Click "Summarise themes with AI" to extract patterns from explanations.
RED
No summary yet.
YELLOW
No summary yet.
GREEN
No summary yet.
Flagged transactions
Date↕
Vendor
Category
Requester
Amount
Score
Explanation
Flag
Rationale
🗺️ Start from another city's procurement guideline
Browse a map of 251 US cities, pick one whose procurement rules match your jurisdiction, and use the AI to convert them into Active rules you can edit and apply. You can also upload your own.
Active rules
Each rule is sent to the AI on every classification. Toggle the switch to disable a rule without deleting it (useful for A/B testing). The "Source" tag shows where the rule came from.
What makes a good rule? (click for examples)
Good rules are specific, actionable, and declarative. They name dollar thresholds, who must approve, or vendor/category expectations. Examples:
"Marketing spend must reference the documented quarterly campaign plan or be flagged YELLOW."
"Office Supplies vendors (Staples, Amazon Business) cannot be justified as Marketing campaigns - flag RED."
"Conference and training expenses must reference an approved training plan or PO; otherwise YELLOW."
If you don't know where to start, click ✨ Draft with AI wizard above - describe a single spending concern in your own words and the wizard will draft a structured policy for you.
Click a group header to collapse / expand
No active rules yet. Click + Add rule manually to add one, or use the ✨ Draft with AI wizard button to have the AI propose one.
Per-category dollar caps (derived from Active rules)
Soft cap = warning level (YELLOW territory). Hard cap = should require explicit pre-approval (RED if absent). Editing a row rewrites the underlying rule in the Active rules panel above - it's a single source of truth.
Soft cap ($)
Hard cap ($)
Source rule
Grouped by category. Click a category header to collapse / expand. Adding a cap creates a new Active rule; removing a row deletes the corresponding rule.
Authorization matrix (derived from Active rules · "Proper authorization of transactions and activities")
Your approval cheat-sheet: who in the organization can approve what, by category and dollar threshold.
Click a group header to collapse / expand
This view is read-only - to change who authorizes what, edit the underlying rule (or the Per-category caps row above).
Documentation requirements (derived from Active rules · "Design and use of documents and records")
What documents must accompany each rule. Click × on a chip to remove a required doc; type in the input + press Enter to add one. Edits here update the underlying rule's metadata.
Required documents
Source rule
Grouped by category, default expanded. Common doc types: receipt, PO, invoice, contract, training plan, Concur entry, campaign plan reference.
Policy coverage (derived from Active rules)
A quick map of what your policies cover and when they kick in. A zero count usually means a gap worth a closer look.
When the policy kicks in
What the policy is doing
Disabled rules are excluded from these counts. Click ↻ AI on a rule to re-extract its classification.
Risk map (derived from Active rules · How likely × How big)
Where each policy sits on a 2×2 risk grid. The four corners show the standard response (Accept it / Transfer it / Control it / Avoid); policies whose declared approach disagrees with the quadrant's recommendation are flagged.
Small impact
Big impact
Happens often
Control it
Avoid
Rare
Accept
Transfer it
"Medium" likelihood/impact is plotted as "Rare/Small" for the grid. Rules without both classification fields appear below the grid; use ↻ AI on a rule to classify them.
Coverage gaps & AI suggestions (derived from Active rules · click "Draft a rule" to send the gap to the AI wizard)
Where your control framework has holes — categories with no authorization rules, function-types you've missed, anti-circumvention checks not yet written. Each suggestion can be sent straight to the AI rule wizard with the strategy pre-filled.
Analyzing your ruleset...
The deterministic gaps above are computed locally and refresh as you edit rules. The Deep AI analysis sends your full ruleset to the model for a narrative critique — about 3-5 seconds.
Detection sensitivity
How aggressively the system flags transactions in the first place. Higher values flag fewer (only the most extreme) transactions; the exact algorithm is set by the researcher.
● Unsaved changesSaved.
Researcher Dashboard
Algorithm control, system-effectiveness analyses, and experimental policy deployment · switch role
How often the department manager reclassifies a previous AI decision. High rates mean the AI / team-member process and the manager disagree a lot - either trust is low or the policy is unclear.
Each team member scored on how often their classified expenses end up RED (or get downgraded by a department manager). Top of the list = highest risk.
3. Event study - effect of rule changes
For each rule-change event, compares the average flag/override rate in the 30 days before vs the 30 days after. A negative Delta means the rule reduced anomalies.
4. Parallel-trends test (treatment vs control)
For each experimental treatment, compares the change in flag rate from pre- to post- treatment in the targeted group vs everyone else (control). The Diff-in-Diff column estimates the treatment effect.
5. Institutional efficiency indicators
Charter-school operational metrics computed from the activity log + program roster.
Active anomaly-detection algorithm
Team members and managers only see the sensitivity slider - they do not see which algorithm is
in use. Switching algorithms takes effect immediately for all users.
Recent events
Every override, classification, rule change, and treatment deployment is logged here for
analysis. Used as the source data for the event-study and parallel-trends tables in Pass 2.
When
Type
Actor
Target
Payload
Loading...
Experimental treatments
Apply a new rule to a subset of requesters/departments (or a random sample) and compare
their behaviour to a control group via the parallel-trends table on the Overview tab.
Started
Name
Target
Notes
Loading...
Institutional efficiency indicators
Computed from the event log and the bundled program roster. Click "Seed demo data" on the
Overview tab if these are blank.
Reclassify expense
Change password
Set a new password for the app. Other tabs/devices will be signed out.
Manage users
Add accounts for other people. Admin users can also manage other users; non-admins cannot.
Username
Role
Last password change
Add a user
Reset password
New experimental treatment
Pick a target group and the rule changes that should apply only to that group. The rest of the population becomes your control.
Create new business rule (AI-guided)
Describe the spending concern you want this policy to address. The wizard will ask a few clarifying questions, then propose a candidate policy for you to accept.
Describe the behaviour you want to flag in plain English - one specific situation, not a list.
The wizard will ask 2-4 short follow-up questions, then propose a structured rule you can accept.
Proposed rule additions
Supporting documents
AI Spending Check
A guided interview that classifies your expense without back-and-forth with your manager.
Attach a supporting document (PDF / image, up to 5 MB), or click Skip if you don't have one.
Uploading…
Procurement library
Pick a city's procurement guideline as a starting point. Green = rules already extracted, click to use. Yellow = source uploaded, click to extract. Grey = no source yet, click to add one.