Product Assistant analytics

This page explains how to read Product Assistant performance so you can make the right business decisions. It is not a technical API reference, but a practical interpretation of the metrics.

Where to find the report

Path: Insights › Analytics › Product assistant analytics (/admin/settings/product-assistant-analytics)

At the top, choose the period 7d, 30d, 90d or 1 year — the report recalculates right away. To see the report you need permission to view reports (reports.view). You manage the assistants themselves in Insights › Analytics › Product Assistants.

It's best to compare equivalent periods (e.g. the last 30 days vs. the previous 30 days), not random time spans.

How to read the assistant funnel

The basic shopper journey:

1. Impression - the assistant was shown.
2. Start - the shopper began answering.
3. Result view - the shopper reached the results.
4. Product click / Add to cart - interaction with a product.
5. Purchase / Assisted purchase - a purchase attributed to the assisted channel.

If the funnel drops already between impression and start, the problem is usually the text, the placement or the relevance of the first step. If it drops between start and result view, the cause is usually the quality of the questions or overly strict rules.

Meaning of the main cards

• Impressions - how many times the assistant was shown.
• Starts - how many times the assistant was started.
• Result views - how many times a shopper reached the results.
• Add to cart - add-to-cart actions from the assistant.
• Purchases - the total number of purchases attributed to the assistant.
• Assisted purchases - purchases where the assistant helped with the decision.

Key ratios:

• Start rate = starts / impressions x 100
• Result view rate = result views / starts x 100

The start rate can be above 100%. That in itself is not a report error, but a property of how events are counted.

Purchase vs. assisted purchase

• Purchase: the purchase is directly attributed to the interaction with the assistant for that product.
• Assisted purchase: the assistant played a supporting role on the way to the purchase.

How to read it in practice:

• high purchase = the assistant can close the decision,
• high assisted purchase = the assistant helps shoppers find their way, even if the last step of the purchase happens elsewhere.

Both metrics matter. Look at them together when you evaluate performance.

Events by type

Use the Events by type table to find where shoppers drop off:

• many starts, few result views -> the questions are hard to get through,
• many result views, few clicks/ATC -> weak relevance of the results or a weak offer,
• many clicks, few purchases -> the problem may lie outside the assistant (product detail, cart, checkout).

Top products

The Top products from assistant results table is a good signal for optimization:

• which products the assistant naturally pushes forward,
• which products get clicks but don't bring purchases,
• which questions probably lead to well-performing products.

If the top products consistently don't reflect your business priorities, adjust the answer rules in the assistant.

By assistant

If you run several assistants, compare:

• which one has the best start (start rate),
• which one has the best progression to results,
• which one brings real purchases.

Don't judge an assistant by impressions alone. What matters is the whole journey through to the purchase.

Common wrong conclusions

• "We have lots of impressions, so the assistant works."
  Being shown doesn't mean quality yet.
• "Few purchases = the assistant is bad."
  The problem is often in the product, the price or the checkout, not in the assistant.
• "Looking at one metric is enough."
  Only the combination of funnel metrics gives the right picture.

When purchases stay low

Check:

• whether the assistant's results include products customers actually buy,
• whether the rules are too narrow (few relevant results),
• whether the shopper goes through the same browser flow all the way to the order confirmation,
• whether problems only arise after the click, outside the assistant (product detail, cart, checkout).

Recommended optimization cycle

1. Change one part of the assistant (a question or the rules).
2. Watch how the metrics change for at least 1-2 weeks.
3. Evaluate the impact on start rate, result view rate and purchases.
4. Only then make the next bigger change.

This way you can identify which change really improved performance.