Q4 2026 pilot cohort. Applications close 30 September 2026.

Product

Solutions

Resources

Pilot applications close 30 September

Product

Private Banks

Solutions

Family Offices

Family Offices

External Asset Managers

Independent Advisors

Trust Center

Resources

Blog

Q4 2026 pilot cohort. Applications close 30 September 2026.

Product

Solutions

Resources

The productSynthetic data throughout

Your book, understood.

Luscent reads the communications your firm already has, shows every relationship manager the next right conversation, and generates the suitability evidence as the work happens.

Reads
Mail, calendar, meetings
Writes back
Nothing
Health Score
0 to 100, daily
Compliance confidence
Always 100
Model inference
Scaleway EU platform
Non-EU model APIs
None

[01] The book

127 relationships · ranked by what needs attention
#ClientHealthWhat movedAUM
01De Vries Holding54Consolidation intent61.200.000 EUR
02Klein Family Office61Review declined twice42.000.000 EUR
03Weiss & Partners74Suitability review due128.400.000 EUR
04Mercier Patrimoine81Target market breach87.300.000 EUR
05Lindqvist Family84Next generation active33.900.000 EUR
06Aaltonen Holding88Reporting date met214.600.000 EUR
07Bergmann Privat92No change76.500.000 EUR
Synthetic data. Health Score 0 to 100, low is attention. 3 of 127 shown as needing it today.
[02] The scoreFig. 1

One number per relationship, and you can open it.

Every relationship carries a Health Score from 0 to 100, rebuilt daily from the communications your firm already produces. Low means attention, not risk. The number is only worth having because it resolves: every score opens onto the four or five things that moved it, and each of those opens onto the messages they were drawn from.

Fig. 1Product: Signals, Health Scores and Evidence | LuscentA Health Score of 62, down 11 over thirty days. Response time contributed minus 14, who opens the exchange minus 11, meeting cadence minus 9, and the suitability file nothing because it is complete and current. Every contribution names the observation behind it. Synthetic data.KLEIN FAMILY OFFICE · HEALTH SCORE 62 · DOWN 11 IN 30 DAYS · SYNTHETICResponse time-14Doubled since AprilMeeting cadence-9No review booked in 92 daysWho opens the exchange-11Last three opened by the clientSuitability fileno changeComplete and currentContributions are what moved the score over the last 30 days. They are not a forecast.
A Health Score resolved into its contributions. Response time and who opens the exchange did most of the damage; the suitability file did none, because it is complete.

That is the difference between a score a relationship manager argues with and a score a supervisor can hold a conversation about. It survives being questioned, because the questioning is the point.

[03] Use case 01 · RetentionFig. 2

The client who is leaving tells you first.

A departure is a sequence, not an event. The sequence is in the mailbox for months before it is in the AUM report, and it is legible: replies get shorter, the client starts opening the exchanges, a competitor gets named, a review gets declined twice.

Fig. 2Product: Signals, Health Scores and Evidence | LuscentSynthetic. Replies shorten on day nought, the client begins opening the exchanges by day 34, a competitor is named on day 62, a second review is declined on day 96, and the transfer request arrives on day 141. The first four are legible in the mailbox; the fifth is the one the AUM report sees. 141 days separate the first from the last.DAY 0Replies get shorterENGAGEMENT · 71DAY 34Client opens the lastthree exchangesCHURN · 84DAY 62A competitor is named inpassingPATTERN · 88DAY 96Second review declinedCHURN · 91DAY 141Transfer requestAUM REPORTSynthetic. The shaded span is the part that is legible in the mailbox: 96 days before the transfer request arrives.
A departure drawn as the sequence it is. The distance between the first legible signal and the transfer request is the whole opportunity.

Luscent surfaces the sequence while it is still a sequence. What the relationship manager does about it is theirs.

[04] Use case 02 · GrowthFig. 3

The money is already in the conversation.

Liquidity events, successions, an inheritance, a business sale, a tax question that only makes sense if something is about to happen. Clients say these things in passing, to someone who is reading for something else, and they do not come back.

