All articles Comparisons

Self-Hosted vs SaaS Fraud Detection: What You Actually Control

The self-hosted vs SaaS argument is usually framed as control versus convenience. That framing skips the useful question: control over what? “Control” covers at least six separate things: what data leaves your systems, who can read the results, how long data lives, where it is processed, how decisions are made, and what it costs. A managed service hands some of these to the vendor and leaves the rest with you. Self-hosting keeps all of them, together with the work of running the service.

This post goes through each one, says what a managed service like Prynt lets you control today and what it does not, and ends with the cases where self-hosting is still the right answer.

The control map

What you want to controlSelf-hostedManaged service (Prynt today)
What leaves your systemsNothingDevice signals and the visitor’s IP reach the API; you decide what else to attach
Who can read each resultOnly youYour server, with the secret key; the browser gets a requestId or a sealed blob
How long data livesYour scheduleYour plan’s retention window, deleted automatically every day
Erasure and exportYour toolingErase by visitorId or linkedId, export from the console
Where it is processedWherever you run itUnited States, on servers Prynt operates; no other region today
How decisions are madeYour codeRules, weights and reason codes you can see and tune; not the code itself
CostInfrastructure plus peopleFlat monthly plans with an identification quota

Two rows are hard limits of any managed service: the region and the code. The other five are mostly about configuration and contract, and you have more say in them than the usual framing suggests.

What leaves your systems

A managed service has to receive device signals; that is how it identifies a device. What you control is everything around them:

  • Identifiers you attach. linkedId should be your opaque internal user id, never an email or phone number. tag should carry an action name such as signup, not personal data. See linking accounts after signup.
  • Optional inputs. Email and message content are analyzed only if you send them for those checks. If you don’t need email hygiene or content spam signals, don’t send either.
  • Consent. The browser agent honors Global Privacy Control by default, and requireConsent: true holds identification until your banner calls setConsent(true). A withheld request stores no signals and returns no visitorId. More in GPC and Do Not Track and consent banners.
  • IP addresses. In Prynt’s shared cloud, IPs are stored in full today, because VPN, proxy and reputation signals need them. Dedicated deployments on the Enterprise plan can store them truncated (a /24 or /48 prefix) or not at all. This is a deployment setting, not a switch in the console.
  • The reputation network. Accounts share one-way salted hashes of entities they confirmed as fraud, never identities or raw values, and any account can stop contributing from the console. See the reputation network.

Who can read each result

The browser never needs the verdict. It sends a requestId with the form, and your server fetches the result with the secret key:

curl "https://api.pryntid.com/v1/events/REQUEST_ID" \
  -H "Authorization: Bearer sk_live_your_secret_key"

If you don’t want even the visitorId or confidence to reach the page, turn on sealed results and gate identification on consent:

<script src="https://api.pryntid.com/cdn/prynt.umd.js"></script>
<script type="module">
  const prynt = await Prynt.load({
    apiKey: 'pk_live_…',
    endpoint: 'https://api.pryntid.com',
    requireConsent: true,   // nothing is identified until the visitor agrees
    sealedResult: true,     // the browser receives an encrypted blob, not the verdict
  });

  // Call this from your consent banner's "accept" handler.
  prynt.setConsent(true);
  const { requestId, sealedResult } = await prynt.identify({ tag: { action: 'signup' } });
  // Send requestId and sealedResult to your server; only it holds the sealing key.
</script>

The sealed result is AES-256-GCM encrypted under a key you generate in the console and keep on your server. Be precise about what that buys you: extensions, injected scripts and the visitor can’t read or alter the result. It does not hide anything from Prynt, which computes the result in the first place. Sealed results explained has the decryption code.

How long data lives, and how it is erased

Retention is set by plan and enforced by a daily deletion job: 30 days on Free, 90 on Pro, 365 on Scale, and as agreed on Enterprise. Database backups are kept for 7 and 14 days on two servers Prynt operates, so deleted data leaves the backups when they expire. You can erase a single device by visitorId, or every event, label and list entry that carries one of your user ids by linkedId, and export visitors and events from the console for access requests. The details are on the trust page.

