The Insights Feed: What Changed, and Why It Matters

Most of the interesting signal in a set of financials shows up for a moment and then disappears. A margin band shifts, an effective tax rate jumps eight points year over year, working capital days swing further than usual, and if you happened to be looking at that page at that moment, you saw it. If you weren’t, it was gone the next time someone reloaded. Rexfin already computed all of this on every render. It just never kept it anywhere.

The insights feed is that memory. It takes findings the product was already generating and gives them a home: new, acknowledged, resolved, scoped to your workspace, and each one still linked back to the numbers that triggered it.

Why this matters to a finance team

The value of catching an anomaly early evaporates if nobody wrote it down. A finance team reviewing five or ten portfolio companies, or a single company across many periods, cannot rely on someone remembering that the effective tax rate jumped last time they looked. They need a list that persists, that someone can act on, and that does not silently reset every time the underlying data recomputes.

Just as important is what the feed does not do. It does not let a language model decide what counts as noteworthy. Every entry comes from the same deterministic detectors that already run on the statements and segment pages, the same anomaly logic, applied consistently. No new rules get invented on the fly, and no narrative-writing layer gets to mint a finding of its own. An AI-generated summary can reference what the feed already found. It cannot add to it.

How it works, at a glance

Each finding is keyed by what triggered it, the company, the subject (a specific ratio, an effective tax rate lens, a segment shift), and the period, not by its value. That matters because it means a recompute lands on the same row instead of spawning a duplicate every time the page reloads. You get one line per real issue, not one line per refresh.

The lifecycle is deliberately plain: a finding starts new, a team member acknowledges it, and eventually someone resolves it. There is no required sign-off or mandatory comment to close one out. But the feed does track whether the number behind a finding moves after you’ve acted on it. If you acknowledge a working-capital anomaly and the underlying figure shifts afterward, say a later restatement changes the picture, the finding flags that the number moved since you looked at it. If you resolved it and the number moves again, it reopens on its own. And if a finding simply stops firing because the inputs changed, the feed does not quietly delete it. It fired once, which is a fact worth keeping visible until someone decides to clear it.

On the interface side, there’s a dedicated feed you can filter by company, severity, or status, with the filter state held in the URL the same way quick company comparables pages are shareable. Each company also gets a compact strip of its own open findings, and an unread count shows up where you’d expect it, though never on a page built for external sharing. Severity and status are written out in plain language rather than left to color alone, so the feed reads the same whether or not you can see the color coding.

Every finding cites the inputs behind it and deep-links to the source, the same click-to-source mechanism that runs through the rest of Rexfin. A margin flag isn’t just a red badge. It’s a badge you can open and trace to the actual line items that produced it.

Where the feed stops, honestly

The current feed covers what the existing detectors already catch: year-over-year swings, margin-band shifts, working-capital jumps, and effective-tax-rate anomalies including MENA-specific tax dynamics. It does not yet flag plan-versus-actual variance as an outlier the way KPI variance analysis surfaces variance for review, and it does not track covenant or minimum-liquidity breaches, since those depend on thresholds a team would need to enter rather than anything derivable from the filings alone. Both are natural extensions, not silent gaps, and they’re a fair question to ask if a threshold-based alert is what you actually need today.

What’s there now is the thing that used to disappear on every reload: a durable, deterministic, cited record of what changed and when someone dealt with it. For teams tracking more than a handful of companies, that’s the difference between noticing an anomaly once and actually managing it. See the Rexfin product tour for how the feed fits alongside the rest of the reporting surface, and how it pairs with comments and review when a finding needs a real conversation.