Every small team has a version of this.
A thread goes out. A reply is expected — from a client, a partner, a government body. The day moves on. Somebody meant to follow up. Then a week passes. Then another. By the time anyone thinks to check, the thread has gone quiet and the memory of it has gone with it.
This isn't disorganisation. It's the nature of running a business with a small team and a fast pace. Things slip — not because people are careless, but because the volume of things-to-track always outruns the bandwidth available to track them. The misses are quiet. Nobody knows they happened until they cost something.
This is what AEGIS was built to catch. And this year, for the first time, we ran it on a real company's real data to see if it actually would.
The pilot
The company is 3i Solutions, an IT services firm led by Elvis Smakaj. They work across Albania and Italy, managing complex enterprise systems for clients in banking, government, and regulated industries. Tight deadlines. High-stakes relationships. The kind of environment where a missed reply to a government authority isn't a minor inconvenience — it's a relationship problem, or a compliance problem, waiting to happen.
Elvis agreed to let us connect their business mailbox to AEGIS — read-only, scoped to a single inbox, through a Microsoft Entra application-only credential. No send permission. No write access. No ability to act on their behalf in any way. AEGIS could read and surface. Nothing else.
That boundary isn't a limitation we apologise for. For a firm whose work touches banking and government, it's the entire point.
What it found
AEGIS read the mailbox. It ranked threads by what it thought mattered. And it surfaced a catch.
An outbound thread — sent to an Albanian government certification authority — had gone quiet. A follow-up had been owed for some time and nobody on the team was tracking it. Not because they'd decided to ignore it. Because the thread had simply stopped surfacing. There was no broken process, no missed alarm, no one to blame. The communication had just… slipped below the waterline.
AEGIS flagged it. And it showed its work.
This is the part I want to dwell on for a moment, because it matters more than the catch itself. Every item AEGIS surfaces points back to the exact source it reasoned from — the message, the thread, the moment it's drawing on. The reviewer doesn't have to trust the system's judgment. They can see the evidence and decide for themselves. Approve it, dismiss it, act on it. Nothing moves without a person choosing.
The catch without the source is a suggestion. The catch with the source is something a person can actually evaluate. That distinction is what separates a vigilance system from a notification system.
What we learned
The pilot wasn't perfect. We were honest with Elvis about that from the start.
The catch AEGIS surfaced was real and correctly traced to its source — but it was an older thread, not a recent one. The radar's selection logic at the time favoured recency of ingest over recency of the thread itself, and a re-ingest had made older mail look new. We could have dressed it up as a feature. We didn't. We called it an artifact, explained the mechanism, and told him we were fixing it.
Because we were. The catch itself told us something important about the shape of miss that actually costs money: not just "went cold" but specifically "we sent something, it needed a reply, we got silence, and we stopped chasing." That's a distinct pattern — outbound, seeking a response, with no follow-up and no resolution.
We built a detector specifically for that shape. It handles the cadence of business communication across languages — including the way Albanian is actually written in professional email, with and without diacritics. It now runs as a dedicated tier in the vigilance loop, above general cold-thread ranking.
The pilot didn't just prove that AEGIS works. It told us what to build next.
What comes next for 3i
The next step is straightforward: widen the aperture. More of the mailbox. A second channel. A 3i team member in the seat, reviewing catches in their own words, approving or dismissing — so we can learn from what they keep versus what they drop.
That feedback loop is how the system sharpens. Not by us deciding from the outside what matters to a SAP consulting firm managing enterprise clients. By watching what a person who knows that world considers worth acting on.
And when a catch turns into a concrete outcome — a reply sent, a relationship recovered, a deadline not missed — we'll ask Elvis for one sentence. Not a testimonial. A result. The kind that means something because it's specific and because it happened.
If you're thinking about what a real AI pilot looks like — not a proof-of-concept on synthetic data, but a system connected to your actual operations — this is what it looks like. Smaller than you expect. More honest than most vendors would be. And the catches, when they're real, are exactly the ones that would have cost you something.