Why would you trust it with everything?

4iG is an eDiscovery and information-governance platform for enterprise legal teams, US-facing. Customers hand it large volumes of unstructured material — email, notes, PDFs, files pulled from Office 365, SharePoint and network shares — and it turns that into something structured: indexed, searchable, and stripped of sensitive content before it leaves the building. I was the only designer on it, and one of the Angular developers building it.

The screens below are not images. Every one is live HTML I wrote, rebuilt from scratch with invented data — open one and the marks beside it explain what you are looking at.

Nothing works until everything is in. The system has to ingest, OCR and index a whole collection before anyone can search it or redact a single line.

So the product’s opening request to a new customer is: hand over everything you have, then wait. And “everything” is the material an organisation is most careful with — privileged correspondence, personal data, regulated records. The collections I designed against ran to millions of documents.

The first run is the hardest moment the product has. It asks for the most and gives back the least.

My domain came almost entirely from one person. The founder had worked in eDiscovery before, and we spoke daily — he walked me through Relativity with dummy data and sent screenshots. From that I learned the shape of a review workspace: native view versus extracted text, work product, tagging, burn and QC on a redaction job, load files, production volumes, connectors, data points.

Two things surprised me. First, the dominant job wasn’t reading — it was mass automated redaction, driven by user-supplied keywords and PII patterns across an entire collection at once. Second, navigation was the real difficulty, not the individual screens. 4iG’s menu ran deep with multi-level sub-navigation, and the redaction screen alone carried a top action bar, two collapsible side rails, and an OCR-rendered document preview in the middle that had to display highlights, masks, borders and cross-lines over text the system did not fully control. I read GDPR guidance on presentation to understand what could legitimately be shown and how.

What I did not have: a lawyer. Not one, in two years.

The founder’s model was straightforward and fast. Point the system at everything, apply keywords and redaction rules in the background as it ingests, show results when they’re ready. One motion, no waiting around for a two-step process.

I argued against it and reversed the order.

Default the first upload to a single small file. It lands in seconds, and the user gets their own document back — folders, sub-folders, extracted metadata, the keywords it found. Then they apply the filters themselves and watch what actually happens to their own content. Only after that does anyone point the system at gigabytes.

The rejected path was faster on paper and worse in the only moment that counted. Bulk-first asks a legal team to hand over every sensitive document they own and wait, with nothing in return but a progress bar.

The upload isn’t a data-transfer step. It’s the trust step, and it’s the only one the product gets.

It also solved a commercial problem. The company was demonstrating to prospective buyers and investors with a product that was only partly built. A small file that processes live, in front of someone, shows the machinery working with whatever exists today. A background job on a large collection shows nothing at all.

The product had no onboarding sequence before this. I built one — and only after it does the collection arrive, by whichever of three routes the customer’s data actually takes.

I optimised the first run for demonstrability. The path that got my best attention was the one a founder shows to a buyer — a single small file, processing live. The path that carried the actual work, a collection of two million documents and the navigation depth that came with it, got less. I spent my design argument on the moment the product is introduced, not the months it is used.

I left before launch. 4iG went live in 2021 and is listed on Relativity’s AppHub, but I can’t tell you which of these decisions survived, because I wasn’t there to see it — and neither the onboarding sequence nor the small-file first run existed in the product as it stood when I left.

What did land, inside the build: the middle-pane screens had been written individually by different developers, same output, different logic, so every bug went back to whoever wrote it. I made the case for consolidating them into one reusable template. It took a month and removed the dependency. Colour and structure went into documented tokens at the same time.

I designed, the founder reviewed, I revised. That was the whole loop, for two years. He knew the domain and I didn’t, so I treated his feedback as the requirement rather than as one informed opinion — and never went and got a second one. I could have found a paralegal to sit with for twenty minutes. I never tried. I’d now treat “the person briefing me knows this domain” as a reason to validate faster, not a reason to skip it.