Skip to content
architecture

CIAM Reference Architectures: Four Patterns and the Vendors That Fit

Updated · 14 min read · By

On this page

Key takeaways

  • There is no single best CIAM architecture. There are four common patterns chosen by identity shape and constraints.
  • B2C mobile-first optimizes for conversion, passwordless, and cost per MAU at volume.
  • B2B multi-tenant lives or dies on the Organizations model, per-tenant SSO, and SCIM.
  • Hybrid B2B2C needs two clean identity planes, not one tangled user pool.
  • Regulated and self-hosted is decided by data residency and a compliance floor first; the real cost is operations.

What is CIAM architecture?

The layers of a CIAM architecture. Customer channels reach identity services through a login experience; identity data feeds downstream systems through APIs, webhooks, and SCIM. A risk engine scores events at the identity-services layer.
The layers of a CIAM architecture. Customer channels reach identity services through a login experience; identity data feeds downstream systems through APIs, webhooks, and SCIM. A risk engine scores events at the identity-services layer.
  • Experience layer. Hosted login pages, native SDKs, and embedded components. This is where conversion is won or lost.
  • Identity services. Authentication (passkeys, MFA, social, enterprise SSO), plus the session and token service that speaks OIDC and OAuth 2.1.
  • Identity data. The user and organisation store, profile attributes, and consent records. Its schema outlives any vendor, so model it deliberately. See identity data modeling.
  • Risk and fraud. Signals and scoring for signup, login, and account changes. See AI fraud detection in CIAM.
  • Authorization. Roles, organisations, and fine-grained policy for what a signed-in customer may do. See API authorization patterns.
  • Integration layer. Management APIs, webhooks, and SCIM that push identity events to CRM, CDP, analytics, and the SIEM.

CIAM architecture principles

  1. Standards over proprietary flows. Standard protocols keep applications portable. A migration off a vendor that only speaks its own SDK is a rewrite; a migration between OIDC providers is configuration.
  2. API-first. Every capability, including registration, profile updates, consent, and admin, is reachable through documented APIs. A hosted UI is a convenience on top, not the only path.
  3. Authentication and authorization are separate concerns. The CIAM proves who the customer is and issues tokens; applications and policy engines decide what they may do with them.
  4. Privacy by design. Progressive profiling instead of long forms, purpose-bound consent, and data residency decided up front. GDPR turns these from good practice into obligations.
  5. Built for peaks. Customer traffic spikes on launches, sales, and outages elsewhere. Validate tokens at the edge, cache signing keys, and plan multi-region failover. The high-scale CIAM guide covers the numbers.

Architectural styles: monolithic, microservices, API-first, serverless

StyleWhat it looks likeWhen it fits
MonolithicLogin, profile, and admin in one deployable, often a self-hosted productSmall teams, regulated self-hosting, predictable load
MicroservicesSeparate services for authentication, tokens, profile, and consentLarge platforms that scale and release each capability independently
API-first SaaSA managed CIAM exposed through APIs and SDKs, with optional hosted UIMost teams; the default for new builds
Serverless extensionsFunctions or hooks that run on identity events (pre-signup, post-login)Custom logic such as fraud checks, enrichment, and CRM sync

These are not exclusive. The common 2026 shape is an API-first managed CIAM, extended with serverless hooks, fronted by a custom UI, and feeding a microservices back end with tokens.

How to use this guide

Match your situation to one of the four patterns, confirm the non-negotiable capabilities, and then take that capability set to the vendor index and comparisons. If you are not yet sure which pattern you are in, work through Start here first. Each pattern below names vendor types rather than a single winner, because cost behavior and constraints still decide the final choice.

Pattern 1: B2C mobile-first

High-volume consumer identity where every extra step costs conversion and cost-per-MAU dominates the bill.

