For your CAUpdated 12 Aug 20268 min

Report insights, anomalies and your digest

Findings above your reports (unusual spending, revenue swings, possible duplicates) worked out from your own books with the figures behind each one, an optional plain-language summary whose every figure is checked, a working-capital screen, and a digest you can have emailed.

Several reports now carry a short strip of findings above the numbers (unusual spending, a revenue swing, something that looks entered twice), with the real figures behind each one. You can ask for a plain-language summary of a report, see how fast money comes in against how fast it goes out, and have a daily digest emailed to you.

Where do these appear?

The findings strip is on Profit & Loss, Outstanding, GST & returns and, in full, on Exceptions. Working capital and Your digest are screens of their own. They need the Reports module.

How does it decide something is unusual?

By comparing a month against the middle of your own history rather than the average, so one freak month last year cannot make this year look normal. A category is only flagged when three things are true at once: it is at least two and a half times its usual level, the difference is more than about a thousand rupees, and it is far enough outside your normal spread to be worth a look. Spending is judged on the month in progress; revenue on the last complete month. Possible duplicates are receipts or bills of the same amount from the same party within a week.

You can move those thresholds. Only the workspace owner sets them, since they change what everyone gets flagged for.

Why does it say it needs more history?

Because with fewer than three months of activity there is nothing to call unusual, and it says so rather than guessing: anomaly checks need N more months of history before they can spot unusual months. Months with no activity do not count as history, so a brand-new expense category is never flagged on its first entry. When there genuinely is nothing to report, it says nothing unusual in your books right now.

Findings that are only possibilities are worded as such, for example this receipt may have been recorded twice, and each one carries the actual amounts, dates and counts, so you judge it rather than take its word.

The findings in full on Exceptions. Each one carries the amounts, dates and counts it was raised from, so you can judge it. (desktop view)
The findings in full on Exceptions. Each one carries the amounts, dates and counts it was raised from, so you can judge it.

What is “Explain this report”?

It is the one part of this that uses the assistant, and only when you tap it; nothing fires when the page loads. Finocket computes a small set of facts from the report, hands only those to the assistant, and asks it to put them into a few sentences. Every sentence is then checked back against those facts. A sentence is dropped if it has a figure that is not one of them, a date that is not in the report, or a citation of a fact that does not exist. Each surviving sentence shows numbered citations you can tap.

The summary adds no numbers of its own. The figures are already on the page, and the line underneath says as much: numbers come from your books, AI wrote the words. When nothing survives the check it says there was nothing worth adding to these numbers, and when your daily allowance is spent or the provider is unreachable it says which. In all three cases the report itself is unchanged.

Treat the prose as a summary, not as advice. The assistant is instructed not to predict the future or give tax, legal or investment advice; what is enforced in code is that every figure is one of your own.

What is on the Working capital screen?

How many days of sales are sitting in your receivables, how many days of purchases in your payables, and the gap between them, plus who your money is stuck with and how much of it rides on one customer. Every tile shows the two numbers it was worked out from, and the screen names its own convention: closing balances against a window you choose (30, 90 or 365 days), counted inclusively.

It also names its own blind spot on the page: cash sales are not counted as revenue here, so a mostly-cash business reads a higher receivable-days figure. Where a figure cannot be worked out (no revenue in the window, say), it prints NA with what is missing, never an infinity and never a plausible-looking guess. Nothing on this screen is AI-generated, and it says so.

Working capital: receivable and payable days and the gap between them, each tile showing the two figures it was worked out from. (desktop view)
Working capital: receivable and payable days and the gap between them, each tile showing the two figures it was worked out from.

What are the five models?

Models is a separate screen with five answers worked out the same way as everything above. They come from your own sales, stock and customers, never from an AI, so every figure can be checked against the report it came from. What will sell next month; what to reorder and by when; which regulars have stopped coming; how each marketplace is doing; and how the counter is selling.

Each model says which of four things happened: it answered, it has some history but not enough, nothing has been recorded for it yet, or the read failed. A model that could not be read never shows you a zero. A forecast is a range: the low and the high are an 80% band from how much your day-to-day sales actually vary. A product that sold on fewer than eight days in twelve weeks is listed as left out rather than forecast at near zero. Reorder dates depend on your supplier’s lead time; where none is recorded the row says the lead time was assumed, so set the real one on the vendor and the dates move.

How do I get the digest by email?

It is one of the reports you can schedule. Open Scheduled reports, add a schedule for Insight digest, and pick daily, weekly or monthly. Only someone with write access to reports can create or pause a schedule.

Email is the only way it is delivered. There is no push and no WhatsApp digest. If no email provider is connected on your deployment, the run is recorded as skipped with that reason and nothing is faked; the screen tells you and links you to sending setup. Each recipient goes through the same consent check as any other message, so one blocked address does not stop the others.

The digest has five sections: cash outlook, money you are owed, money you owe, GST deadlines, and things worth a look. It distinguishes three cases that look identical if you are careless: a section that is genuinely empty is left out, a section your schedule switched off is silent, and a section that could not be read says this couldn't be read just now, so it's left out rather than shown as zero. The emailed copy carries exactly the numbers the screen shows.

The digest on screen before you schedule it, with the same sections and figures the emailed copy carries. (desktop view)
The digest on screen before you schedule it, with the same sections and figures the emailed copy carries.

The digest's optional written summary needs your own AI key or the Pro add-on, because an unattended nightly job will not spend a free workspace's chat allowance. Where it is unavailable, off, or the summary did not match your numbers, it says which and the figures are unaffected.

Related: Cash outlook & forecast, Scheduled reports, Budgets & ratios.

Does this work offline?

Frequently asked questions

How does it decide something is unusual?

It compares a month against the middle of your own history rather than the average, so one freak month cannot skew it. A category is flagged only when it is at least two and a half times its usual level, the difference is more than about a thousand rupees, and it sits far enough outside your normal spread. Possible duplicates are same-amount records from the same party within a week.

Why does it say it needs more history?

With fewer than three months of activity there is nothing to call unusual, so it says how many more months it needs instead of guessing. Months with no activity do not count, so a brand-new category is never flagged on its first entry.

Can the AI summary invent a figure?

Every sentence is checked back against the facts Finocket computed for that report. A figure that is not one of them, a date not in the report, or a citation to a fact that does not exist: any of these and the sentence is dropped. Where nothing survives it says there was nothing worth adding to these numbers.

Is the summary advice?

No. Treat it as a summary of numbers already on the page. The assistant is instructed not to predict the future or give tax, legal or investment advice; what is enforced in code is that every figure it states is one of your own.

Why does a working-capital tile say NA?

Because the figure genuinely cannot be worked out, for example no revenue in the window. It prints NA with what is missing rather than an infinity or a plausible-looking guess. The screen also names its own blind spot: cash sales are not counted as revenue there, so a mostly-cash business reads a higher receivable-days figure.

How is the digest delivered?

By email, as a scheduled report: daily, weekly or monthly. If no email provider is connected the run is recorded as skipped with that reason and nothing is faked. A section that could not be read says so rather than showing as zero.

Can something check for unusual spending without me opening the report?

Yes. The Unusual spend watch automation at /agents runs this same finding on a schedule and tells you only on the days something is flagged. It reads the thresholds you already set here, so it never flags anything this page would not.

If it flags nothing for months, have I wasted the money?

A run that finds nothing never reaches the AI, so a quiet day costs nothing. But be honest about the monthly figure: a daily watch is about ₹8.70 a month whether or not it ever catches anything, because you are paying for the watching rather than the finding. Six months of silence you never acted on is a reason to review the automation, not to keep paying for it.

Related articles