How to A/B Test Your Lead Routing

Most routing changes get evaluated by gut feel. Someone moves leads around, close rates tick up that month, and the team calls it a win. Then six months later nobody can remember what actually changed or whether it held.

If you want to know whether a routing change works, you need an experiment. Not a before/after comparison. An actual controlled test with a holdout group, defined metrics, and a run time that outlasts your natural sales cycle.

This guide walks through how to do that cleanly, what to watch out for, and where most teams go wrong.

Why Before/After Comparisons Almost Always Lie

The obvious way to test a routing change is to flip the switch, wait a month, and compare. The problem is that almost everything else changes at the same time: your reps' tenure, your product releases, your market, seasonal demand, which accounts are in the pipeline.

A before/after comparison can't separate the effect of your routing change from any of those other variables. You might see a genuine improvement. You might be seeing Q4 seasonality. You can't tell.

A controlled experiment solves this by running both versions at the same time on equivalent leads. Whatever external forces affect one group affect the other. The only thing that differs is the routing logic.

Set Up a Clean Holdout Group

The foundation of a good routing experiment is a holdout group: a slice of incoming leads that continues to route under your current logic while the rest go through the new logic.

A few things matter here:

• Randomize at the lead level, not the rep level. If you assign certain reps to the new logic and others to the old, you're also testing rep differences, not just routing differences. Assign each incoming lead randomly to control or treatment.
• Keep the split proportional. A 50/50 split gives you the fastest answer. If you can't accept 50% of leads going to the old logic, use 70/30 or 80/20, but know that a smaller holdout means you need more time to reach statistical significance.
• Don't let reps know which group is which. If reps know some leads are "test leads," their behavior changes. Keep the experiment invisible at the rep layer.
• Lock the assignment. Once a lead is assigned to control or treatment, it stays there. Don't reassign leads that didn't convert quickly.

What to Measure

Routing experiments fail in two ways: teams measure too little, or they measure the wrong thing.

Primary metric: close rate or qualified pipeline rate

Close rate is the clearest signal. Did more leads in the treatment group result in closed-won deals? This is what routing is actually trying to move.

If your sales cycle is long, you can use an intermediate metric like opportunity creation rate or SQL rate as a leading indicator, but track it carefully. It's easy to see more opportunities created and assume routing is working, when really the new logic is just sending leads to reps who are more aggressive about creating opps regardless of fit.

Secondary metrics to track alongside:

• No-show rate (a routing change can shift this significantly)
• Time to first meeting
• Average deal size in each group (to check if you're routing better-fit leads to treatment)
• Rep utilization and meeting volume per rep

The metric to actively avoid: revenue credited in-period

If you're running a 60-day experiment and your average sales cycle is 90 days, in-period revenue tells you almost nothing. Most of the treatment group's deals won't have closed yet. Measure pipeline stages and use close rate on deals where the cycle has had time to complete.

How Long to Run the Experiment

Short answer: at least one full sales cycle, often two.

If your average deal closes in 45 days, run the experiment for at least 45 days after the last lead enters the test, and preferably longer. Leads that enter in week one of your experiment might close in week six. If you call the experiment in week four, you're evaluating incomplete data.

Beyond the sales cycle, you need enough volume to be meaningful. A 2-point difference in close rate between groups of 50 leads each could easily be noise. The same difference across 500 leads per group is much harder to dismiss.

Use a standard sample size calculator if you want a precise target. As a rule of thumb: if you're looking for a 5-point improvement in close rate and your baseline is around 40%, you need several hundred leads per group before you can trust the result.

The Traps That Make Results Look Better Than They Are

Cherry-picking the end date. If you check results every week and stop the experiment the moment the treatment group looks good, you've introduced bias. Pick your end date before you start, or commit to a statistical significance threshold and don't peek early.

Contamination between groups. This happens when leads assigned to control accidentally get routed through the new logic (or vice versa). Audit your assignment logic before you launch, not after.

Rep quality imbalances. If treatment leads happen to land with your top three reps more often due to a flaw in your randomization, your results will look inflated. Check the rep distribution across both groups when the experiment ends.

Novelty effects. Reps sometimes perform better on leads they know are part of something new. If your reps know an experiment is running, you may see a temporary bump that fades.

Attribution timing mismatches. If you change your routing logic mid-experiment, you've invalidated the test. Freeze the logic for the duration.

What Good Results Actually Look Like

In one B2B sales case study across 2,420 meetings and 1,281 deals, the measured close-rate gap between the best-matched rep and the worst-matched rep for a given deal type was nearly 30 percentage points. That is the ceiling of what better routing can theoretically capture. Modeled uplift from routing alone came to roughly 17% in that study. The gains weren't guaranteed to any specific account, but the signal was clear: who handles the lead matters as much as how many leads you generate.

You can read more about how that analysis was structured in the Salescadia case study.

A clean A/B test of your own routing can tell you whether you're actually moving toward that potential, or just making changes that feel productive.

FAQ

How small can a holdout group be?

Technically you can run a 90/10 split, but the smaller your holdout, the longer you need to run the experiment to detect a meaningful difference. If you're time-constrained, a 70/30 or 50/50 split gets you to a reliable answer faster.

Can I run a routing experiment without a dedicated testing tool?

Yes. Many teams use a simple field in their CRM to tag leads as control or treatment and then segment reporting manually. It requires discipline to keep clean, but it works. The more important thing is randomizing the assignment, not the tool you use to track it.

What if my sales cycle is too long to wait for closed-won data?

Use a leading indicator like opportunity creation rate or stage-2 conversion as your primary metric, and clearly document that this is a proxy, not a final answer. Then plan a follow-up review once enough deals have had time to close.

How do I know when to stop a losing experiment early?

Define this in advance. Common practice is to stop early only if the treatment group is performing significantly worse than control, not just because it's underperforming your expectations. If you stop every experiment that doesn't show early promise, you'll never finish one.

---

Better routing is a testable hypothesis, not a gut feeling, and testing it correctly is the fastest path to more closed deals from the leads you already have. More revenue. Same pipeline.