 Mobile / web app
       |
   [ CIAM ]  --- social login, passkeys, magic links, email/SMS OTP
       |            |
   sessions     fraud / bot signals (device, velocity, risk)
       |
   profile + consent  --->  marketing / CDP, analytics

Non-negotiable capabilities. Passwordless and passkeys as the primary path; social login at scale; strong registration conversion; bot and fraud signals; and a pricing model that stays sane past 100k MAU. Tenancy is flat, so the Organizations model is irrelevant.

Failure mode. Choosing a B2B-leaning platform "to be safe" and watching the per-MAU bill become the largest line item by the time you reach scale. The second failure is treating passkeys as a checkbox rather than designing the enrollment and conditional-UI flow that actually drives adoption.

Vendor type that fits. Developer-first and B2C-specialist platforms, with hyperscaler-native and open-source options once cost dominates. See the consumer-apps vertical and the passwordless guide. When cost is the binding constraint at volume, weigh the Auth0 alternatives and Firebase alternatives.

Pattern 2: B2B multi-tenant SaaS

Your customers are companies. Each is a tenant with its own users, roles, and, once you sell upmarket, its own identity provider.

                 [ CIAM ]
                /    |    \
        Org A      Org B      Org C
        |  \        |          |
     users  SAML   users      users
            IdP    OIDC IdP   (password/passkey)
        + SCIM provisioning per org
        + per-org roles, policies, audit

Non-negotiable capabilities. A first-class Organizations / tenant model; per-tenant Enterprise SSO (SAML and OIDC) with self-service IdP setup; SCIM for user provisioning and de-provisioning; per-tenant roles and policies; and tenant-scoped audit logs. This is the pattern where the Organizations and tenants and enterprise SSO guides are required reading.

Failure mode. Bolting tenancy onto a flat B2C platform after the first enterprise deal demands SSO. Retrofitted multi-tenancy leaks: shared user pools, no per-tenant policy isolation, and SSO that an admin cannot configure without your engineers. The cost shows up as deal friction and security-review findings.

Vendor type that fits. B2B-first platforms and enterprise-readiness layers built around Organizations, SSO, and SCIM as primitives. See the B2B SaaS vertical and the multi-tenant architecture guide.

Pattern 3: Hybrid B2B2C

You serve consumers directly and you serve business tenants whose end users also log in. The mistake is collapsing both into one pool.

   Consumer plane                 Business plane
   (direct B2C users)             (tenant end users)
        |                              |
        |                          Org A / Org B ...
        |                          per-org SSO + SCIM
         \                            /
          \                          /
            [ CIAM with two planes ]
            shared primitives, separate tenancy + policy

Non-negotiable capabilities. Everything from Pattern 2, plus a clean way to run a consumer plane alongside the tenant plane without identity collisions: separate user directories or strict tenant scoping, distinct branding and flows per plane, and an authorization model that never lets a consumer identity resolve into a business tenant or vice versa.

Failure mode. One tangled user pool where a consumer who later joins a business tenant becomes two conflicting identities, or where tenant isolation depends on application code instead of the platform. Account linking and policy boundaries are the hard part.

Vendor type that fits. Platforms with a strong Organizations model that also handle high-volume B2C cleanly, or an architecture that pairs a B2B enterprise-readiness layer with a B2C platform. This is the pattern most likely to justify a two-component design; confirm the seam before committing.

Pattern 4: Regulated or self-hosted

Data residency, sovereignty, or a compliance floor (FedRAMP, HIPAA, strict in-region storage) sets the architecture before any feature does.

   Your cloud / region / VPC
        |
   [ self-hosted or compliant-cloud CIAM ]
        |          |              |
   data stays   audit /        HA topology you
   in-region    evidence        own and operate
        |
   identity verification / step-up where mandated

Non-negotiable capabilities. The specific compliance attestations you require, in-region or in-account data storage, exportable tamper-evident audit logs, and (often) the ability to self-host. Strong step-up and risk-based authentication are common in this pattern; see adaptive risk-based authentication.

Failure mode. Underestimating operations. Self-hosting trades a vendor invoice for on-call, patching, upgrades, and HA you own. The other failure is assuming a SOC 2 badge satisfies a FedRAMP or HIPAA requirement; the floor is specific, and the capability matrix scores each attestation separately.

