Best practices

We sit on a lot of data about what works. This page distills the patterns the highest-converting Zubby stores share. None of these are rules — every store is different — but each is a defensible default.

Setup week

Install on Monday. You want a clean week of data before drawing conclusions.
Spend an afternoon on the knowledge base. Write 30 great FAQ entries. Highest-impact single activity in the first two weeks.
Set a brand voice. Pick a tone preset and add 5 opinions and 5 do-not-say items. The default AI is fine; a brand-tuned AI is exceptional.
Turn on cart rescue with the Classic 3-step template. Don’t customize. Run the default for two weeks to get a baseline.
Set a sensible cost cap. 1.5× your projected monthly spend. Catches infinite-loop scenarios early.

System prompt patterns

Things to include and exclude:

Do include: brand POV, hero product callouts, do-not-say list, escalation triggers, tone in 1 sentence.
Don’t include: long product taxonomies (the AI retrieves them anyway), exhaustive pricing rules (use the discount band setting), case studies, marketing copy.

Knowledge base patterns

Write FAQs as questions, not declaratives. Retrieval matches better.
Pull policies in verbatim. Don’t summarize. The AI quotes them.
Add one section per “first principle” — what makes us special, what we don’t do, our point of view on the category.
Refresh quarterly. Stale knowledge erodes trust faster than no knowledge.

The 30-FAQ test

Pick the 30 most common questions your support team answers. Add them to the KB. Retrieval coverage will jump 15-25 percentage points — guaranteed.

Widget patterns

Launcher on every page is the right default.
Add an inline widget on landing pages where the AI is the hero — gift-finder, quiz, product comparison.
Don’t use exit-intent popups on cart. They feel desperate. Cart abandonment recovery via email works better.
For mobile-first stores, consider fullscreen on a dedicated /ask page with a deep-link from your nav.

Recovery patterns

Send from a human address with a real reply-to. Reply rates double.
Don’t discount on the first email. The Classic 3-step waits until email 3.
Keep frequency cap at 1 email / day / shopper. Resist the urge to stack with Klaviyo.
Browse rescue at 4 hours outperforms 24 hours by ~30%. Faster is fresher.

Training patterns

Triage the gaps inbox once a day for the first two weeks. Weekly after that.
Accept or edit, rarely reject. Even a partial fix improves coverage.
Turn on auto-promote after 50 manual accepts.
Replay last 50 conversations after big changes to see real impact.

Analytics patterns

The metric that matters: assisted revenue / billable conversation. Watch the ratio, not the raw numbers.
Use 1-day attribution by default. 7+ days is for considered purchases.
The Monday weekly review is the highest-signal artifact in the dashboard. Read it religiously for the first 8 weeks.

Operations patterns

Configure SLAs per channel. Email has different expectations than SMS.
Route VIP escalations to a single owner — the founder during early stage, the CX lead at scale.
Re-train quarterly. Set a calendar reminder; product copy and brand voice drift.
Audit handoff reasons monthly. A growing “low confidence” share means the gaps inbox needs attention.

Anti-patterns to avoid

Over-customizing the system prompt — keep it under 500 words.
Building too many widget variants — three is plenty.
Discounting on the first recovery email — sets a bad expectation.
Ignoring the weekly review.
Mixing staging and production stores in one workspace.

Where to go next

Setup week

Install on Monday. You want a clean week of data before drawing conclusions.

Spend an afternoon on the knowledge base. Write 30 great FAQ entries. Highest-impact single activity in the first two weeks.

Set a brand voice. Pick a tone preset and add 5 opinions and 5 do-not-say items. The default AI is fine; a brand-tuned AI is exceptional.

Turn on cart rescue with the Classic 3-step template. Don’t customize. Run the default for two weeks to get a baseline.

Set a sensible cost cap. 1.5× your projected monthly spend. Catches infinite-loop scenarios early.

Knowledge base patterns

Write FAQs as questions, not declaratives. Retrieval matches better.

Pull policies in verbatim. Don’t summarize. The AI quotes them.

Add one section per “first principle” — what makes us special, what we don’t do, our point of view on the category.

Refresh quarterly. Stale knowledge erodes trust faster than no knowledge.

The 30-FAQ test

Pick the 30 most common questions your support team answers. Add them to the KB. Retrieval coverage will jump 15-25 percentage points — guaranteed.

Widget patterns

Launcher on every page is the right default.

Add an inline widget on landing pages where the AI is the hero — gift-finder, quiz, product comparison.

Don’t use exit-intent popups on cart. They feel desperate. Cart abandonment recovery via email works better.

For mobile-first stores, consider fullscreen on a dedicated /ask page with a deep-link from your nav.

Recovery patterns

Send from a human address with a real reply-to. Reply rates double.

Don’t discount on the first email. The Classic 3-step waits until email 3.

Keep frequency cap at 1 email / day / shopper. Resist the urge to stack with Klaviyo.

Browse rescue at 4 hours outperforms 24 hours by ~30%. Faster is fresher.

Operations patterns

Configure SLAs per channel. Email has different expectations than SMS.

Route VIP escalations to a single owner — the founder during early stage, the CX lead at scale.

Re-train quarterly. Set a calendar reminder; product copy and brand voice drift.

Audit handoff reasons monthly. A growing “low confidence” share means the gaps inbox needs attention.

Anti-patterns to avoid

Over-customizing the system prompt — keep it under 500 words.
Building too many widget variants — three is plenty.
Discounting on the first recovery email — sets a bad expectation.
Ignoring the weekly review.
Mixing staging and production stores in one workspace.