The standard optimisation playbook assumes high traffic and a same-session purchase. B2B has neither.

What works instead

• Session recordings and user interviews with actual buyers — a dozen well-chosen conversations usually beat an underpowered test.
• Funnel instrumentation so you can see where multi-session journeys break down.
• Friction removal in reorder, search by SKU, and account setup, which are where B2B revenue actually lives.
• Sequential testing or holdout groups where sample size will not support a clean split test.

Measure the right thing

Optimising for add-to-cart rate on a site where half the revenue comes from quote requests will make the numbers look better and change nothing. Agree the primary metric before any work starts, and make sure it connects to revenue.

The purpose of a B2B channel is usually to serve existing accounts more cheaply and more reliably, not to acquire new traffic. Measure that.

Metrics worth a dashboard

1. Self-service adoption — percentage of orders placed online rather than by phone or email.
2. Reorder rate — how often accounts come back without prompting.
3. AOV by account segment — whether the channel grows baskets or just moves them.
4. Quote-to-order conversion and the time between the two.
5. Cost to serve — support contacts per order, which is where the business case usually lives.

Metrics to stop reporting

Bounce rate, time on site and raw sessions. In a channel where a buyer logs in, reorders in ninety seconds and leaves, a low time on site is a success.

Performance work has a reliable order of operations, and skipping to a rebuild usually means the same problems reappear on new code.

Work in this order

1. Measure with field data — real user monitoring, not a single lab score on a fast connection.
2. Images — correct formats, correct dimensions, lazy loading below the fold. Usually the single biggest win.
3. Third-party scripts — audit every tag. Most sites carry several nobody can account for.
4. Render-blocking CSS and JS — defer what is not needed for first paint.
5. Server and cache — full-page caching, query optimisation, and a CDN configured for your catalog.

Set a target that means something

Pick a threshold tied to revenue rather than a score out of a hundred, and track it per template — category pages, product pages and checkout behave very differently and average badly.