Vendor type that fits. Open-source self-hosted platforms for maximum sovereignty, or compliant-cloud vendors with the exact attestations. See the open source CIAM alternatives, the government vertical, and the financial-services vertical.

Choosing between patterns

If your situation isYour patternThe capabilities that decide it
Consumer app, high volume, cost-sensitiveB2C mobile-firstPasskeys, social, fraud signals, cost per MAU
Selling SaaS to companiesB2B multi-tenantOrganizations, per-tenant SSO, SCIM
Both consumers and business tenantsHybrid B2B2CTwo clean planes, account linking, tenant isolation
Residency, sovereignty, or compliance floorRegulated / self-hostedAttestations, in-region data, self-host, audit

Pin your pattern, carry its non-negotiable capability set into the vendor index, and let cost behavior and your hard constraints break the final tie. Every rating behind these recommendations is explained in the methodology.

Related guides

Related vendors

Where to next

FAQ

What is CIAM architecture?
CIAM architecture is the set of components and data flows that handle customer identity. It covers the login experience, identity services for authentication and tokens, a user and consent store, a risk engine, and an integration layer that feeds CRM, CDP, and analytics. It is built for millions of external users, self-service, privacy law, and conversion, unlike employee-focused IAM.
What are the core components of a CIAM architecture?
Six components cover most deployments. They are an experience layer (hosted login and SDKs), identity services (authentication, MFA, SSO, sessions, and tokens), and an identity data store with consent. The other three are a risk and fraud engine, authorization, and an integration layer of APIs, webhooks, and SCIM.
What is API-first CIAM?
API-first CIAM exposes every identity capability, including registration, login, profile, consent, and administration, through documented APIs before any UI is built on top. It lets teams build native login on any channel, automate identity operations, and swap front ends without touching the identity core. Most modern CIAM platforms are API-first with an optional hosted UI.
How is CIAM architecture different from IAM architecture?
IAM architecture is built around a known workforce, a central directory, and access to internal systems. CIAM architecture is built around self-registration, millions of external users, social and passwordless login, consent and privacy controls, and sudden traffic peaks. The protocols overlap, but scale, UX, and data-protection requirements differ.
Can one platform serve more than one of these patterns?
Yes, several can, but check the seams. A platform strong at B2B Organizations can usually also do flat B2C, and a few do hybrid B2B2C cleanly. The risk is the reverse: a B2C-first platform asked to grow real B2B tenancy, SSO, and SCIM tends to express those as add-ons rather than primitives, and the gaps show up under enterprise procurement.
Do I need to pick the architecture before the vendor?
Pick the pattern first. The pattern tells you which capabilities are non-negotiable (Organizations and SCIM for B2B, cost-per-MAU and passkeys for B2C, residency and self-host for regulated), and that capability set is what narrows the vendor list. Choosing a vendor first tends to bend your architecture toward what that vendor happens to do well.
Where do passkeys fit across these patterns?
Passkeys are primary-auth in all four, but they matter most in B2C mobile-first, where conversion and phishing resistance are direct business metrics. In B2B and regulated patterns they are increasingly a compliance and security expectation. Synced passkeys give consumer-grade UX at AAL2; device-bound passkeys give higher assurance with more operational overhead.
How does agentic identity change these architectures?
It adds a fourth principal type. Until recently you modeled users, tenants, and machine-to-machine clients. AI agents acting on behalf of a user need scoped, attenuated tokens distinct from the user's own session, plus an audit trail that separates agent actions from human actions. In each pattern below, treat agent identity as a first-class principal rather than reusing a user or service token.

Sources

  • CIAM Compass vendor index and capability matrix
  • FIDO Alliance passkey and AAL guidance
  • OAuth 2.1 and OpenID Connect specifications
  • SCIM 2.0 (RFC 7643 / 7644)
Last reviewed 2026-09-28.