Think about what's actually inside your shop software. Your customer list. Your prices. Every text and call with a client. Your quotes, your invoices, the contracts people signed. It's the one part of the business you'd never hand to the shop down the road — and you've handed it to a vendor.
Ask that vendor how they protect it and you'll hear two words: bank-level. It sounds like a standard. It isn't one. No auditor checks it, no document defines it, and nobody has ever failed it. It's a phrase, and a phrase can't be measured.
Standards can. Security engineers don't argue about adjectives; they grade software against published lists of what goes wrong and what stops it. So that's what this post does. We took the standards the industry actually uses and put how Wrapstart is built next to each one. Where we meet the bar, we say so. Where we go past it, we say so. Where we haven't finished, we say that too.
First, what actually goes wrong
The picture of a breach most people carry around — a hooded figure, green text, a vault door blown open — is almost never how it happens. The real failures are boring, and the industry has counted them.
Read those together and the story is clear. Software leaks when a page forgets to check which shop is asking, when a stolen login gets used, and when a vendor you trusted didn't deserve it. Those are the three things worth benchmarking a shop app against — and they're exactly where the standards below point.
The benchmark
Ten lines. Each one names the standard, what it asks for, and what Wrapstart actually does — taken straight from our security page, which goes deeper on every point. No line is marked "meets" because of a policy document; each one is marked because of something the system enforces.
Deny by default. Check on the server, for every single record, that the person asking is allowed to see it — never only in the screen.
PostgreSQL row-level security is on for every table, scoped to your shop and your role. Our server functions re-check that the records and files a request names belong to your shop before touching them. The database refuses a save from the app that would tie your work to another shop's customer, job, quote, invoice or other core record.
Two independent enforcement points: the application checks, then the database checks again on its own.
Strong cryptography whenever data crosses a public network. In practice TLS 1.2 is the floor; SSL and early TLS don't count as strong.
Every connection to Wrapstart's services is HTTPS over TLS 1.2 or 1.3. Older, weaker protocol versions are refused. At rest, your data is encrypted with AES-256, with daily encrypted backups.
Protect stored account data. The council's own advice: the surest way to protect card numbers is not to have them.
Card numbers never touch Wrapstart. Your customer types their card into the processor's own secure fields — Stripe, Square, Clover or SumUp, each certified PCI DSS Level 1 — and the number never reaches our servers. Payment notices from Stripe are signature-checked before anything acts on them.
That also keeps your shop in the lightest PCI category for online payments.
Offer multi-factor sign-in. Never let one shared password be the whole door. Treat text-message codes as a restricted authenticator and offer something stronger alongside.
Passwords are handled by a dedicated authentication service, never by our own code. Two-step sign-in is available on the web, with an authenticator app or a text code. Every team member signs in with their own account; invitations expire after 7 days unused.
NIST's note applies to us too: the authenticator app is the stronger choice. Use it.
Verify that data and messages really came from who they claim to, before acting on them.
Stripe payment notices are checked against Stripe's signature. Incoming texts and calls are checked against our carrier's signature before they reach your inbox. E-signed documents carry a SHA-256 fingerprint of both the original and the signed file, so any later change can be detected.
People get the access their job needs and no more. Privileged access is granted deliberately, limited in time, and logged.
Roles are set area by area — customers, jobs, messages, the AI assistant — separately for viewing, creating and editing, and enforced by the database or our servers, not just the menu. Plan changes are owner-only and checked on every request. Our own support team gets in only when you grant a session, for 24 hours to 7 days, with each sitting capped at 12 hours. You can revoke it any time, every session is logged, and the app tells you when one starts.
On shared infrastructure, one tenant must never read, write or link to another tenant's data — by accident or by crafted request.
One shop, one vault. Row-level security, server-side shop checks, refusal of cross-shop links to core records, and contracts, documents, recordings and receipts kept in private per-shop folders that are checked before every read, copy or share, with path-climbing tricks like ../ refused in any encoding. A person who works for two shops sees only the shop they've switched into.
Changes are reviewed, tested and reversible. Security controls are designed in, then proven, not bolted on.
More than 12,000 automated tests, including tests that deliberately reach for another shop's data and confirm the door stays shut. New shop-boundary protections ran in watch mode first, logging what they would have blocked, before we switched them on. Putting the file and link checks back into watch mode is a single switch. Database changes are written, reviewed and applied by a person.
Honour STOP immediately. Record consent. Don't record or auto-dial without the right permission.
STOP, UNSUBSCRIBE, CANCEL and the usual words opt a customer out until they text START; every change is recorded. A texted STOP also blocks AI calls and marketing email to that person, and email unsubscribes take effect at once. On Wrapstart phone lines, a "do not record" contact is never recorded or transcribed in either direction, and AI-voice calls follow the consent level you've captured for each contact.
