Forecast report
In the app: Reports → Forecast
Needs the 'View Reports' permission.
Estimates the next 8 weeks: how many tickets to expect, the refund rate of the orders of those weeks, and how many people are needed in the busiest week. Every forecast figure is an estimate with a range, not a promise.
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.
A forecast needs at least 8 weeks of tickets, counted from the first day with a ticket through the Sunday before the forecast starts. Until then the page shows "Too little history" with the number of days Squid has so far. When the nightly forecast has not been built yet, or failed, the page says so instead of showing numbers.
Store and how the forecast is made
"Store" chooses one store or "All stores". The menu appears when you have more than one store, or whenever your access is limited to particular stores. A teammate whose access is limited to certain stores can only pick their own stores and has no "All stores". The nightly build makes a forecast for all stores together, then for up to the 25 busiest stores of the last 8 full weeks. Individual stores can be deferred when the build runs out of time.
The forecast is rebuilt once a night from up to a year of daily ticket and order counts. It uses statistics only: the pattern of the past weeks, including the rhythm of the weekdays. It does not know about a sale, a campaign or a product launch you are planning; use "What if" for that.
What is counted. A ticket counts on the day it was created. Deleted tickets, tickets merged into another one, the example ticket and outbound campaign tickets the customer never answered are left out, as in Reports. An order counts on the day it was placed; test orders and cancelled orders are left out. Days are UTC days.
Every estimate is shown as a middle value with a range. The range is where the figure is expected to fall four times out of five: one time in ten it will be lower, one time in ten higher.
A line under the chart names the method and the forecast quality: "good", "fair" or "poor". Squid tests the method on your own past weeks, by forecasting up to four weeks ahead from earlier dates and comparing with what happened. The line gives the average weekly error and the share of past weeks that fell inside the range. "Good" means an error of at most 10% with at least 60% of the weeks inside the range; "fair" means at most 25% with at least 40%; anything else is "poor". With fewer than four past weeks to test, the quality is not rated yet.
Forecast tiles
- "Tickets · week of …": the tickets expected in the first forecast week, named in the label. The forecast starts on the Monday of the week in which it was built, so this is normally the week that is running now, not the week after it.
- "Tickets · busiest week (…)": the tickets expected in the busiest of the 8 weeks.
- "Refund rate · orders of these weeks": the share of the order value of the coming weeks that is expected to be refunded in the end. A refund comes days or weeks after its order, so recent weeks always look better than they will turn out. The forecast therefore uses the last 12 weeks of orders that are at least two weeks old, each corrected for the refunds that are still expected to come. It needs at least four such weeks and enough refund history to know how long refunds take; otherwise the tile says why there is no figure.
- "People needed · busiest week": see "People needed". When the figure is under a tenth of a person, or cannot be shown because no worked hours have been added, a note under it says which share of tickets people handled in the last 4 weeks.
A tile without enough data reads "Not enough data to estimate this" instead of a number.
Tickets per week
A chart of the actual tickets of the past weeks followed by the 8 forecast weeks, with the range drawn around the forecast. Weeks start on Monday and are UTC weeks.
People needed
"People needed in the busiest week" turns the ticket forecast into a number of full-time people, per language.
- "Language" and "Share of tickets": the reply languages of the tickets created in the last 4 full weeks. The eight largest languages are listed; the rest are added up as "Other languages". Tickets whose language was not determined are shown as "Language not detected".
- "Handle time": the measured minutes a person actively spends on a ticket in that language, averaged over the tickets for which Squid measured activity. A language with fewer than 20 measured tickets uses the average over all languages.
- "Needed" and "Range": the people needed for that language. The "Total" row adds the languages up.
How it is worked out: the forecast tickets, times the share of tickets a person handled in the last 4 weeks, times the handle time, divided by the productive hours of one person in a week. A ticket counts as handled by a person when a teammate wrote and sent the customer at least one reply on it, with or without the AI. An AI draft that a teammate approved and sent counts as the AI's reply here, not as the teammate's. Productive hours per day come from Settings → Costs; until you set them Squid uses 6.5 hours a day, 5 days a week. The card states which values it used.
People needed can only be worked out once worked hours have been added under Settings → Costs; until then the card says why and links there. It is a planning estimate. It is not a saving and not a cost.
What if
Recalculates the forecast with your own assumptions. Nothing is saved, and the forecast itself does not change.
- "Orders above the forecast (%)": expected extra orders, for example 20 for a sale. It can be set from -50 to 200. Tickets rise or fall by the same percentage in every forecast week.
- "AI resolution share (%)": the share of tickets on which every reply to the customer was written by the AI, including AI drafts a teammate approved and sent. It is not the share the customer confirmed as resolved, and it is counted differently from "Handled by AI alone" on the "AI" tab of Reports, where an approved draft counts as a person's reply. The field starts at the share of the tickets created in the last 4 full weeks, rounded to a whole number. The field is disabled when that share is not known yet.
What changing "AI resolution share (%)" does. The difference between the number you type and the measured share is moved between the AI and people. Typing a share 10 points higher lowers the share of tickets a person handles by 10 points, and typing a lower one raises it. Tickets that got no reply at all stay where they are. The tickets a person handles, and with them the people needed, follow in proportion: if people handle 40% of tickets today and you raise the AI share by 10 points, people handle 30%, a quarter fewer tickets and a quarter fewer people. One recalculation can move the share by at most 50 points, and the share a person handles never goes below zero. Leaving the field as it started changes nothing. The ticket forecast, the handle time per ticket and the refund rate are not changed by this field.
- "Recalculate" shows the result: the people needed in the busiest week with a range, and the tickets a person handles over the 8 weeks.
- People with the 'View customer service costs' permission also see "CS cost per order" for the scenario. It is built from the 28 days before the forecast’s starting Monday: refunds, vouchers and replacements per order, plus tool and staff costs spread over the orders. Extra orders spread the tool and staff costs over more orders and lower it. More automation changes the people needed, not the staff cost line: Squid never puts a price on handling time.
The last line names the assumptions that were applied.