The cost of B-book exposure monitoring depends primarily on reliable trade-event integration, position reconciliation and operating coverage, rather than the dashboard or AI model. Compare suppliers using a three-year total cost of ownership that separates licences, implementation, integration, evaluation and ongoing operation against the same risk and recovery requirements.
Price the monitoring obligation first
A dealer monitoring one trading platform during a staffed session has a different requirement from a group monitoring several platforms, currencies and legal entities around the clock. Define the instruments, account populations, valuation measures, acceptable event lag and response hours before requesting prices. State whether monitoring is advisory or supports an operational control on which the desk intends to rely. Those decisions determine the engineering and support burden.
The initial scope should name the authoritative source for positions, contract specifications, market marks and currency conversion. Specify how often the projection must reconcile with platform snapshots, who investigates a break and what the interface displays while data is stale. A low headline price can become expensive if it assumes clean identifiers, uninterrupted event delivery and no history to replay. Ask bidders to list these assumptions explicitly.
An illustrative three-year TCO
This is a hypothetical budgeting exercise, not a KJ Capital quote, vendor price survey or promised delivery schedule. Amounts are GBP excluding VAT, financing and inflation. Assume one platform, a read-only monitoring service, existing data-access rights, one supported deployment and staffed business-hours operation. The assumed delivery rate is £800 per person-day; actual effort and rates require discovery.
| Cost category | Transparent assumption | Year 1 | Years 2–3 | Three-year total |
|---|---|---|---|---|
| Software licence | Hypothetical £500/month monitoring component | £6,000 | £12,000 | £18,000 |
| Implementation | 35 person-days × £800: ledger, limits, interface and runbooks | £28,000 | £0 | £28,000 |
| Integration | 20 person-days × £800: events, marks and reconciliation | £16,000 | £0 | £16,000 |
| Evaluation and acceptance | 12 person-days × £800: replay, failure and dealer tests | £9,600 | £0 | £9,600 |
| Hosting and observability | Assumed £400/month | £4,800 | £9,600 | £14,400 |
| Maintenance and support | 2 person-days/month × £800, all three years | £19,200 | £38,400 | £57,600 |
| Total before contingency | Sum of the six categories | £83,600 | £60,000 | £143,600 |
The example excludes platform API fees, market-data entitlements, 24-hour cover, disaster-recovery facilities and the broker's internal staff time. Budget these separately where required; do not treat a zero entry as evidence that a cost cannot arise.
Why integration can dominate the build
An adapter is more than an API connection. It must handle symbol aliases, account currencies, partial closes, amended trades, duplicate delivery and recovery after an interruption. If historical events are unavailable, the team needs an agreed bootstrap from authoritative snapshots. If the platform interface is restricted by contract, obtaining permission may affect the schedule even when coding effort is modest. Validate access before committing to a fixed delivery date.
Ask for an integration price per named source with its expected data contract, recovery behaviour and acceptance tests. A second platform may reuse the ledger and screen while still requiring substantial mapping and reconciliation work. Quote changes to instrument coverage separately: gross currency exposure and complex derivative sensitivities do not represent equivalent calculation requirements. The responsible risk specialist must approve the chosen measures and their limitations.
What the licence comparison must include
- Charging unit: platform, account, legal entity, concurrent dealer, event volume or deployment. Model growth against that unit rather than assuming a flat bill.
- Rights to export raw events, position history, alert evidence and configuration in a usable format. Include exit assistance and data-retention responsibilities.
- Included feeds and entitlements. Confirm whether market marks, bridge access and platform interfaces are licensed separately and whether non-production testing is allowed.
- Support scope: incident response hours, restoration responsibilities, upgrade compatibility, backup testing and the boundary between the vendor and internal platform team.
Buy, build or combine the two
Buying is attractive when an existing product supports the broker's platforms, exposure measures and working practices without extensive adaptation. A configured product still needs independent reconciliation and acceptance. Request a trial using controlled representative records, including a disconnect and replay. Assess the dealer's ability to explain each alert and reach the underlying position data, rather than judging a polished demonstration alone.
A custom build becomes more defensible where cross-platform reconciliation, entity boundaries or desk-specific risk workflows are central to the requirement. It brings responsibility for releases, incidents, data contracts and specialist knowledge. A combined approach can use licensed calculation or feed components beneath a broker-owned workflow. Put ownership, export rights and replacement of individual components into the contract so that the architecture remains maintainable.
Evaluation is a budget line, not a launch ceremony
Reserve effort for independent expected results and repeatable replay tests. Include delayed and reordered events, trade corrections, missing prices, currency conversion and limit changes. Compare the system with authoritative snapshots over representative operating periods. Set tolerances with the desk and record unresolved discrepancies. The output should make a degraded source obvious; a plausible number generated from incomplete data is not a successful acceptance result.
Keep deterministic calculations outside the language model. If AI drafts an explanation, evaluate whether it cites the correct calculation, acknowledges stale evidence and stays inside the user's permissions. Monitoring must continue when the model is unavailable. Cost the ongoing review of new instruments, platform upgrades and configuration changes, because acceptance evidence can become outdated as the operating environment changes.
Use the business case without inventing a return
Measure the existing time spent collecting evidence, investigating reconciliation breaks and preparing shift handovers. A hypothetical saving of 30 minutes per day across four operators is two staff-hours per day; multiply by the broker's own loaded hourly cost and actual staffed days. That is a capacity estimate, not guaranteed cash savings. Do not present avoided trading losses as a predictable return without a defensible counterfactual.
Approve a contingency according to the uncertainty in data access and calculation scope, and show it separately from the base TCO. Before procurement, require a named operating owner, rollback procedure and funding for support. The useful decision is whether the broker can sustain accurate monitoring and respond to exceptions over the full period, with clear responsibility when an integration or feed fails.
Related questions
Is an AI model the largest running cost?
Not necessarily. Trade-data handling, retention, observability, specialist support and dealer review may cost more. Price the intended model workload separately and keep core exposure calculations deterministic.
Can we start with a spreadsheet export?
A controlled export may support discovery or periodic reconciliation, but it does not meet a requirement for current event-driven exposure. Agree the operating limitation before using it as a production control.
Does custom software remove licence costs?
It may remove a monitoring-product licence, while platform access, market data, database, hosting and other dependencies can still carry fees. Include maintenance and exit costs in the comparison.
Book the KJ Capital AI Diagnostic to define the monitoring scope, integration risks and a broker-specific costed build plan. Use the readiness assessment as a supporting step before committing to implementation.