An outside auditor confirms, over months, that the controls above exist and operate as described.
Today we audit ourselves — a function-by-function review of how our servers decide which shop a request belongs to — and we publish how. We have not yet completed a third-party attestation. When we do, it will be on this page and on the security page, with the report available on request.
Benchmarked October 17, 2026 against OWASP Top 10:2025, PCI DSS v4.0.1, NIST SP 800-63B-4 and the SOC 2 Trust Services Criteria.
Nine lines met, one amber. The amber one is the line most vendors quietly leave off the page, which is exactly why it's on ours.
The line that matters most, in shop terms
If you only read one row, read the first. Broken access control has been the number one web application risk for years, and it's the one that would actually hurt you: not a stranger guessing a password, but another shop seeing your customer list because a page forgot to check who was asking.
Most software guards against this in one place — the application code. A developer writes "only show this shop's records" on every screen, and hopes nobody misses one. Wrapstart writes that rule in two places. The application checks. Then the database checks again, on its own, because the rule lives inside PostgreSQL itself. A screen with a mistake in it still can't show you a row that isn't yours, because the database never hands it over.
If one layer ever got something wrong, the next would still stop it.
That's the whole philosophy, and it shows up everywhere on the list: payments where card numbers never arrive, links that open a page instead of a file store, support access you grant rather than we take. Not one lock. Several, each held by a different part of the system.
Payments: the best way to protect a card number is to never have it
Here's a useful test for any shop software: ask where a customer's card number goes when they pay a deposit. If the honest answer involves the vendor's own servers, their PCI obligations just became your problem.
In Wrapstart, the card number never arrives. Your customer types it into Stripe's, Square's, Clover's or SumUp's own secure fields, and the number never reaches our servers. All four processors are certified to PCI DSS Level 1, the highest tier of the card industry's standard. We couldn't leak a card number if we tried, because we never had one.
Sign-in: the front door
Eighty-eight percent of breaches from basic web application attacks run on stolen credentials — a password stolen from somewhere else, scooped up by malware, tried against a hundred sites until one opens. That isn't a vendor's weakness to fix; it's the shared front door between you and us.
Our half: every team member gets their own login, passwords are handled by a dedicated authentication service rather than our code, and two-step sign-in is there to switch on, on the web. Your half: switch it on, and pick the authenticator app over the text code. NIST calls text-message codes a restricted authenticator for good reason — SIM swaps and number porting are real — and the app takes thirty seconds to set up. Ask your front desk to do it this week.
Who else touches your data
Thirty percent of breaches now involve a third party, so it's fair to ask who ours are. These are the companies that handle your data on our behalf, and what each one does:
- Supabase runs the database and sign-in, and Vercel hosts the app. Both hold SOC 2 Type 2 reports.
- Twilio carries your texts and calls. Resend sends your email.
- Stripe, Square, Clover and SumUp take payments, whichever one your shop uses.
- OpenAI, Anthropic and Bland AI power the AI: drafting replies, transcribing calls and answering the phone. Every AI draft is reviewed by your team before it goes out.
If a breach ever affected your data, you'd hear from us within 72 hours where the law requires it. And when you close your account, your data is deleted within 30 days.
Every claim above, one layer deeper
How the vault works, how links expire in ten minutes while messages stay useful for weeks, how support sessions are granted, and how we ship security changes in watch mode before they block anything.
What we haven't done
We'd rather you hear this from us. Wrapstart has not yet been through a SOC 2 or ISO 27001 audit. Everything on the list above is enforced by the system and tested by us — twelve thousand tests, a function-by-function review of our shop-boundary checks, changes rolled out in watch mode before they bite. But "tested by us" and "confirmed by an auditor who doesn't work for us" are different sentences, and only one of them earns a stamp.
We'll get the stamp. Until then, you get the thing most vendors give you instead of one: a page that tells you exactly what's behind the door, which you can hand to anyone you trust to check it.
Why this is on a shop software blog
Because the data we're asking you to trust us with is the data this whole product exists to work: the leads you already paid for. Wrapstart is built on the idea that a lead, a quote, a job, a message and an invoice should all know about each other — that's what makes closing, not getting, the thing software can actually fix. Connected data is the product. Connected data is also precisely what has to be walled off, shop by shop, without exception.
So the two aren't separate conversations. The architecture that lets your follow-up fire on time is the same architecture that makes sure the shop down the road never sees who you were following up with. You can't have the first without getting the second right, and we built it in that order.
Found something that doesn't look right? security@wrapstart.com. We read every report, and we'd much rather hear from you than not.
Hamza Ishaq · Founder