Fig. 3Product: Signals, Health Scores and Evidence | LuscentSynthetic. Five opportunities drawn from client conversations, plotted by estimated value in millions of euro against the system's confidence. A business sale at 4,2 million with confidence 88 and a consolidation at 3,4 million with confidence 76 sit in the shortlist; a tax question at 0,9 million with confidence 62 does not. A large number the system is unsure about is not a lead.SHORTLIST5060708090CONFIDENCE1234EUR MBusiness sale mentionedDe Vries Holding4,2M · 88Inheritance in progressLindqvist Family2,6M · 81Consolidation intentWeiss & Partners3,4M · 76Synthetic. Three of five clear the shortlist. A large number the system is unsure about is not a lead.
Opportunities plotted by what they are worth against how sure the system is. The shortlist is the top-right corner, and it is deliberately small.

Ranked by what it is worth and how sure the system is, so a book of 127 relationships produces a list of three.

[05] Use case 03 · EvidenceFig. 4

The file writes itself, while the advice happens.

Suitability files get assembled afterwards, from memory, unevenly across a population of relationship managers. The reconstruction is the risk: it is slow, it is uneven, and at an inspection it is the thing that is hardest to defend.

Fig. 4Product: Signals, Health Scores and Evidence | LuscentA suitability file reconstructed after the fact is recalled weeks later, its basis inferred, its source not attached, and its quality uneven across a desk. A file written at the point of advice records the basis, attaches the messages it was drawn from, and has the same shape every time.RECONSTRUCTED AFTERWARDSRecalled, weeks laterBasis inferredSource not attachedUneven across the deskWRITTEN AS THE ADVICE HAPPENSWritten at the momentBasis recordedMessages attachedSame shape every timeThe difference is not effort. It is what survives an inspection.
The same advice event, recorded two ways. One of them is defensible at an inspection.

Compliance outcomes come from rules, never from a probability, and the rule that fired is on the record with its source.

[06] How it worksFig. 5

Seven stages, and no path around them.

The order is the product, not the list. Consent cannot be recorded after the reading. Identifiers cannot be removed after the model has already seen them. A rule cannot be checked against an output nobody explained. Skipping a stage is not a setting, which is why there is no switch for it anywhere in the product.

Fig. 5Product: Signals, Health Scores and Evidence | LuscentEvery signal passes through seven stages in order: consent, redact, infer, rules, explain, decide, audit. Inference is the only stage a model touches. Rules are deterministic, the relationship manager decides, and the audit log is immutable. There is no path around it.EVERY SIGNAL, EVERY TIME, IN THIS ORDERConsentLawful basisrecorded beforeanything is readRedactIdentifiersremoved beforeinferenceInferOpen-weightmodels, servedin the EURulesDeterministic.No modeldiscretionExplainEvidence andrule attached tothe outputDecideThe relationshipmanager, alwaysAuditImmutable log ofwhat happenedand whoThe dashed stage is the only one a model touches. Solid stages are deterministic or human, and neither can be skipped.Tabular models for structured data are on the roadmap and are not in this run today.
The spine, end to end. Inference is the only stage a model touches, and it is the only one drawn in dashes.

Compliance Signals carry a confidence of 100 because a rule produced them. Signals derived from communications carry a number, and the number is always on the face of the Signal rather than in a footnote.

[07] What the model is givenFig. 6

The shape survives. The identity does not.

A promise about redaction is worth less than the redacted string, so here is the string. This is a synthetic client message, and beside it is what Luscent actually sends to a language model.

Fig. 6Product: Signals, Health Scores and Evidence | LuscentSynthetic client message. As received it names the relationship manager, the client's son and the month of his birthday. As sent to a language model, every identifier is replaced with a placeholder while the structure of the message survives: a succession event is still legible, the people are not.AS RECEIVEDMarc, thank you for the quarterly pack.Before we meet I should tell you that my sonJulian has asked to be included in theseconversations. He turns thirty in March.AS SENT TO THE MODEL[NAME], thank you for the quarterly pack.Before we meet I should tell you that my son[NAME] has asked to be included in theseconversations. He turns thirty in [MONTH].REDACTSynthetic message. Every identifier is replaced; the sentence structure is untouched.
Before and after. A succession event is still legible on the right; the people are not.
[08] What it will not doFour constraints, stated as constraints

Four things it is built not to do.

01

It does not send your clients' words to a US model.

