Transaction cost analysis you work through with agents: every trade's cost explained in plain words, every figure cited, every decision left to you
Every trade's cost explained in plain words, by broker, algorithm and venue, by morning.
See one case, screen by screen ↓Transaction cost analysis (TCA) measures what each trade cost against a fair benchmark, in basis points, and why, so a buy-side desk can show the execution its clients received and compare brokers, algorithms and venues. Transaction Cost Analysis gives that work to eight agents: overnight they cost every equity, fixed income and FX order, split the cost into delay, spread, timing, impact and fees, explain every expensive order in plain words with each figure cited, check each other's work and flag dealers whose behavior changed. In the morning the head of trading reads their findings, asks follow-up questions, opens the evidence and approves; traders propose and the best execution committee ratifies.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Transaction cost analysis is how a buy-side desk shows what execution cost its clients: each order measured against a fair benchmark in basis points, split into delay, spread, timing, impact and fees, and compared across brokers, algorithms and venues. Under the SEC's interpretation of June 5, 2019, every adviser that chooses brokers is expected to evaluate that execution periodically and systematically. Yet TCA still arrives as a vendor portal, a monthly PDF and spreadsheets. Equity TCA is well served; fixed income TCA depends on which fair price the tool picked when a bond's evaluated price is days old; FX last-look rejections sit in platform logs nobody reads.
And the numbers never say why. An order cost 38 bp, and the reason is rebuilt from EMS notes, chat and memory, so the committee cannot tell poor execution from a halt or a guidance cut. A market-on-close order is measured against arrival and a fix order against a mid it was never meant to beat. A dealer that lengthens its last-look window shows up only in a quarter-end average, brokers are ranked on a few dozen orders, and a broker change is decided in a meeting and an email with no maker-checker.
Estimated split for one month of post-trade TCA at a multi-asset desk, by hand and with the solution.
Eight agents benchmark, cost, explain and check every order; traders propose, the head of trading approves.
Quillfield Capital, a global asset manager, traded 652 orders on Friday, October 16. By Monday at 9:10 a.m. in New York the agents have costed every one, explained the expensive ones and lined up what needs a person. Here is how the Global Head of Trading works through it with them — reading their findings, asking back, opening the evidence and approving — screen by screen, in the working solution.
Sam's morning screen opens with four numbers: 652 orders and 8,114 fills analyzed, $6.48B, every one costed; 4.9 bp against benchmark with a peer median of 5.5; 23 beyond threshold, 19 ready to accept and 4 needing him; 2.03 million fills in the last 12 months. Beside “Needs you”, with one next step each, the agents say what they did between 1:00 and 1:22 a.m., each line cited. A late bond quote log was retried at 6:52 and 14 bond orders re-analyzed, so the counts tie.
Anomaly agent · 1:15 a.m.: “Thornbury Bank: 16.8% of requests rejected since Oct 1, after it lengthened its hold window to 200 ms.”
Lukas Meyer bought €50,000,000 against dollars for the International Equity Fund, request for stream to five banks. On the price chart: arrival mid 1.17416, three requests to Thornbury Bank rejected after 172, 176 and 178 ms holds, then Thornbury filling €25M at 1.17427, 0.6 pip worse than its first offer, and Ardmore Bank and Larchfield FX filling the rest higher. The cost stack beside it splits 2.1 bp five ways: delay 0.1, spread 0.4, timing 0.6, impact 0.7, fees 0.3, against a peer median of 0.7 bp.
Benchmark agent: “The PM asked for a risk transfer, ‘complete now’; this is not a fix order, so the policy’s arrival benchmark is the fair one.”
Down the right, each agent says what it did, with its plan, sources and confidence. The Explanation agent writes why in plain words: Thornbury showed the best offer, 1.17421, then rejected Lukas three times while the price rose, and filled 0.6 pip worse; without the rejections the order would have cost about 0.8 bp. The Checker agent recomputes from the raw fills, gets 2.09 bp, confirms the 172, 176 and 178 ms holds and agrees with the explanation.
Neither agent may accept the explanation. “Accept the explanation?” waits for a person.
