Inveris Digital — You cannot transform what nobody can explain

Why the AI spend returned nothing

The models were never the constraint. The knowledge they had to reason over was.

Every enterprise AI programme rests on one assumption nobody stress-tested: that what the machine needs to know is already written down. In the systems that actually run your company, it never was.

Every dollar you have spent on AI so far went to the fraction of your knowledge that somebody happened to document. The rest — the reasoning, the edge cases, the why — was never in the corpus, and no amount of model capability recovers it. This is not a gap in the technology. It is a missing floor underneath all of it.

There is now a whole category selling you the missing context — architectural maps of your codebase, indexes across your documents and applications. It is real work and it helps. But every one of those tools is built from what somebody already recorded. The reasoning that decides your edge cases was never recorded anywhere, and no connector reaches it. That is the part we go and get — from the people who still hold it, while they are still here.

You bought the models.Same licences, same morning, available to every competitor you have.

You wired up the data.Warehouses, lakes, pipelines — all of it pointed at what had already been written down.

You waited for the return.It did not arrive, and the diagnosis landed on the model, the vendor, or the change programme.

The first rung is the one everybody assumed they already had.

You own the audit.An investigator will ask you to explain a system nobody left can explain.

You own the submission.The design rationale is in firmware and in two engineers’ heads.

You own the migration date.Three to seven years of programme, sitting on an as-is nobody wrote down.

What your peers are working on

Six problems running in life sciences right now — and the shape of the work that resolves each one.

No client names and no case studies. These are the situations we see repeatedly across regulated MedTech, and the form the work takes when IRIS is pointed at one of them. Each is a project you could scope on Monday. If one describes something sitting in your estate, that is the conversation.

01

Read the payer logic out of the platform that runs it

The situationTwenty years of claims and contract rules inside a billing platform. The two analysts who understand it are the same two keeping it running in production.

Pointed atThe source, the modification specs, and the analysts — in their own language.

An auditable specification of how the rules actually resolve — so the processing path can be automated against rules somebody has verified, and defended when the pricing is questioned.

02

Write FDA-ready documentation from the firmware that ships

The situationA submission waiting on design rationale that exists in two engineers and nowhere else.

Pointed atDevice software, the existing design files, and the engineers who wrote them.

DHF content, hazard analysis, FMEA and traceability with provenance on every line — and traceability that stays live as the code changes.

03

Delete the discovery phase of an ERP migration

The situationFifty-plus bolt-on applications around a two-decade-old instance, a handful of business analysts, and a three-to-seven-year programme resting on an as-is nobody has written down.

Pointed atBolt-on codebases — RPG, COBOL, Java, PL/SQL — reconciled against the modification specs, plus the base-configuration knowledge only the analyst holds.

A documented as-is and fit-gap input in weeks rather than months of workshops — and an as-is that stays current through the programme instead of ageing out in month four.

04

Make the quality system explainable before an investigator asks

The situationQMSR is in its first inspection cycle. Deviation routing and CAPA escalation live in code and in precedent, not in documents.

Pointed atThe quality systems, the workflow logic, and the QA veterans who apply the precedent.

Documented decision logic with provenance — continuous gap analysis instead of a pre-audit scramble, and an audit response assembled in days.

05

Understand an acquired platform before the people leave

The situationPost-close, orphaned systems, and the team that understood them falls outside the org chart within twelve months.

Pointed atThe inherited estate and the parent’s experts, while they are still reachable.

What it does, what it commits you to, and what breaks if you switch it off — so retire, replicate or rewrite is decided per module rather than assumed for the estate.

06

Capture a retiring expert while they are still here

The situationOne person holds fifteen years of system logic. The knowledge still exists — it has exactly one point of access, and it is leaving.

Pointed atTheir reasoning, elicited conversationally and reconciled against what the system actually does.

A successor onboarded in weeks instead of a year — and a single-person dependency that stops being a board-level risk.

Every one of them, scoped the same way One system you name Six weeks POC at near-zero cost Your cloud Your reviewer signs

The join

The questions that need three experts and a shared calendar.

No single function holds the answer. Today that means the decision waits for whoever is free.

A supplier change

One question — can we do this, what does it oblige us to file, what does it cost. The knowledge sits in three different heads, and the answer waits for whichever two are free on Thursday.

QualityRegulatoryCommercial

A 483 response

The finding spans manufacturing logic, quality workflow and design history. Four teams spend several weeks reassembling something the organisation already knew.

ManufacturingQualityDesign

Day one after an acquisition

What does this platform do, what has it committed us to, and what breaks if we switch it off. Three questions, three functions, and nobody holds all three.

EngineeringLegalCommercial

The point is not that we know more than your experts. It is that your organisation stops depending on which two of them are free on Thursday.

What the gap costs before anyone calls it a problem

The knowledge is missing long before the programme admits it.

Discovery is a small share of a programme’s budget and a large share of its outcome. The published evidence is consistent on both halves.

