Opaque
a score nobody can unpack
When a score cannot be broken down, the review then becomes a debate concerning trust.
Agentic AI · Nova
Nova Ask it. It does the work. You keep the final say.
Nova works on every page of the platform and inside the AI tools your team already uses. Give it any manual task in your own words, ask it anything about your data, and it runs the task. It also reads your screening profiles, rules, customers, and monitoring schedules, then returns tuning suggestions with the evidence attached. Settings never change without your approval.
Ask Nova
Ask it in your words. Nova does it and shows its work.
Request the five highest-risk customers for this week, or instruct Nova to assign a case and set its status; the system will carry out the task, state its findings and those things it could not confirm, and suggest the next step.
Explained
Every part of the score
Linked
Back to the source record
Proposes
Before anything bulk or permanent
Kept
Every command and suggestion logged
Why this exists
Set once at go-live, worked by hand ever since
Profiles and rules are configured at launch by people who have since moved on. Nobody touches them, because nobody can predict what a change will do. Meanwhile analysts click through the same menus to pull the same reports. Nova takes on both: the tuning nobody dares start and the manual steps nobody should repeat.
Untuned
set at go-live, left alone
A parameter isn't altered if its effect can't be measured.
Undocumented
the why was said, not written
The reason for setting a threshold was discussed in a meeting that produced no minutes.
How it works
Scroll to advance
Find it, size it, explain it, hand it back
-
Step 01
Detectors count what already happened
Deterministic detectors proceed by going through the screening profiles, applying the rules, taking the dispositions, calculating the customer scores, and checking the monitoring schedules. At this stage, no model makes a call. These are facts derived from your own data.
-
Step 02
Your baseline and the sector filter it
A signal is only important if it stands out; Nova compares each one against your personal history and anonymized sector benchmarks and removes the normal ones.
-
Step 03
Every finding comes with its proof
The items discovered, why they matter, the supporting records, Nova's confidence, and the anticipated impact on false positives. All the information an approver needs is contained on a single card.
-
Step 04
Your team makes the call
When Nova reaches the proposal, approving it opens the draft rule or pending setting change in your normal review process; if it is rejected, the reason is sent back, so the same suggestion won't reappear unchanged.
nova / detect
- Alerts read14,206
- Rules checked38
- Customers21,480
- Signals raised96
deterministic detectors · your data only.
nova / benchmark
- Signals in96
- Above baseline41
- Above sector24
- Kept16
own baseline + anonymised sector medians.
nova / explain
- Findings16
- With evidence16
- Confidence scored16
- FP estimated11
evidence + rationale + confidence + expected impact.
nova / hand over
- Approveddrafts only
- Rejectedwith a reason
- Deferredreturns later
- Loggedall three
maker-checker · nothing applied automatically.
Detect
Benchmark
Explain
Hand over
In your workflow
Nova works where your team already works
You don't need a second login or a copy-paste procedure; Nova detects the context of the screen you're on, carries out what you request, and then returns the decision.
What it looks at
Everything behind the answer
Score decomposition
Each point in the score traces back to its module, weight, and source.
Evidence references
The evidence points to the record rather than paraphrasing it.
Threshold proposals
A suggested value and the effect it has on alert volume. This can only be applied following approval.
Suggestion history
Each proposal, whether it was accepted or rejected, and when.
Original language kept
The source content is presented alongside any translation, regardless of the original language.
Works inside the case
The answer is located beside the decision it relates to. There is no separate tool needed.
Guardrails
Your team holds the final yes
Whether you approve, reject, or defer, all of your choices are recorded; and when a suggestion is rejected, the reason for the rejection is sent back to Nova so that the same suggestion will not afterward reappear as it was.
Maker-checker
3
Audit
1
No suggestions for this product right now.
Approve: it opens as a draft
When a suggestion is approved, it becomes a draft rule or a pending change to a setting, along with the evidence, and then proceeds through your normal change process.
Rule
Confidence 100%
Rule-based
Why Nova flagged this
Nova can write drafts. It cannot change live configuration.
Evidence
- outcome draft created
- applied by your approver
- audit event written
Reject: the reason stays with Nova
A short note is added to the finding when a rejection occurs, so the next reviewer knows the reason and Nova stops raising that point.
Setting
Confidence 100%
Rule-based
Why Nova flagged this
A suggestion engine becomes noise within a month without feedback.
Evidence
- reason stored on the finding
- repeat suppressed
- audit event written
Defer: paused, not dropped
An answer isn't too late right now; the result will appear in the next run window rather than remaining unread in the queue.
Setting
Confidence 100%
Rule-based
Why Nova flagged this
The suggestions have the same effect on a team as alerts do, so Nova keeps a record of what was seen and when.
Evidence
- outcome snoozed
- returns next run window
- audit event written
Every run stays on record
With the regulator pack, all analysis runs, all findings they produced, and all decisions your team made are saved and exported.
Setting
Confidence 100%
Rule-based
Why Nova flagged this
When an auditor asks why a threshold moved, the response is the finding, the evidence, the approver, and the date.
Evidence
- run id, window, detector versions
- finding evidence + confidence
- decision who, when, why
Approve · Reject with a reason · Defer — all three are logged
Agentic AI
The rest of the agent suite
FAQ
Before you switch Nova on
What is Nova?
An assistant that works on every page of the platform and inside the AI tools your team already uses. You give it tasks and questions in your own words; it reads the context of the screen you are on, carries out the request and shows its work.
What can I ask Nova to do?
Manual tasks such as listing this week's five highest-risk customers, assigning a case or setting its status. Nova carries out the task, states what it found and what it could not confirm, and suggests the next step.
Does Nova act without asking?
It runs your request, but proposes first before anything bulk or permanent. A single case assignment happens when you ask; a change across many records waits for your confirmation.
Can Nova change a live setting?
No. Nova can write drafts but cannot change live configuration. An approved suggestion opens as a draft rule or a pending setting change and goes through your normal change process, with the approver recorded.
What does Nova look at when it suggests tuning?
Your screening profiles, rules, dispositions, customer scores and monitoring schedules. These are usually configured at go-live and left untouched, because nobody can predict what a change will do. Nova measures that before anyone decides.
How does Nova decide what is worth suggesting?
Deterministic detectors count what already happened in your data, and no model makes a call at that stage. Nova then compares each signal with your own baseline and anonymised sector benchmarks, and keeps only the ones that stand out.
What comes with each suggestion?
The finding, why it matters, the supporting records, Nova's confidence and the expected impact on false positives, all on one card. Evidence references point to the record itself rather than paraphrasing it, so the approver can check every claim.
Can Nova break down a customer's score?
Yes. Each point in the score traces back to its module, weight and source record. A review then starts from what built the score, not from a debate about whether to trust it.
What happens when we reject a suggestion?
The reason is stored on the finding, so the same suggestion does not come back unchanged. The next reviewer sees why it was turned down, and Nova stops raising that point. A deferred suggestion returns in the next run window instead.
Can an auditor see Nova's work?
Yes. Every run, finding and decision is recorded, with detector versions, evidence, confidence, and who decided and when. The history is exported with the regulator pack, so when an auditor asks why a threshold moved, the answer is already on file.
Let Nova look at your own queue
We run it against a sample of your screening and monitoring history and walk through what it found.
Minutes
To first rule
No code
To change one
Dry run
Before it touches a case