How a broker chooses liquidity: the conversation almost nobody gets right

Most conversations between a broker and an execution provider start badly, and always for the same reason: spread gets discussed before depth, and latency before anyone asks where it's measured from. It ends with an agreement that looks good in the deck and breaks on the first day of non-farm payrolls.

This guide is what a desk head should take into that first call. There are no brand recommendations here: just the criteria, the exact questions, and — more useful still — the answers that should make you hang up the phone.

First: what you're actually buying

There's a confusion worth resolving before anything else, because it determines who you're talking to and what they can legally offer you.

When a broker "contracts liquidity," in practice it's doing one of two very different things:

It onboards as a client of another entity and executes through it. Your entity opens an institutional account, sends its flow, and the counterparty acts as intermediary in the execution of those transactions. You keep the relationship with your clients and the commercial terms.

It receives a wholesale feed to trade against its own book. Here the counterparty supplies you prices and you manage execution internally.

They look the same from the commercial side, and they are not the same from the regulatory side. In many jurisdictions the second activity requires different — and more expensive — permissions than the first. Before going any further, ask which of the two they're offering you and under what license. If the answer is vague, you've already learned something important.

The six criteria, in order of importance

1. Depth, not headline spread

The spread that appears on the website is the best case, at the best moment, in the most liquid pair. It's useless to you.

What you need is the distribution of spread across the sessions and instruments your desk actually trades. A venue advertising "from 0.0 pips in EUR/USD" should be able to show you how that spread behaved during the Asian session, in the London–New York overlap, and at the US close. If it can't produce that data, the headline is advertising.

And depth matters as much as the touch. Two venues can both quote 0.0 pips at the top of book and give you radically different fills on 5 million. A 0.0 spread on one million is worthless if your average ticket is ten and walking the book costs you 0.3 pips.

Ask for this: the spread distribution and the depth in EUR/USD and gold over the last 30 days, broken out by session.

2. Latency measured from where you are

Latency is a relative number. It's measured from where your stack sits to where the matching engine sits, and any other figure is internal marketing.

A provider with a New York presence and a 250-microsecond cross-connect to NY4 means nothing to a prop firm in São Paulo reaching it over 120 milliseconds of public internet.

A serious provider helps you map the round-trip from your specific point of presence, and sometimes the honest answer is "as you're set up you won't improve; you need to move your infrastructure." That answer is worth more than a generic "sub-millisecond."

Ask for this: given my current point of presence, what's the realistic round-trip to your LD4 and NY4 hubs, and what would I have to change to lower it?

3. Connectivity: FIX, MT5 and REST

In 2026 the institutional baseline is native FIX 4.4 and 5.0, with a production-parity sandbox available before any commercial commitment. Anything less is a venue that isn't targeting institutional flow, whatever its website says.

For brokers on MetaTrader — still the majority in LATAM, the Middle East and part of Asia — a production-tested MT5 bridge is non-negotiable. Symbol mapping, A/B/hybrid routing and white label deployment have to work out of the gate, not "in the next release."

The most revealing question isn't which protocols they support, but what happens when the session drops at three in the morning. If the reconnection and sequence-reset behavior isn't documented, you'll discover it in production.

Ask for this: the FIX certification timeline, the MT5 bridge architecture, and the reconnection documentation.

4. Transparency of the risk model

There are three models: A-book (straight pass-through to the market), B-book (internalization) and hybrid (routing by criteria). All three are legitimate at the institutional level. What separates a serious provider from a cosmetic one is whether you can see and configure which one applies to your flow.

A serious provider publishes its rejection rate for each upstream counterparty when it uses Last-Look, and gives you a dashboard where you see how your flow is routed. One that isn't serious dances around the question.

Another signal: if the risk function is operationally separated from the trading desk, with daily reconciliation, there's less structural chance of P&L leaking against you.

Ask for this: what's your rejection rate by upstream, and how is risk routing configured for my flow?

5. Real multi-asset coverage

A broker that started in Forex in 2024 probably runs a multi-asset book today: indices, metals, energies and a basket of crypto CFDs. A provider that only does Forex well, or does Forex well and metals badly, is a bottleneck waiting to happen.

The reasonable standard today is 800+ instruments in a single normalized stream, so your stack doesn't have to reconcile feeds from several vendors.

Ask for this: show me the coverage and depth in the asset classes that aren't Forex.

6. Senior contact, during and after

The criterion that gets promoted least is the one that ends up mattering most. An institutional integration that goes live in 48–72 hours needs a senior person running it, not a chain of handoffs between sales, onboarding and integration.

The same person who takes the first call should be the one who answers at 3 a.m. when something breaks. If the provider can't commit to that — or worse, laughs at the expectation — it isn't built for institutional flow.

Ask for this: who is my assigned senior contact, and is it the same person across onboarding, integration and incidents?

Signs you're wasting your time

Some answers should end the conversation:

• "The spread is from 0.0 pips" without being able to show the distribution. That's a headline figure, not an execution condition.
• "Sub-millisecond latency" without asking where your infrastructure sits. They're quoting you their number, not yours.
• "We use Last-Look but it's industry standard" without publishing the rejection rate. Last-Look isn't the problem; opacity is.
• Not being able to tell you under what license they operate and which activities it covers. If the counterparty can't precisely describe what it's permitted to do, the problem is yours too.
• A sales contact who disappears after signing. Ask explicitly who stays afterward.

The most expensive mistake: signing before testing

Most integration problems don't show up in the demo. They show up in the first volatility event, with the book loaded and clients watching.

That's why a production-parity sandbox isn't a technical detail, it's the guarantee. A sandbox that behaves differently from production pushes bug discovery into your first live session. Demand sandbox credentials before any commercial commitment, and test specifically for:

• Behavior in high-impact news windows
• What happens when the FIX session drops and reconnects
• Fills at the ticket size you actually trade, not a demo one
• The full symbol mapping of your catalogue, not a sample

How to use all this

Take the six criteria into your next first call — including ours. The provider that can answer all six in production terms, with data and not adjectives, is the one built for institutional flow in 2026.

At Exura Prime we work with these criteria because they're the ones we apply to ourselves: brokers, prop firms and funds onboard as institutional clients and execute through us. If you want to test us against this list, talk directly with our institutional team.

To go deeper into the mechanics behind these criteria: how liquidity aggregation actually works.

Frequently asked questions

What does it mean for a provider to be "connected to Tier-1"? The Tier-1 label refers to the global banks — JP Morgan, Goldman Sachs, Citi, Deutsche Bank, UBS, Barclays, BNP Paribas, Bank of America, Morgan Stanley and the like — that run wholesale FX desks. A venue is "connected to Tier-1" when its book aggregates feeds from those banks, not when it operates on prices from a retail aggregator. It's a verifiable difference: ask for the list.

How long should an institutional onboarding take? Between 48 and 72 hours of effective work from a complete application: application, KYB and beneficial-owner verification, sandbox integration testing and cutover to production. Slower than that points to a process problem; much faster usually means compliance controls are being skipped, which is not good news.

Do I need FIX or is the MT5 bridge enough? It depends on your client base. If you run a MetaTrader brokerage, the bridge is enough. If your internal stack speaks FIX natively — typical in prop firms, funds and managers — the FIX session is the right path. Many brokers start with the bridge and add FIX as they scale.

Does depth matter more than spread? For institutional tickets, yes. Always evaluate spread together with depth at the size your strategy actually trades. An excellent spread you can't access with your volume isn't an advantage, it's a figure.