21–78×

the cost of fixing a requirements error once it reaches integration and test, rather than at requirements

Haskins et al., INCOSE International Symposium, 2004 · NASA NTRS 20100036670

45%

average overrun on IT programmes above $15M — delivering 56% less value than predicted, across 5,400+ projects

McKinsey & the BT Centre for Major Programme Management, University of Oxford, 2012

The clock

FDA QMSR has been in force since 2 February 2026, incorporating ISO 13485:2016 by reference into 21 CFR Part 820. This is the first inspection cycle. Demonstrated system understanding is now something you get asked for — not something you plan for.

Does this apply to us?

What IRIS reads

If your estate is written in one of these, it is in range. Volume is not the limiting factor — a million lines is ordinary.

COBOLRPGPL/IABAPPL/SQL.NETC / C++JavaPythonAssemblerDevice firmware

And what it reads alongside the code

The written record is reconciled against the running system, so the two are compared rather than assumed.

SOPsDesign history filesValidation protocolsConfluenceSharePointFunctional & technical specsModification specs

And the part no tool reads: the people who still understand why it works that way.

What the knowledge is for

Knowledge is not the deliverable. It is the precondition.

A specification nobody acts on is a better-organised archive. The reason to capture what your systems and your people know is that none of the work you actually want is possible until it exists — and most of it becomes straightforward once it does.

The migration that stops stalling in discovery

Design begins from a documented as-is instead of a blank page and twelve analysts in a room — and the as-is stays current through the programme rather than ageing out in month four.

The submission that stops waiting on two engineers

Traceability, hazard analysis and design rationale written from the firmware that actually ships, confirmed by the people who wrote it.

The audit you can answer in days

An investigator’s question met with provenance rather than an internal archaeology project across four teams.

The automation that finally has something true to run on

Agents acting on your real rules, with a citation behind every decision and a boundary around what they are allowed to do unattended.

Transformation is not blocked by ambition, and it is rarely blocked by budget. It is blocked by the fact that nobody can fully explain the thing being transformed.

Where this goes

Retrieve. Author. Monitor. Act.

Validated knowledge is the only kind that can be automated against in a regulated environment — because the action inherits the defensibility of the knowledge beneath it. The rungs cannot be skipped.

01

Retrieve

Answers with citations, drawn from your code, your documents and your captured expertise.

Today

02

Author

Produces the artifact — specification, hazard analysis, fit-gap, as-is — for a named human to sign.

Today

Where this goes

03

Monitor

Watches the estate and flags drift: code changed, traceability broke, a rule diverged from the SOP that governs it.

Act on approval

Drafts the CAPA, opens the change request, computes the submission delta — and stops for a signature.

Act within bounds

Closed-loop execution inside a validated envelope, every action carrying its provenance.

Trajectory

The boundary

IRIS is deliberately bounded, and that boundary is what makes the intelligence defensible. The reviewer reviews rather than authors.

The spend, reframed

You are not buying knowledge. You are paying someone to re-derive what you already own.

It is in your source code and in your people, and it has been the whole time. Conventional discovery assembles a large integrator and your own scarce analysts into a room for weeks — and every hour spent there is an hour not spent running a live business. At the end you receive a document, and the understanding behind it leaves with the people who wrote it.

Every engagement you have ever run produced understanding that walked out of the building. This one leaves the asset behind, assigned to you.

What you are holding at week six

Not a tool your team has to learn. Artifacts they can sign.

A business specification of the system you namedReviewed and signed by your own analyst — who reviews rather than authors.
Traceability from requirement to design to test to riskGenerated from what the system actually does, not from what the file says it does.
A gap view against the standard that governs itQMSR, IEC 62304, ISO 14971, Annex 11 — as applicable.
Provenance on every statementThe code path or the named expert it came from.
The IP, assigned to you outrightAnd a named engineer’s signature on the deliverable.

Acceptance criterion

Your reviewer confirms 80%+ specification accuracy — or the pilot has not succeeded. It is written into the SOW, not implied by it.

In your calendar, not ours

You will know whether this works before your next board review.

What goes in the SOW

POC at near-zero cost — we underwrite it Production engagement under $100K, fixed fee Runs in your own cloud Nothing leaves your environment Zero data retention — your code trains no model A named engineer signs IP assigned to you outright

Procurement

Inveris Digital is listed on AWS Marketplace. You can buy through your existing AWS agreement and draw it down against commitment you have already made — no new-vendor onboarding, no separate procurement cycle. For most enterprises this is the difference between starting this quarter and starting next year.

When not to call us.

If your systems are documented, your specifications match what is running, and the people who understand them are not going anywhere — you do not need us. That is a real answer, and it is more common than the market pretends. If two of those three are untrue, the conversation is worth thirty minutes.

Start here

Which of these is yours?

Pick the one that sounds like your week. It opens a short form — nothing you would need approval to share — and puts the right conversation in front of you instead of making you hunt for an email address.

Swipe through them

Every one of these is a six-week question. Most people have been treating it as a six-month one.

The proof of concept costs you close to nothing, and it can be bought through AWS Marketplace against commitment you have already made — so the first step needs a decision, not a procurement cycle.

Or email us directly

You cannot transform what nobody can explain.