The part only you control is your own copy. If your database keeps every requestId, verdict and risk score forever, the vendor’s retention window doesn’t help you. Store the decision you made and the reason, and apply your own expiry. Data minimization for fraud signals covers what to keep.

Where it is processed

This is where a managed service is least flexible. Prynt’s API, database and console run in the United States on servers Prynt operates, not on a third-party cloud. The sub-processors are listed on the trust page: Cloudflare for DNS, CDN and TLS (it sees request metadata in transit, including IPs), Stripe for billing and a transactional email provider. There is no EU or other regional processing today.

Under GDPR that means a transfer to the US, covered by the DPA. For many teams that is a normal vendor review. If your rule is “no device data leaves the EU” or “no processing outside our country”, Prynt does not meet it today, and no amount of configuration changes that. Self-hosting for data residency goes through those rules in detail.

How decisions are made

Self-hosting gives you the code. A managed service gives you, at best, the reasoning. In Prynt each decision carries a suspect score, reason codes and the signals behind them, and you can add your own rules, adjust risk weights and report outcomes so labels feed back into scoring (the outcomes feedback loop). What you can’t do is read or change the detection code: Prynt is not open source.

For most teams, explainability per decision is what auditors and disputing customers ask for. If you need to change how a signal is computed, or prove to a regulator line by line how a model works, that requires code you own.

What it costs

Self-hosting costs servers plus people: someone to deploy and patch the service, carry the pager when the fraud check sits on your login path, and keep detection current as bots and anti-detect browsers change. That last item is the one teams underestimate. Detection that isn’t updated degrades quietly.

A managed service moves that work to the vendor. Prynt prices with flat monthly plans that include an identification quota (pricing). There is no per-call overage: at the quota, identification returns 429 quota_exceeded until the next month or an upgrade. A bot flood can’t run up your bill, but it can use up your quota, so leave headroom and gate abusive traffic at the edge. The build vs buy guide models both sides.

When self-hosting is still the right answer

Be honest about these cases. A managed service doesn’t fit when:

  1. A law, regulator or contract forbids sending device signals to a third party, or out of a specific country. That decides the question.
  2. You need processing in a region the vendor doesn’t offer. For Prynt today, that is anywhere outside the US.
  3. The system must run without outbound calls, for example in air-gapped or on-premise environments.
  4. Detection is your product. If you have the team and the volume to treat fingerprinting and bot detection as a core competency, owning the code is worth the cost.

In those cases the options are building on open-source fingerprinting libraries and doing the server, matching and detection work yourself, or buying from a vendor that sells an on-premise edition. Either way, you own detection freshness.

A short decision checklist

  • Does any rule forbid sending device signals to a vendor or out of the country? If yes, self-host.
  • Do you need processing outside the US? If yes, Prynt is not an option today.
  • Can you get the controls you need from the vendor? Opaque ids, consent gating, sealed results, retention, erasure, a DPA. If yes, a managed service is usually less risk for less work.
  • Who would own a self-hosted service at 3 a.m.? If nobody, that answers it.

Frequently asked questions

Is self-hosted fraud detection more secure than SaaS?

Not by default. Self-hosting keeps device signals inside your network, but security then depends on how well you patch, monitor and staff the service. A managed service that receives only what it needs, hands the browser nothing it can read and deletes data on a schedule can be the lower-risk option for a small team.

What can I control when fraud detection is a managed service?

What you attach to each request, who can read the result, how long data is kept, how it is erased and exported, and the contract that governs it. With Prynt that means opaque account ids instead of emails, consent gating, sealed results, plan retention enforced daily, erasure by visitorId or linkedId, and a DPA. You do not control the processing region or the detection code.

When is self-hosting the right choice?

When a law, regulator or contract says device signals may not leave your infrastructure or a given country, when you need processing in a region your vendor does not offer, when the system has to run without outbound calls, or when you have the team to run detection as a product of your own.

Self-hosted vs SaaS is not one decision but six. Write down which kinds of control you actually need, check each against what a managed service offers, and self-host only for the ones it can’t meet. If those are none, start free and read the docs; if residency is the blocker, read why teams self-host fraud detection first.

Try it free

Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.

Keep reading