Inference runs on open-weight models served on Scaleway's EU platform. No OpenAI, Anthropic or Google API sits in the path of client data.

EU serving boundaryOpen-weightNo fine-tuning
02

It does not start on your real clients.

The pilot runs on synthetic data first. Real correspondence enters once the firm has seen what the system does with the invented kind.

Synthetic first90 days3 to 5 seats
03

It does not act on its own.

Every Signal is a recommendation. The relationship manager decides, and the decision is recorded against the client with the person who took it.

SignalRMDecision recorded
04

It does not ask you to take its word.

Every Signal carries the rule that gated it, the messages it was drawn from, and the model and version that read them.

Rule citedSource messagesModel and version
[09] The first monthFig. 7

Four steps, each with something true at the end of it.

A rollout plan with no stated end state per step is a list of activities. These four each finish with a sentence you can check, which is also how the pilot is judged.

Fig. 7Product: Signals, Health Scores and Evidence | LuscentConnect, Week 1 to 4: The system can see the book. First look, Week 4: You know whether it agrees with you. First Signal acted on, Week 5 to 8: There is something to judge. Compounding, Week 9 onward: It stops being a pilot.01 · WEEK 1 TO 4ConnectRead-only, through your ownconsent flowThe system can see thebook.02 · WEEK 4First lookThe book, ranked, onsynthetic dataYou know whether it agreeswith you.03 · WEEK 5 TO 8First Signal acted onA recommendation a personacted onThere is something tojudge.04 · WEEK 9 ONWARDCompoundingThe representation getsdenserIt stops being a pilot.
The first month, and the sentence that has to be true at the end of each step.
StepWhat happens
ConnectMicrosoft 365 and Salesforce, authorised by the firm, scoped to what the firm chooses to share. Nothing is written back and nothing is migrated.
First lookEvery relationship carries a Health Score and a reason. The first conversation is about whether the ranking matches what the desk already believes.
First Signal acted onOne Signal, one decision, one record. False positives are triaged weekly with the compliance function, and the rate is one of the success metrics agreed before day one.
CompoundingA relationship representation accumulates from your own communications. Month six is worth considerably more than month one, and no amount of capital shortens that for anyone.
[10] Where it landsThe screens a relationship manager opens

Five surfaces, one of them daily.

SurfaceWhat it is
DashboardThe prioritised feed. What needs attention today, in order.
ClientsThe book, in three views, filtered however the desk works.
Client DetailSeven tabs: Monitor, Detect, Execute, Connect, Allocate, Compliance, Documents.
SignalsBook-wide inventory, searchable by type, urgency and date.
TasksWhat a Signal turned into, and who owns it.
[11] What it coversProduct regulation only

What the rules core actually produces.

These are obligations on your firm, and this is the record Luscent generates to evidence them. Anything on the roadmap says so.

RegulationReferenceStatusWhat Luscent produces
MiFID IIArticle 25LiveSuitability evidence written at the point of advice, with its source attached
GDPRArticles 28 and 32LiveLawful basis recorded, identifiers removed before inference
EU AI ActArticle 50LiveModel outputs labelled and sourced, a named person decides
CVM and ANBIMARes. 30, 178, 35, 19, 50LiveBrazilian resolutions run in the same rules engine
FinSA and FinIASuitability and documentationLiveSwiss-booked relationships produce the same file
LGPDData handlingAcknowledgedRecorded in the way Brazilian client data is handled
UK Consumer DutyNot shippedRoadmapOn the roadmap and not in the product. We would rather say so

This describes what Luscent does, not what your firm is obliged to do. It is not legal or compliance advice. The platform generates evidence and surfaces Signals; decisions remain with the firm and its professionals at all times.

The pilot is ninety days, free, with three to five relationship managers, starting on synthetic data. Success metrics are agreed before day one, so at the end there is something to judge rather than an impression.

Join the pilot programme

Luscent S.A. is a société anonyme in formation in Luxembourg, at House of Startups, 9 Rue du Laboratoire, L-1911 Luxembourg. Luscent stores and processes client data in the European Union only, on Scaleway infrastructure in Paris with encrypted disaster recovery in Warsaw. Sub-processors, certifications and security controls are documented in the Trust Center.