Insights AI
The assistant in the web dashboard — how-to answers for every store, answers about your own sales and stock with the add-on, in English, Filipino, Taglish or Bisaya — and the separate insight panels on report pages.
7 min read · Updated 30 Oct 2026
On this page
Two different things in Kassly are called insights, and they work in completely different ways. This page covers both.
| Feature | Where | Needs |
|---|---|---|
| Insights AI — you type a question | Sparkle button, bottom right of the web dashboard | How-to answers: every store. Answers about your sales and stock: the Insights AI add-on |
| Insight panels — Kassly flags something | Top of the report pages and the Reports hub | Business Reports add-on |
Asking a question
The sparkle button sits at the bottom right of every page of the web dashboard, including in a phone's browser and in Kassly Desktop's Web Admin. It stays off the POS and kitchen screens, where it would cover buttons. The phone app doesn't have it.
Type a question the way you would say it — "Magkano benta ko kahapon?", "How do I void a sale?" — or tap one of the suggestions. The answer appears as it is written, in the language you asked in: English, Filipino, Taglish or Bisaya.
Ask a follow-up such as "and last month?" — the assistant reads your last few messages. Your conversation is kept on that device, the last fifty messages, so closing the panel and coming back does not lose it. It does not follow you to another device, and nobody else in your store sees it.
How to use Kassly — every store
Any question about doing something in Kassly is answered from the app's Help & Manual page: voiding a sale, adding staff, connecting a receipt printer or cash drawer, finding a Z-reading. The answer gives the steps and links the Help articles it came from, so you can open the full article.
Anyone who can open the web dashboard can ask these, in any store, with no add-on.
Your sales and stock — with the add-on
With the Insights AI add-on, the assistant also answers about your own store:
| Ask about | For example | What comes back |
|---|---|---|
| Sales for a period | How were sales today? | Total sales, number of sales, average sale |
| A comparison | How does this week compare to last week? | Both periods side by side with the change. A period still running is compared up to the same point: Monday to Thursday 3pm against last Monday to Thursday 3pm |
| Best sellers | Top 5 products this month | A ranked table |
| One product | How many Chickenjoy did we sell this week? | Units and sales for that product |
| What is running low | Ano ang paubos na? | The reorder list |
| Stock of one product | How many San Mig Light are left? | Stock on hand, per branch |
The numbers come from the same queries as your reports, for the branches you can see. Name a branch to narrow it to one. The assistant only puts the figures into words; it does not work them out itself.
Sales and stock answers need a role that can see reports — Owner and Manager by default. Without the add-on, the assistant says these come with Insights AI and links Billing & Add-ons; it never shows a figure.
The periods it reads
It answers for today, yesterday, this week, last week, this month or last month. A question without a period means today, except best sellers, one product and comparisons, which mean this month. Every answer names the dates it covers — read them before you trust the number. For any other range, such as a month by name or the last 90 days, use the reports.
Insights AI uses plain calendar days, not your business-day cutover. Every other report in Kassly buckets a 1am sale under the trading day that opened the previous morning. Insights AI does not — it uses midnight-to-midnight in Philippine time. If you trade past midnight, chat answers will disagree with your dashboard and your reports, and the reports are the ones that are right. A store with the cutover left at midnight sees no difference at all.
What it won't do
The assistant is for your store and for Kassly. It declines anything else — writing posts or homework, general knowledge, code — and anything that tries to change how it behaves. It never sees another store's data. A person whose questions are declined five times in a day is paused until the next day.
Limits and record-keeping
| Limit | Value |
|---|---|
| Questions per minute | 10 per person |
| Questions per month | 300 per store, how-to and sales questions together. Resets on the 1st |
| Question length | 500 characters |
| Conversation kept | Last 50 messages, on that device only |
Every question is written to your audit log — what was asked, what kind of question it was, and whether it was answered or declined. Owners can read that back under Audit logs.
To read your question and write the answer, Kassly sends the question, your last few messages and, for sales and stock questions, the figures that answer it to its AI service providers.
The footer of the panel says "Insights AI can make mistakes. Verify important numbers in your reports." Take that literally. The figures are the ones the reports use, but reading your sentence — especially the period — is the part that can go wrong.
Insight panels on report pages
Completely separate feature, part of Business Reports. These are the coloured cards at the top of a report page that say something is off. You do not ask for them; they appear when a rule fires.
They are threshold rules, not learning. Each one is a fixed comparison, computed over whatever date range the page is showing, and cached for five minutes.
| Signal | Fires when |
|---|---|
| Sales drop | A day inside your range is 25% or more below the same weekday one week earlier. Marked high at 40% or more. Up to three days flagged |
| Discount outlier | A cashier's discount-to-sales ratio over the last 30 days is at least twice the store median. High at three times. Cashiers with fewer than ten sales are skipped |
| Margin issue | A product's margin is half its category's average or worse, having sold at least ten units. Up to five products |
| Transaction gap | Three or more consecutive hours with no sales at all between 9am and 9pm. High at four hours or more. Up to three gaps |
| Branch variance | A branch running at 70% or less of the average of its peers |
| No-show rate | Services businesses only — 15% or more no-shows over the range |
| Dead stock | Stock that has not moved in 60 days is worth more than 10% of your tracked inventory value |
| Product drift | Products that entered or dropped out of your top ten against the previous period |
Some of these deliberately stay quiet:
- Branch variance is skipped for a single-branch store, and skipped when you have filtered the page to one branch. Comparing a branch against itself means nothing.
- No-show rate needs at least ten appointments in the range, so one cancellation on a quiet week does not raise a flag. It is skipped entirely for retail and food-and-beverage stores.
- Dead stock is a retail signal. It is skipped for services businesses, and it needs your inventory value to be non-zero.
- Margin issue and dead stock are hidden from anyone who cannot see financial figures, because their headline numbers are money.
Compare thresholds with the ones on the Dashboard. The dashboard's large-discount rule is a different rule with a different definition, and the two are not meant to agree.
If a signal cannot be computed — a detector that hits missing data, say — it is skipped quietly rather than blanking the whole panel. So an empty insights panel means "nothing to flag", not "something broke".
Which to use
The panels are better at finding problems, because they run without being asked and they compare against a baseline. Insights AI is better for a specific question you already have, without navigating to a report — and for asking how to do something in Kassly.
Neither replaces the reports themselves. For anything you are going to act on — or send to your accountant — go to Sales reports or Profit reports and read the figure in context.