Sam types the question under the thread. The Explanation agent answers from the quote log: at 14:20:05.840 Thornbury offered 1.17421, Ardmore 1.17422, Larchfield FX 1.17424, Sundgren 1.17426 and Kinloch 1.17427. Thornbury was best by 0.1 pip, which is why the platform kept sending to it; sending to Ardmore after the first rejection would have saved an estimated 0.4 bp.
Every answer carries its citation, and the question and answer go on the audit trail.
Each figure on the replay is numbered. Clicking the rejection opens the FX platform's trade request and rejection log for Thornbury Bank: request time, price, amount, decision, hold in milliseconds and the mid at request and at decision. Three requests rejected after 172, 176 and 178 ms, one accepted after 41 ms, and the reject reason field reads “PRICE_CHECK” for all three.
“The record is read-only, fingerprinted and kept with the order's audit trail.”
Since October 1 Thornbury rejected 36 of 214 requests, 16.8%, against a panel median of 0.8% and its own 0.9% in Q3. The median hold rose from 38 to 176 ms, and 33 of the 36 rejections came when the price moved against the bank during the hold. The agent read the bank's disclosure dated September 28, which raised its maximum last-look window from 50 to 200 ms, and estimated the cost at $244,000: 0.9 bp of re-fill cost on $2.71B rejected.
Anomaly agent: “I drafted an inquiry letter asking Thornbury for the reasons behind the 36 rejections and its hold-time policy, citing the FX Global Code. It needs the head of trading's approval before it is sent.”
Lukas proposed A-1041 at 08:46 London: move Thornbury Bank to Review, out of the EUR/USD and GBP/USD request-for-stream panels above €20M, streaming below €20M continues, an inquiry letter on rejection reasons and hold-time policy, re-assess November 16. Sam opens it, keeps reason code EXQ-LL, execution quality: last-look rejections, and ticks the attestation before he signs.
“I have reviewed the evidence attached to this proposal and I am not its maker.” Signed, fingerprinted and added to the audit trail.
The review queue holds 19 costs explained and checked by two agents: Daikoku Precision's special buy quote at the open, read from a Tokyo notice in Japanese; Sorvino Industria's volatility auction, from a Milan notice in Italian; Morrow Health Systems' trading pause; a USD/JPY fix order measured against the fix at 0.2 bp; a Hollins Water bond at 2.9 times average daily volume; Ultra 10-year Treasury futures in a thin book. Sam selects all 19 and accepts with one reason code.
“Accepting signs them as documented reasons.”
Hannah Brooks sold $25,000,000 of Ashgrove Utilities 5.10% 2036 to Ostrander Markets at 99.42, cover 99.38, after an RFQ to six dealers. The evaluated price, 99.71, is two business days old, so the Benchmark agent built a fair price of 99.55, confidence 88%: 13.0 bp, where the stale price would read 29.1 bp. The Checker agent uses the 10s30s curve move rather than the 10-year move for a 2036 maturity, gets 99.53 and 11.1 bp — 1.9 bp apart, above its 1.0 bp tolerance.
“Which fair price should stand?” Use 99.55 · Use 99.53 · Ask Hannah.
The committee's screen for the meeting on Thursday, October 22 at 10:00 a.m. ET shows October so far: equities 10.8 bp, bonds 4.3 bp, FX 1.9 bp against a peer median of 1.8, and 3 decisions to ratify, each maker-checked with its evidence attached, among them A-1041, Thornbury Bank to Review, approved by Sam that morning. Harriet ratifies what the desk decided; Calloway Markets' move to the watch list goes to a committee vote. The Report agent drafted the brief overnight: FX above peers for the first time this year, because Thornbury Bank began rejecting 16.8% of requests.
Committee brief: “The desk proposed Review the same day the pattern was found.”
The same Friday, seen by five of the people who trade it, approve it and oversee it — what TCA looked like before, and what it looks like now.
Eight agents benchmark, cost, explain and check every order; traders propose, the head of trading approves.
Gathers orders and PM instructions from the order management system, fills and routes from the FIX drop copy, FX quote and rejection logs, bond RFQ quotes and market data; ties the counts and re-pulls late fills and late logs.
Picks a fair benchmark for every order from the execution policy and the order's instruction, and writes one sentence on why; builds a fair bond price from dealer quotes, prints and the curve move when the evaluated price is stale.
