SquidSupport
Product reference / Reports

Investigations report

In the app: Reports → Investigations

Needs the 'View Reports' permission.

Helps find out where a number changed: which store, contact reason, channel, country or product is behind a rise or a drop, what changed around that time, and which tickets or refunds are behind it. It has three views: "Why did it move", "Compare cohorts" and "Money".

When the report is available

The tab is only in the Reports strip for organizations that have Squid Agent switched on. For other organizations the page shows "Available with Squid Agent" with a link back to Reports, and no figures.

What an investigation shows

An investigation compares two periods and shows which groups the difference comes from. What it finds is an association: two things that moved together. It does not prove that one caused the other. Only an A/B test shows a cause.

  • Each view asks one fixed question with a fixed list of numbers and ways to split them. There is no free query.
  • Numbers about conversations come from figures that are built once a night, so today is not complete. Refund amounts are taken from the refunds recorded in Shopify, the same way Reports → Refunds counts them.
  • Groups that are too small to say anything about are not shown by themselves. They are added together as "Small groups", and the groups after the six largest contributions as "Everything else". A group is small when both periods have fewer than 30 records for a rate, or fewer than 5 for a count or amount.
  • A teammate whose access is limited to certain stores only sees those stores.

Why did it move

Choose a number, a way to split it, and "Last 7 days vs before" or "Last 30 days vs before". The period ends yesterday in UTC and is compared with the same number of days before it. The default is 7 days, "Refunded", split by "Refund path". Click "Investigate" to run the chosen question; changing the controls alone does not update the result.

Numbers about conversations. All four are counted over the conversations created in the period, in UTC days, with deleted, merged and example tickets and unanswered campaign tickets left out:

  • "Conversations": the conversations created in the period.
  • "Customer messages": the messages customers wrote in those conversations, also the ones written after the period.
  • "Handed to a person": the share of those conversations that the AI handed to a person at some point.
  • "Reopened": the share of those conversations that were reopened at least once. Reopens are only known from the day Squid started recording them.

Numbers about refunds. "Refunded" is the amount refunded, "Refunds" the number of refunds and "Orders refunded" the share of orders that got a refund. They are counted the way Reports → Refunds counts them: the period selects orders by the day they were placed, and the refunds on those orders count when they were made within 60 days of the order. Test orders are left out; cancelled orders remain included. Refund amounts use the organization’s reporting currency, converted at each refund’s date. Refunds without an exchange rate are left out of amounts and reported in a warning.

This matters when you compare two periods. "Last 7 days vs before" compares the refunds on the orders placed in the last 7 days with the refunds on the orders placed in the 7 days before. It does not compare the refunds that were paid out in those weeks. The newest orders have had the least time to be refunded, so the latest period tends to look lower than it will turn out; a drop in a refund number is not by itself good news.

Splits: conversation numbers can be split by "Contact reason", "Store", "Channel", "Handled by" and "Ship-to country". "Refunded" can be split by "Refund path", "Refund reason", "Product", "Ship-to country", "Contact reason" and "Store". "Refunds" has the same splits except "Product". "Orders refunded" can be split by store and country only.

The table shows each group's "Contribution" to the change:

  • For a count or an amount, a group's contribution is its own change, and the contributions add up to the total change.
  • For a rate, a group can contribute in two ways: its own rate changed, or its share of the total changed. A store with a high handoff rate that suddenly gets more of the conversations raises the overall rate without its own rate moving. The contribution combines those two parts; the table shows their combined contribution, not separate values for them.
  • "There is no change to split" means one of the two periods has no data.

"Show records" on a named group’s row opens "Records behind …": the refunded orders or the conversations of that group. Refunds show the order, store, country, path, reason and amount. Conversations show ticket numbers with their store, channel and creation date. Up to 20 records are listed. For "Handed to a person" and "Reopened", the records are the conversations that were handed over or reopened. "Orders refunded" has no records view: its button returns an error asking you to use "Refunded" instead. Listing the tickets also needs the permission to view conversations; without it you only see the count.

"What changed around then" lists the changes recorded in the period and in the days before it: changes to the AI configuration, knowledge, rules and integrations, a configuration set assigned to a store, A/B tests started or ended, and Squid releases. When data is available on both sides, a change shows the number for up to 7 days before and up to 7 days starting on its date. Counts and amounts are shown as daily averages; rates are shown as percentages. A rate needs at least 30 records on each side. The timeline includes the comparison period and the 7 days before it, and lists at most 200 changes. A change near a movement is a lead to look into, not proof. Some changes outside your access are counted but not listed. Configuration assignments and A/B tests are omitted without a hidden count when you lack their viewing permission.

"Save as finding" keeps the conclusion with its evidence level. It changes nothing in your setup. Saving the same question again updates the same finding when its leading group is unchanged. "Save as finding" appears only when there is a named group contributing in the same direction as the overall change.

Compare cohorts

Puts two groups side by side for one number. Choose the number, then a "First cohort" and a "Second cohort": for each, what to split by and which value, for example ship-to country code DE against FR. It always uses the last 30 full UTC days. Click "Compare" to run it; both cohort values are required.

Each cohort shows its number and how many conversations, orders or refunds it is based on. A cohort that is too small reads "Too few … to compare (n = …)" and gets no number, because a rate over a handful of cases says little. Rates need at least 30 orders or conversations; counts and amounts need at least 5 conversations or refunds. The difference is shown when both cohorts qualify. For rates it also shows a 95% range and says when the gap may be chance.

Money

Two cards, for "Last 7 days vs before" or "Last 30 days vs before". The default is 30 days, ending yesterday in UTC. Period and split changes update the cards automatically.

"Why refund cost moved" splits the refunded amount of the period and its change against the period before. Buttons choose the split: refund path, refund reason, product, ship-to country, contact reason or store. The product split allocates each refund across the order’s products by their paid value; it does not identify the particular product that was refunded. Without a comparison period the card is titled "What refunds consist of" and shows each group's share. The amounts are Shopify's recorded refunds. The Refund Hub is only used for the reason of a refund, not for amounts. This card is also shown on Reports → CS economics. Like the refund numbers above, it counts the refunds on the orders placed in the period. Its total therefore differs from the "Refunds" line of the CS cost, here and on Reports → CS economics, which counts the refunds paid out in the period.

"CS cost by …" splits customer service cost by "Contact reason", "Store" or "Country". It uses the same cost lines and total as Reports → CS economics. Refunds are counted for their own order's store, country or contact reason; every other cost is shared out by each group's share of the conversations created in the period. This card needs the 'View customer service costs' permission; without it, a note says an Owner or Admin can add it to your role. When no worked hours are uploaded for the period, labour is not included and the card links to Settings → Costs.