BRK // For Brokers · Dealing Desk AI

Your dealing desk, running in seconds instead of the next morning.

The modern broker dealing desk is an AI problem, not a headcount problem. We ship real-time toxic-flow scoring, per-client and per-symbol A/B routing recommendations, LP scorecards and exposure alerting — plugged into your existing bridge without touching the terminal.

BuyerHead of Dealing · Head of Risk← All broker capabilities
The problem

The dealing desk problem, precisely.

A retail broker running 20,000 active traders across MT5, three LPs and a mixed A/B book has a state-space no human dealer can hold. By the time last night's report shows that a Vietnam cluster ran a 34% win rate on Gold or that an EA on EURUSD is 92% correlated across 41 accounts, the exposure has been running for hours.

This is not a problem you solve by hiring another dealer. It's a problem you solve by giving the dealers you have a system that watches every ticket in real time, scores it against models trained on your own flow, and surfaces the two things that actually matter this minute.

What good looks like

What a real dealing desk AI looks like.

Not a dashboard. A system that sits in the flow and produces four artefacts:

  • A per-ticket toxic-flow score, produced in <500ms, that combines behavioural signals (latency to fill, session concentration, symbol clustering, EA fingerprints, IP/device graph) with historical PnL curves.
  • A per-client and per-symbol routing recommendation — A-book, B-book, or partial — that a dealer can approve, override or automate.
  • A live LP scorecard: fill quality, rejection rate, mark-up leakage, hedging cost per LP per symbol per session.
  • Exposure alerts that page the right dealer when a correlated position crosses a jurisdictional or capital threshold — not when the daily report reveals it tomorrow.
What KJ Capital ships

What KJ Capital ships.

A bespoke Dealing Desk AI, wired into your PrimeXM, oneZero or Centroid bridge (or your custom OMS), your MT4/MT5 Manager API and your risk warehouse. We deliver:

  • A production toxic-flow model trained on your own historical flow, evaluated against a hold-out cohort with published precision/recall.
  • A routing recommendation service the desk consumes via API and a lightweight UI overlay.
  • An LP scorecard dashboard refreshed per-session, exportable to your BI stack.
  • An alerting layer routed to Slack, Teams, email or SMS with clear ownership.
  • The full source, deployed in your cloud tenancy. No black box, no lock-in.
Typical outcomes

What brokers actually see after go-live.

Toxic-flow detection latency
sub-second
Typical B-book protected PnL uplift
6–14%
Dealer alert-fatigue reduction
60–80%
Time from kickoff to production
8–10 weeks
Frequently asked

What buyers ask us about this build.

Q01Does this replace our dealers?+
No. It replaces the moment where a dealer stares at a screen trying to spot patterns humans cannot see at that speed. Your dealers become approvers and exception-handlers, which is where they add most value.
Q02Do you touch the MT4/MT5 terminal itself?+
No. Everything runs alongside — Manager API, bridge feeds, risk warehouse. You stay MetaQuotes-safe and license-safe.
Q03How is the model trained?+
On your own historical flow: tickets, fills, PnL curves, KYC attributes, session data. We do not train on other brokers' flow, and your data does not leave your tenancy.
Q04What if we don't have a bridge?+
We build the ingestion layer for whatever you have — Manager API, custom OMS, direct LP feeds. It adds a week or two, not a project.
Q05What's the price?+
Productised Build: £120k, 8 weeks, fixed scope. Continues on the AI Operator retainer (from £8k/mo) once live.
Next step

Two weeks. £15k. A written blueprint and a working proof-of-concept on your data.