Computes each order's cost in basis points and splits it into delay, spread, timing, impact and fees, against the pre-trade estimate and the peer median.
Writes why each expensive order cost what it did, in plain words, with every figure cited; reads trader notes, news, halts and exchange notices in the original language, such as Japanese or Italian.
Recomputes every cost and its split from the raw fills, independently and on a different model family, and tests each explanation against the market record.
Watches each dealer's reject rate, hold time and asymmetry, algorithm participation against the instruction and venue markouts; reads dealers' last-look disclosures and drafts inquiry letters.
Ranks brokers, algorithms and venues per asset class with sample sizes, cost against expected and against peers, and a confidence label.
Builds the daily cost summary, the monthly report, the committee brief, client TCA reports, the quarterly execution-quality summary for compliance on the EU and UK entities (MiFID II Article 27, FCA COBS 11.2A) and the examiner pack, every figure tied to the cost record.
Propose, accept, approve, ratify. The agents propose; a named person decides.
Every order and every working list has a box to ask the agents or tell them what to do. Answers point to their source, and nothing goes out without a person.
Why did EUR/USD cost 2.1 bp on Friday?
The €50M EUR/USD buy cost 2.1 bp against a peer median of 0.7 bp. Thornbury Bank rejected three requests, each after a hold of under 180 milliseconds, then filled 0.6 pip worse than its first offer; the second half paid about 2 pips more as the price moved. A second agent recomputed it: 2.09 bp. Opening the replay.
What did the other banks quote?
At 14:20:05.840 the five banks offered: Thornbury 1.17421, Ardmore 1.17422, Larchfield FX 1.17424, Sundgren 1.17426, Kinloch 1.17427. Thornbury was best by 0.1 pip, which is why the platform kept sending to it. Sending to Ardmore after the first rejection would have saved an estimated 0.4 bp.
What would it have cost without the rejections?
Filling all €50M at the first offers (Thornbury 1.17421 for €25M, Ardmore 1.17422 for the rest) gives about 0.8 bp, $4,700 including fees, so the rejections and what followed cost about $7,630.
Draft a note to the PM
Draft for Ruth Okafor: “Your €50M EUR/USD buy cost 2.1 bp against 0.7 bp for peers. One bank rejected three requests in just over a second, each after a hold of about 175 ms; we have proposed moving it to review. Fund impact: $12,330, of which about $7,600 is attributable to the rejections.” I have not sent it: the draft is open for you to edit and send.
13 screens from the working solution, on its sample data. Pick one to see it large.
Benchmarks, thresholds and alert rules are settings that follow your execution policy. A change is a proposal: the agents back-test it on last quarter, then compliance and the head of trading approve it.
| Setting | Default | Choose from |
|---|---|---|
| Benchmarks per asset class | Equities: arrival (VWAP shown for VWAP orders, close for market-on-close) · Bonds: evaluated mid at RFQ time, RFQ composite when stale (pending A-1038) · FX: mid at arrival, fix orders against the fix · Futures: arrival in ticks | Changed by proposal; the TCA analyst proposes, compliance and the head of trading approve |
| Equities threshold | Twice the pre-trade estimate, never below 20 bp | Against arrival; edit and propose |
| Bond thresholds | Investment grade 10 bp · high yield 35 bp · GBP and EUR investment grade 5 bp | Against the evaluated mid or fair price; edit and propose |
| FX and futures thresholds | FX 1.5 bp against mid, or 1.0 bp above the peer median · futures half a tick above expected | Edit and propose; in every asset class, orders under $250,000 are costed, not explained |
| Dealer reject rate alert | Above 5% (was 8%) or 3× the panel median, on at least 50 requests | On · Off |
| Last-look alerts | Median hold above 100 ms in a week · more than 75% of rejections when the price moved against the bank · a new disclosure version | On · Off, each |
| Algorithm above instruction | Participation above the cap by 5 points for 10 minutes | On · Off |
| Service levels | Dealer alerts same day · trader requests 24 hours · algorithm checks 2 business days · cost and benchmark reviews 5 business days · broker follow-ups 10 | Escalate to the head of trading after the clock, to compliance after twice the clock; a change is a proposal the head of trading approves |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample trading data · “target” = the design goal, measured in the live solution · “estimated” = our estimate (the split of days); the $244K and $41.2M figures are the working solution's own estimates on its sample data (0.9 bp × $2.71B of rejected requests; 1 bp × $412.2B of quarterly flow) · “typical” = published figures (Coalition Greenwich, May 27, 2026 — almost 60% of US buy-side fixed-income desks use TCA; traders planning to implement a TCA solution rose from 11% to 23% in a year). Context: SEC Commission Interpretation Regarding Standard of Conduct for Investment Advisers, June 5, 2019; FX Global Code; MiFID II Article 27 and FCA COBS 11.2A for the EU and UK entities' execution-quality summary. People, firms, banks, brokers, funds and securities named on this page are fictional — characters and sample data in the working solution.
