Isolation is provable
Every record is tenant-scoped and enforced by row-level security at the database layer, a control the application runtime cannot bypass. Automated tests cover the isolation boundary and run on every change.
Tidal Wave · Trust
Trust posture, in plain terms.
How Tidal Wave handles your data, what we will show you, and what we will not claim. Wave sits beside your stack, reads through native connectors, and writes back only what you can audit, roll back, and switch off.
Quick answer
Tidal Wave is a per-tenant marketing intelligence platform built on three verifiable claims: tenant isolation is enforced at the database layer and covered by automated tests, consent is enforced at the boundary and fails closed, and every CRM write is logged with its before and after values and reversible in one click. We demonstrate each claim live in security reviews rather than asking you to take it on faith.
The trust facts
- Row-level tenant isolation the runtime cannot bypass
- Consent enforcement that fails closed
- One-way suppression: opt-outs never auto-reverse
- Every write logged with before and after values
- One-click rollback and a two-layer kill switch
- One model per tenant, no shared training
- Read-only and no-API-key onboarding paths
Last updated September 2026
The three claims
Three claims to hold us to.
Security reviews run on proof, not prose. These are the claims Wave is built around, and every one of them is demonstrable in a live session.
Consent fails closed
Do-not-contact people are never written to. Do-not-predict people are never scored. Suppression is a one-way ratchet: an opt-out can suppress a record, and nothing ever auto-un-suppresses it.
Every write is audited and reversible
Each CRM write is logged with the previous value, the new value, and the model version that produced it. Rollback is a first-class operation, and a two-layer kill switch stops writeback instantly.
Verify, do not trust
What a security review gets: a working session, not a slide.
We would rather show the controls than describe them. A review session walks through four live demonstrations on a working tenant.
- 01
Watch the isolation suite run
The automated cross-tenant isolation tests run live, so you see the database refuse cross-tenant reads and writes instead of reading about it in an architecture diagram.
- 02
Inspect the writeback preview
Before Wave writes anything, the console shows exactly which fields it would write per contact, with suppressed people visibly withheld. You review the writes before they exist.
- 03
Flip the kill switch
Watch writeback stop instantly when the per-tenant switch flips, and confirm nothing reaches your CRM while it is off.
- 04
Read the audit trail
Walk the write log: previous value, new value, model version, timestamp, and the rollback path for any write Wave has made.
Six receipts
Six more things you can check, in plain terms.
Each one is a control you can ask us to show you, and the artifact that proves it.
- 01
Off by default.
Every capability that writes to your CRM starts switched off. You turn each one on, one at a time, after seeing a preview of exactly what it would write. Two off switches sit above that: one for your account, one for the whole platform.
Receipt: The writeback preview and the per-capability switches, shown live.
- 02
Wave proves its own lift, and says when it cannot.
Wave splits your contacts into a group that gets its predictions and a control group it leaves alone, then compares the two every day. It reports the difference with a confidence range and a minimum sample size. If there is not enough data to call it, it says so instead of printing a number that looks better than it is.
Receipt: The lift methodology whitepaper in the security pack.
- 03
Autopilot is earned.
Nothing in Wave sends on its own today. Every play is handed to a person, and the one send that exists is a single, human-clicked email from your own domain, off until you turn it on. Automation in Wave is earned one step at a time: the content has to be there, the committee defined, the motions aligned, and the control group has to say it works. Then, and only then, does a step go on autopilot, with an off switch you hold. Wave's HubSpot connection is tested on every build to make sure it cannot publish or send.
Receipt: The build test, shown on request.
- 04
Tested against prompt injection on every release.
Wave points AI at your CRM and at web pages, so someone could try to hide instructions to trick it. Wave runs more than a hundred of these attacks against its own defenses on every code change. If one gets through, the release is blocked.
Receipt: The red-team results summary in the security pack.
- 05
Every quote is real.
When Wave researches an account, every quote it shows has to match a page Wave read. If it cannot find the words on the page, the quote is dropped, and Wave tells you why.
Receipt: The drop reasons on every research finding.
- 06
When Wave does not know, it says so.
