Switching Liquidity Providers Without Breaking Execution Quality

Migrating a wholesale liquidity provider is one of the riskiest operational changes a broker can make. It is also one of the most under-rehearsed. Most teams treat it like an IT cutover — flip the connection, monitor for an afternoon, declare success — and only notice the damage two months later when the P&L review shows fill ratio is down four points and average spread is up six tenths.

This article is the migration playbook we use when we onboard new counterparties at Drovix. It assumes you are switching from any wholesale provider to any other wholesale provider; the principles are vendor-agnostic.

Why migrations fail

Liquidity is a moving target. Your incumbent's quality at the moment you signed is not their quality today, and the new provider's pitch deck is a curated snapshot. The only honest comparison is real flow, matched cohort, over a representative window. That requires planning the migration as an experiment, not a deployment.

The eight failure modes

• 1. Comparing aggregate metrics instead of matched cohorts.
• 2. Migrating during an atypical market regime (post-NFP week, holiday liquidity).
• 3. Not isolating last-look policy differences in the comparison.
• 4. Underestimating the credit-line setup timeline and ending up with a hybrid book during stress.
• 5. Switching FIX session parameters mid-test (e.g., heartbeat interval, MaxMessagesPerSecond).
• 6. Forgetting that drop-copy and audit trail change shape — your reconciliation breaks silently.
• 7. Not running the rollback drill before go-live.
• 8. Communicating the cutover to clients (downstream brokers) too late, or not at all.

Phase 1 — RFP and shortlisting

Limit the shortlist to three. More than three and the decision becomes political. Build the RFP against the twelve questions in our buyer's guide, weight them by what actually matters for your book, and score each candidate blindly. Do not let a relationship manager tell you `we will price that better in production' — anything not in the RFP response is not real.

Phase 2 — Contracting and credit

Contracting takes longer than people expect because credit and KYC run in parallel and either can block. Start collateral wiring early. If you are switching to a pre-funded model from a credit-line model (or vice versa), model the capital impact for at least a full quarter; it is not unusual for the working-capital change to be the biggest single line item in the migration.

Phase 3 — Technical certification

Standard FIX certification covers logon, order entry, cancel/replace, partial fills, and drop-copy. What is usually missed:

• Burst behaviour at 10x normal message rate.
• Behaviour when the primary session is up but stale (heartbeat fine, no quotes).
• Failover to secondary venue mid-trade.
• Symbol mapping for any non-standard tickers (especially exotic FX and index CFDs).
• Time-in-force interactions with rollover windows.

Phase 4 — Parallel running

This is the single most important phase, and the one most often skipped. Connect to the new provider in production, but route a controlled percentage of flow — start at 5% — and replicate that flow against the incumbent's quote stream at the same instant. You are not actually executing against both; you are recording the prices and comparing what would have happened.

After two weeks of parallel data, you have matched-cohort fill quality, matched-cohort effective spread, and last-look reject rates that are directly comparable. Reject any provider that refuses to support a parallel period — that refusal is itself the answer.

Phase 5 — Staged cutover

Migrate one asset class at a time. The standard sequence is:

• Week 1: Move the asset class with the largest execution gap (where the new provider looks meaningfully better).
• Week 2: Move secondary asset classes once stability is observed.
• Week 3: Move benchmarks (EUR/USD, USD/JPY, XAU/USD) — the ones that matter most for your monthly report.
• Week 4: Decommission the incumbent for migrated instruments, keep them warm on the long tail until you are 60 days past cutover.

Phase 6 — Rollback drill

Before going live, run the rollback drill. In UAT, simulate the scenario where the new provider has a 30-minute outage during the London open. You should be able to fail back to the incumbent within five minutes without manual intervention. If that is not true, you do not have a migration plan — you have a leap of faith.

After go-live — the 90-day review

Lock in a formal 90-day execution review. Compare like-for-like quarters. The numbers that matter, in order: fill ratio at notional aggressor cohorts, effective spread by time-of-day bucket, last-look reject rate by venue, average latency at the 95th percentile, and post-trade reconciliation break rate.

If any of those numbers regressed and you cannot explain why, do not normalise it. Either find the cause or roll back. The cost of a second migration is high but it is finite; the cost of slowly bleeding execution quality is unbounded.

Migrating to Drovix specifically

Drovix supports a structured parallel-running mode out of the box. We will give you a shadow FIX session that receives a copy of every quote and computes a what-if fill against your actual ticket stream, so you can build the cohort comparison without any production risk.

If you want to see a sample parallel-run report from a recent migration (anonymised), the institutional sales desk will send one on request.