Transaction cost analysis (TCA) measures what each trade cost against a fair benchmark, in basis points, and why, so a buy-side desk can show the execution its clients received and compare brokers, algorithms and venues. Transaction Cost Analysis gives that work to eight agents: overnight they cost every equity, fixed income and FX order, split the cost into delay, spread, timing, impact and fees, explain every expensive order in plain words with each figure cited, check each other's work and flag dealers whose behavior changed. In the morning the head of trading reads their findings, asks follow-up questions, opens the evidence and approves; traders propose and the best execution committee ratifies.
For default and urgent equity orders the arrival price benchmark applies, which is implementation shortfall: what the fund paid from the moment it decided. The Cost agent splits that cost into delay, spread, timing, impact and fees and compares it with the pre-trade estimate and the peer median. For VWAP-instructed orders the interval VWAP benchmark is shown as well, because VWAP alone flatters an order whose own buying lifts the average.
When the evaluated price is old, the Benchmark agent builds a fair price from the dealer quotes received, the last prints and the curve move, and states its inputs and confidence. On the sample Ashgrove Utilities sale, the two-day-old price would have read 29.1 bp; the fair price gives 13.0 bp. The Checker agent computes its own figure, and when the two differ by more than 1.0 bp a person decides.
With last look, FX banks may hold a trade request for a short window, measured in milliseconds, and then reject it. The Anomaly agent watches every bank's reject rate, hold time and asymmetry each morning from the FX platforms' millisecond logs, and reads each bank's last-look disclosure. In the sample, it tied Thornbury Bank's new 200 ms maximum window to 16.8% of requests rejected since October 1, 33 of 36 when the price moved against the bank, and drafted an inquiry letter citing the FX Global Code last-look principle (Principle 17) for the head of trading to approve.
The one its instruction calls for, from your execution policy. The Benchmark agent reads each order's instruction and writes one sentence on why: arrival for urgent equity orders, the close for market-on-close, the fixing rate for FX fix orders, the evaluated mid or a fair price for bonds, arrival in ticks for futures. A USD/JPY fix order flagged at 3.4 bp against arrival cost 0.2 bp against the fix.
The Ranking agent ranks brokers, algorithms and venues per asset class with the number of orders, cost against expected and against peers, and a confidence label, and marks a rank “too few to tell” where the sample is thin. Agents never change routing, panels or algo-wheel weights. A change to algo wheel broker selection is a person's proposal: A-1040, cutting Calloway Markets' wheel weight from 8% to 4% after a third quarter above expected cost, goes to the committee to decide.
Yes. Agents explain, check, answer questions and draft; people decide. A person accepts every explanation, broker actions and client reports are maker-checker with a reason code, methodology changes need compliance and the head of trading, and nobody approves their own proposal. Every step, including each question asked of the agents, is on an append-only audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your execution policy, a sample of past orders and fills, and your dealer disclosures — and it goes live once every quality gate has passed. First-time setup connects your systems read-only, maps counterparties, reads the benchmarks from your policy and back-tests a past quarter before go-live.
Committee pack in two hours, every outlier explained, every figure tied to its source.
1-min demoBroker VoteEvery research dollar tied to its vote and evidence, approved in days, not weeks.
1-min demoCommission ManagementEvery research dollar matched to an eligible use and a signed split, before money moves.
1-min demoWe'll run Transaction Cost Analysis on a past month of your own orders, fills, FX rejection logs and bond quotes.