Sign in
Set up for Trading

AI for insights and analytics

AI analytics for trading

A client that trades 45 percent of its inquiries by count and eight percent by size ranks well on the venue report and is sending every large ticket elsewhere. AI analytics reads the inquiry and trade files, weights by size, and answers the head of sales: who is really trading with us, and who has gone quiet.

Ask it in your own words

These are the questions desks ask. Each one maps to a measure our tools compute from your files. The AI model chooses the measure and explains the result; the arithmetic is done by our code, and every total is checked.

The questionThe measure behind itWhat comes back
Which clients trade with us by count but not by size?Size-weighted hit ratioHit ratio by count and by notional per client and product, with the gap between them ranked.
Who has gone quiet?Quiet client listClients whose inquiries fell against their own baseline, by prior revenue, with the last contact.
What share of each client's inquiries do we see?Inquiry shareOur inquiries against the client's reported total where known, by product.
Are clients trading the products we are axed in?Axe hit rateTrades in axed products over axes shown, by client and desk.
What did we trade that no salesperson had sent?Traded unsentTrades with no prior inquiry or axe from us, by client, with the notional.
How fast do we answer, and does it matter?Response time on inquiriesTime from inquiry to quote by desk, against hit ratio in the same buckets.

A conversation, with the figures

Every figure below was computed by a tool from the rows and checked before it was shown. The lines sum; the percentages match; each line opens to the rows that make it.

Where is the size-weighted hit ratio far below the count?

Notional we quoted and did not win is $4800.0m across 3 desks, on clients whose count hit ratio is above 40 percent and size hit ratio below 15. Rates desk holds $2400.0m, 50% of it.

Rates desk$2400.0m50% of the total
Credit desk$1500.0m31% of the total
FX desk$900.0m19% of the total
The 3 lines sum to$4800.0m0 unexplained

Which rates clients?

3 clients account for $2200.0m of the $2400.0m lost notional on the Rates desk, 92% of it. Every trade over ten million at each went elsewhere.

Aldergate Capital$1100.0m45% by count, 8% by size
Northbridge Asset Management$700.0m48% by count, 11% by size
Sable Macro Fund$400.0m41% by count, 6% by size
These 3 are$2200.0m92% of Rates desk

What does the desk change?

Find out where the large tickets go and why: pricing, response time or a relationship elsewhere. The salesperson takes the size split to the client; the desk head reviews response time on tickets over ten million.

Rates salespeopleDiscuss the large-ticket split with the three clientsThis fortnight
Head of rates tradingReview response time on inquiries over 10mThis week
Head of salesReport both hit ratios weekly, by client tierWeekly
Which clients trade with us by count but not by size?Who has gone quiet?What share of each client's inquiries do we see?Are clients trading the products we are axed in?

The measures behind the answers

Each measure has one formula, one source and one meaning. They are computed per salesperson and product and in total, and every one carries an identity that must hold before it is shown.

MeasureFormulaFromWhat it tells you
Hit ratio by client and productInquiries traded ÷ inquiries received, per client, product and channelInquiry log; blotterWhere the desk wins and loses
Size-weighted hit ratioNotional traded ÷ notional inquiredInquiry log; blotterClients who trade small with the desk and size elsewhere
Inquiry shareClient inquiries seen ÷ estimated client inquiries in the productInquiry log; venue or client-stated dataWhether the desk is shown the flow at all
Quiet client listClients whose inquiries in the last 20 days are below a stated fraction of their own norm, where desk volume is normalInquiry logFlow that has moved to another dealer
Client tiers by flow and profitabilityTier from trailing volume and revenue after cost of capital and hedgingBlotter; revenue attributionWhere balance sheet and attention should go
Axe hit rateAxes traded by the client ÷ axes received, within the windowAxe log; distribution records; blotterWhether the distribution lists work
Traded unsentClients trading the axed instrument, right side, in the window, without having been sent itAxe log; blotterClients missing from the lists
Coverage recency by valueRevenue of clients with a two-way contact within cadence ÷ revenueCRM or chat logs; revenueTop clients nobody has spoken to
Revenue concentrationTop ten clients' share; largest client share, by productRevenue attributionDependence on a few accounts
Response time on inquiriesMedian seconds or minutes to quote, by product and channel; hit ratio by response bandInquiry log with timestampsTrades lost to speed, not price

Each measure is worked through, with the export it comes from and what to drop, in Client KPIs for sales and trading desks.

What the AI model does, and what our tools do

The AI model chooses the measure

From your question and the measures declared for trading, the model picks the one that answers it, and the period and comparison the question implies.

Our tools do the arithmetic

Deterministic code reads the rows, computes the measure, and checks the identities below. The same question on the same data gives the same answer, every time.

The AI model explains, and cites

The AI model writes the sentence around the result, naming the salesperson or account behind it. It states no figure that is not in the result, and every figure links to its rows.

The identities that must hold

Hit ratioInquiries = traded + traded away + passed + no quote
AxesMatched trades are a subset of the blotter; each trade matches at most one axe
TiersEvery client is in one tier; client revenue sums to the desk total
CoverageClients = covered + uncovered

What it reads

The exports desks already produce. Column names are mapped once and the mapping is reused. A file is the way in; scheduled delivery and connections to your systems come with the plan, and every source is listed here.

  • Inquiry log
  • Blotter
  • Venue or client-stated data
  • Revenue attribution
  • Axe log
  • Distribution records
  • CRM or chat logs

Who owns each answer

An answer is a list with an owner and a cadence, or it is a chart nobody works.

MeasuresOwnerCadence
Quiet client list; coverage recencyHead of sales; salespeopleWeekly
Hit ratio, both weightings; response timeHead of sales with head of tradingWeekly to monthly
Axe hit rate; traded unsentHead of salesMonthly
Tiers; inquiry share; concentrationHead of sales and tradingQuarterly

Questions desks ask about AI analytics

Why two hit ratios?

Because count says how often the client trades with you and size says how much. A client at 45 percent by count and 8 percent by size is testing prices on small tickets and trading the large ones elsewhere. Both are computed from the inquiry and trade files, and the difference is the finding.

What is a quiet client?

One whose inquiries fell against its own baseline, not a fixed number, ranked by the revenue it produced before. The list carries the last logged contact, so the head of sales can see whether silence was noticed. It is a list to work through, with an owner per row.

Can it read RFQ and trade data safely?

The files are read where you choose to run it, and client identities can be replaced with identifiers before any upload on Enterprise. No figure is sent to a model to compute; the model only writes the sentence around a result a tool has already produced.

Does it work per desk and across the floor?

Every measure is computed per desk, per salesperson and in total, and the totals reconcile. A client that trades rates and credit is one client in the tier and two rows in the desk view, and the page says which.

See it on your data

Bring a few thousand rows. The data map opens next, every column mapped once, and the first question is answered in minutes. Free, in your browser, no account.