From sign-in to a board-ready ARB package

ArchReview AI is multi-tenant: your organization is one tenant, and everyone you invite shares its systems, reviews, and billing plan. Sign-in is Google OAuth only — there's no separate ArchReview AI password. From the login screen, choose Continue with Google.

The first person from a new organization to sign in automatically becomes that organization's tenant, as its Admin. Everyone who signs in afterwards with a different Google account also gets their own new tenant, unless you invite them into yours — there's currently no "join my organization" flow, so keep every teammate signing in with a shared understanding of who's who.

• From the login screen, click Continue with Google
• First sign-in for a new org becomes Admin automatically

Your landing page rolls up every review in the tenant: active versus completed review counts, the average score across every completed review, how many distinct frameworks have been evaluated tenant-wide, a score trend chart by month, and a pipeline breakdown by status.

A brand-new tenant with no reviews yet sees a three-step checklist instead of empty charts: create a review, run it, and export a report.

• New tenant? Follow the three-step checklist on first load
• Otherwise, the dashboard is where every review rolls up automatically

A review always belongs to a system — a piece of your architecture you're evaluating. Creating one is a three-step wizard: name the system, pick its domain and organization size (fed to every framework engine so scoring accounts for scale); multi-select any combination of the six automated framework engines, keeping in mind GCC regulatory compliance always runs regardless of which you pick; then upload as many source documents as you have.

Extraction progress is shown per file, and once it finishes you'll see an estimated review time before you commit to running it. Re-running a review against the exact same set of documents doesn't re-spend an AI call — ArchReview AI hashes the combined document content and reuses the cached result if nothing changed.

• Name the system, choose domain and organization size
• Multi-select frameworks — GCC compliance runs either way
• Upload .pdf, .docx, .pptx, .png or .jpg files

Six frameworks run as fully automated scoring engines, each producing a 0–100 score, a written summary, and a list of findings — all selected engines run in parallel, so choosing more frameworks doesn't multiply your wait time.

• TOGAF — scored phase-by-phase across all 8 ADM stages against expected deliverables
• TM Forum — eTOM process alignment, SID data model coverage, TAM mapping, Open API conformance
• BIAN — service domain boundaries and standard service operations for banking architectures
• SABSA — risk-driven security: trust models, policy architecture, traceability to control
• AWS Well-Architected — reliability, security, cost, and performance among its pillars
• ITIL v4 — service value chain maturity, incident/change/problem management, CMDB discipline

The Framework Library page also lists six reference frameworks — NIST CSF, DAMA-DMBOK, ISO/IEC 42010, Zachman, COBIT, C4 — which inform how findings get written but don't yet run as independent scoring engines.

Every review — no matter which frameworks you picked — is also checked against Gulf Cooperation Council regulatory expectations, producing its own 0–100 GCC score. Four checks run as plain rule logic, not an AI guess, so they're consistent every time: UAE data residency (citing TDRA Decision No. (47) of 2020), Arabic-language / RTL support for UAE/KSA-serving systems, Saudi NCSC approved cloud providers for KSA-context systems, and GCC cloud region residency.

Three additional checks need judgment rather than pattern matching, so Claude reviews the documentation directly for them: Emiratisation workforce commitments, government data classification handling, and ADSIC (Abu Dhabi Systems & Information Centre) alignment. GCC compliance appears as its own dedicated section in every review report.

Once every framework engine and the GCC check finish, ArchReview AI assembles one report: a three-paragraph executive summary in plain English, a visual scorecard (an overall gauge plus a radar chart comparing every framework side by side), findings grouped Critical → High → Med → Low — each with a framework badge, clause reference, evidence quote, and recommendation — and a prioritized remediation plan with an effort estimate and suggested owner per fix.

Export a fully branded .docx (scorecard table, every finding, remediation plan as a numbered list) or a tight 5-slide .pptx deck. The full report is cached against the review's document content — re-opening or re-exporting it never re-runs the underlying AI analysis.

• Read the executive summary and scorecard first
• Work findings top-down, starting at Critical
• Export as .docx or .pptx to share outside the app

Findings tell you what's wrong; this turns them into a redlined draft of what "fixed" actually reads like. From any completed review, Generate Suggested Solution asks Claude to redline your submitted document directly — locating the passage a finding applies to, quoting it as the original, and proposing a specific, actionable replacement. Where a finding needs an entirely new section with no counterpart in your original document, it's labeled New section instead of inventing a fake quote.

Each redline shows the original passage (or "no counterpart"), which finding(s) it resolves, the suggested revision, and a one- or two-sentence rationale. Download the result as a branded .docx — generation is cached per review, so downloading again doesn't regenerate it.

Turn a completed review into what your Architecture Review Board actually needs to see: an arb-submission.docx with context, an architecture proposal summary, alternatives considered (pulled from your own documentation, never invented), the decision required, plus the full ArchReview AI report attached as Appendix A — and a risk-register.xlsx with one row per Critical/High finding, its severity, an owner and target date left blank for you to assign, and the mitigation action.

Once submitted, an Admin can record the board's decision directly on the review — Approve, Reject, or Defer — with optional notes, timestamped and attributed.

Every system in your tenant, at a glance: each shows its latest review as a color-coded badge — green above 80, amber 60–80, red below 60 — a trend arrow against its previous review, the last-reviewed date, which framework was used, and its count of open Critical findings.

Click into any system for a full timeline of every review it's had, plus a trend line chart plotting overall score by review date, one line per framework, so you can see exactly where the architecture improved and where it stalled. Filter the portfolio by domain, score range, framework, or "has a Critical finding open," and export the whole filtered view as a CSV.

Four roles exist on every tenant — Admin (full control, including recording ARB decisions and managing billing), Architect (creates systems, runs reviews, generates deliverables), Analyst (reviews findings, contributes to remediation), and Viewer (read access to reviews, reports, and the portfolio). Today, the one place role actually gates an action is recording an ARB decision; the rest is the intended shape of each role as your team grows into it.

Three plans: Starter ($499/mo, 5 reviews/month, all six framework engines, GCC checks), Pro ($1,499/mo, unlimited reviews, the ARB packager, Suggested Solution, Word & PowerPoint export), and Enterprise (from $5,000/mo, everything in Pro plus white-label branding and API access). On Starter and hit your 5-review monthly limit? A single extra review is $300 as a one-off credit — no need to upgrade the whole plan to finish the month.

• Downgrading a plan never removes deliverables already generated — only new Pro-tier generation going forward