0.40, finding the 30 businesses included. One model reply in an earlier comparison run (a cheaper model, not the measured path) came back as broken JSON and was retried.
The assessment you asked for
Snaplio already keeps Interested and Not Interested, a pipeline, call, text and email actions, and voice to SMS. The addition should build on those, not rebuild them. What follows is from the public app and this prototype; I have not seen your tables or workflows yet.
| Proposed work | Complexity | Planning estimate | Recommendation |
|---|---|---|---|
| Reuse lead status and the call, text and email actions | Low | 2 to 4 days | First. Depends on the current tables and workflows. |
| One seller profile: identity check, evidence, next question | Medium | 1 to 2 weeks | Build. The hard part is ambiguous matches and refusing to guess. |
| Today list in the app and one tested reminder channel | Medium | About 1 week | Build, to test follow-through. It does not guarantee a habit. |
| Several profiles and scoring settings | Medium | Later | Defer until one profile produces useful decisions. |
| Push notifications and lock-screen actions | High | No promise | Separate feasibility test first. |
| CRM integrations | Medium to high | Later | Defer. Keep stable ids and a versioned interface. |
Planning estimates before seeing the app: they assume the current tables hold the capture, one seller profile, English and Spanish text.
The smallest test worth running
- One seller profile, written with offer, territory, customer types and exclusions.
- The identity check, the evidence and the question, only on captures where the lettering is silent or the phone is missing.
- A Today list inside the app, plus an opt-in email each morning that opens it.
- Measure time from capture to first action, share of captures followed up, wrong matches, and use in the second week.
Not yet
- Several profiles per rep and scoring settings.
- A 0 to 10 score shown to reps: it reads like a buying probability it is not.
- Estimates of company size or fleet that no source states.
- Push notifications with actions on the lock screen: Glide deprecated its push notifications in December 2024.
- CRM integrations. Keep stable ids and a versioned contract so they can come later.
How Glide would call it
An additive integration appears feasible; the route depends on your plan, tables and capture workflow. The direct one is the Call API action (Business and Enterprise), which sends the capture and saves the answer, one update per run. The other sends the capture out with Trigger Webhook and gets the answer back through a webhook-triggered workflow. Either way the service returns the same small record: capture and owner ids, profile version, identity status, evidence, recommendation, cost. The service lives outside Glide, so a CRM can use it later.
Next step: look at the real app and plan, then a pilot with 5 to 10 reps for about ten working days. The Free plan's 3 leads a month is too few for that; the pilot needs a test allowance.
How this was built
The businesses are real, found on the public web in Miami-Dade and nearby. The captures are constructed: their public name and phone written as a truck, sign or card would show them, with OCR slips added. Voice notes are typed, so they test Spanish and English text, not speech recognition. The engine never saw the answer key; failures and the refused capture stay in every count.
Pipeline per capture: a small model reads the text, one web search, a plain-code check that the phone on the capture is on the page found, then one recommendation per profile from that page only. Results are frozen; nothing on this page calls a model.
All 30 captures and what came back
| Capture | Identity | Equipment rep | Source |
|---|---|---|---|
| CENTRAL C1VIL | Phone matched | Likely fit | centralcivil.com |
| AM UTILITY | Phone matched | Likely fit | amutilitycontractors.com |
| DEMOLITION BY | Phone matched | Likely fit | absodemo.com |
| THUNDER | Phone matched | Likely fit | thunderdemolition.com |
| “ok note to self, Dade Wrecking, they had a big excavator and a mulcher on a lot on 137th avenue, clearing it, didn't get the number, find them” | Name matched | Likely fit | dadewreckingcontractors.com |
| PAL DEMOLITION | Phone matched | Needs qualification | paldemolition.com |
| UAG | Phone matched | Likely fit | uagconstruction.com |
| SOUTHERN ASPHALT | Phone matched | Likely fit | southernasphaltengineering.com |
| JD'S ASPHALT | Phone matched | Likely fit | jdsasphalt.com |
| DADE ASPHALT INC | Phone matched | Likely fit | findglocal.com (directory) |
| ANOTHER POOL BY | Phone matched | Needs qualification | swimmingpoolsofflorida.com |
| “apunta esto, una obra de piscina de AGLO Pools, AGLO con g, el numero es tres cero cinco, cuatro cinco ocho, uno ocho uno cuatro, tenian una mini excavadora alquilada parece” | Phone matched | Likely fit | aglopools.com |
| MORALES & SON INC | Phone matched | Outside profile | moralesandsonac.com |
| CARVAJAL A/C | Phone matched | Outside profile | carvajalmechanical.com |
| DADE SUPER C00L | Phone matched | Outside profile | dadesupercool.com |
| MARTIN AIR AND HEATING LLC | Phone matched | Outside profile | martinairandheating.com |
| SEAWAY | Phone matched | Outside profile | bbb.org (directory) |
| BRADLEY | Phone matched | Outside profile | bradleyplumbers.com |
| USA PLUMBING | Phone matched | Needs qualification | usaplumbingandseptic.com |
| OSCAR & S | Unresolved | Identity unresolved | none, three candidates (directory) |
| A. PALMER ROOFING INC | Phone matched | Needs qualification | apalmerroofing.com |
| A1 DURAN | Phone matched | Needs qualification | a1duranroofing.com |
| EXTREME ROOFING INC | Phone matched | Needs qualification | extremeroofingmiami.com |
| arxcis inc. | Phone matched | Needs qualification | arxcisconstruction.com |
| HENRY'S FENCE INC | Phone matched | Needs qualification | henrysfenceinc.com |
| “otro, camioneta de Allday Fence, all day todo junto, cercas y muros de concreto, tres cero cinco nueve uno cero sesenta y seis cero uno” | Phone matched | Likely fit | alldayfence.com |
| PURE | Phone matched | Outside profile | puremovers.com |
| NATIVE | Phone matched | Outside profile | nativepestmanagement.com |
| P.H.B. LANDSCAPING | Phone matched | Needs qualification | phblandscaping.com |
| MR. CRUNCH | Name matched | Outside profile | mrcrunchfood.com |