A person with too little history gets no prediction rather than a confident-looking guess. Reports withhold any reading based on fewer than five people. When Wave says 70%, it means about 70%, and it re-checks that claim against what really happened.
Receipt: The calibration report, available to every design partner.
What we do not do
Deliberate non-goals, stated up front.
Trust includes being clear about what Wave is not, and holding the line.
Your sources
Wave
HubSpot
Wave does not become your system of record. It is not a sending platform: no bulk sends, no sequences, no scheduled sends, no SMS, push, or ads. Your marketing platform stays the sender of record for campaigns. It does not train shared models across customers, and it does not sell or share your data. Onboarding can start read-only, and for teams that cannot grant API access at all, Wave onboards by secure file upload or a read-only warehouse view.
Related: No-API-Key Onboarding, AI Accuracy Eval Harness, AI Lift Experiments.
Where we are
Stage-honest: what we claim, and what we do not.
Tidal Wave is early access, working with design partners. We say exactly what is in place today versus on the roadmap, and we claim no certification until it is held.
What we do not claim
What we do claim
SOC 2 certification today.
A SOC 2 readiness program under way, with the controls above demonstrable live today.
A finished compliance story.
A DPA, subprocessor list, security review pack, and technical whitepaper, available on request.
Enterprise scale today.
Early access at design-partner stage, on an architecture built multi-tenant from day one.
Perfect AI.
Measured AI: accuracy scored continuously on your content, gated before changes go live, and lift proven by a built-in holdout.
FAQ
Questions security and procurement teams ask.
Can Wave send email on our behalf?
Today, only when a person clicks send, one recipient at a time, from your own domain, with unsubscribes honored at the moment of sending, and only after you turn that on. There are no sequences, bulk sends, or scheduled sends in Wave today. Our goal is for Wave to carry more of the sending burden over time, one earned step at a time, with a control group as the license and an off switch you hold. Your marketing platform stays the sender of record for campaigns.
Is my data isolated from other customers?
Yes, at the database layer. Every record is tenant-scoped and enforced by row-level security policies the application runtime cannot bypass, and automated isolation tests cover the boundary on every change. Wave also trains one model per tenant, so your data never trains anyone else's.
Do you train models across customers?
No. Wave classifies and predicts with one model per tenant. Your contact and engagement data never enters a shared or cross-customer model, and every prediction is stamped with the model version that produced it.
What happens when someone unsubscribes or opts out?
Suppression flows in from your systems and is enforced at the boundary: do-not-contact people are never written to, do-not-predict people are never scored. Suppression is a one-way ratchet, an opt-out can suppress a record, and nothing in Wave ever auto-un-suppresses it.
Can Wave's writes be undone?
Yes. Every CRM write is logged with the previous value, the new value, and the model version that produced it, and rollback is a first-class operation. Writeback is off by default until you opt in, and a two-layer kill switch stops it instantly, per tenant or platform-wide.
Can we onboard without granting API access?
Yes. For teams that cannot grant OAuth or API keys, Wave onboards by secure file upload or a read-only warehouse view, so the security review can say yes without a standing integration.
Are you SOC 2 certified?
Not yet, and we will not claim it until it is held. A SOC 2 readiness program is under way, and in the meantime security reviews get a live working session plus the security pack, so you can verify the controls directly.
Do you have a DPA and a subprocessor list?
Yes. The DPA, current subprocessor list, security review pack, and technical whitepaper are available on request.
How do I run a security review?
Request the security pack and we will schedule a working session: the isolation test suite run live, the writeback preview on a read-only connection, a kill-switch demonstration, and audit-trail inspection.
Does Wave crawl our website?
Only when a customer specifically asks Wave to research your company, and only your public website. Wave's crawler identifies itself as WaveBot/1.0, checks robots.txt on every visit and honors it immediately, makes no more than one request per second, reads only public pages on your own domain, and does not collect personal data about individuals from the public web. It is not a continuous, internet-wide crawler. See About WaveBot for the full policy and how to block it.
Verify it yourself
Request the security pack and DPA.
Book a session and we will run the review live: the isolation suite, the writeback preview, the kill switch, and the audit trail, on a working tenant.
Request the security pack