Survey

Find out what AI will amplify before you turn it on.

AI amplifies whatever is already there. Pointed at disciplined systems, it compounds them. Pointed at fifteen years of tribal knowledge and untested code, it ships defects faster than your team can catch them; the research puts it at 41% more bugs and an eightfold rise in duplicated code on undisciplined codebases.

A Survey tells you which one you're holding. Fixed fee, three weeks, and you get a map and a sequence, not a slide deck.

Who this is for

You run a manufacturer, broadcaster, distributor, or mid-market operation on software that's ten to twenty years old and still makes the money. Some combination of the following is true:

• The board is asking about AI and every answer so far has been a pilot that didn't ship.
• One or two people hold the system in their heads, and one of them is near retirement.
• You've bought copilot licenses. Nobody can say what changed.
• You suspect the honest answer to "can we bolt AI onto this?" is no, and you want to know what "yes" would cost.

If your systems are two years old and fully documented, you don't need us. Nobody's systems are.

What we examine

Four lenses, because the code is only a quarter of the picture:

1. Structure. Architecture, dependencies, and the seams. Where the system has clean joints and where it's welded shut. The specific question: which parts can be worked on safely, and which parts fight back.

2. Discipline. What's documented versus tribal. What's tested versus hoped-for. How decisions get recorded, how code gets reviewed, how anyone new would learn this system. This is the strongest predictor of whether AI helps or hurts, stronger than the code itself.

3. Seams and data. Can a machine operate this system, or only a person clicking through screens? APIs, contracts, data quality, integration points. Agents are already routing around interfaces; systems without machine-operable seams don't get bypassed. They get abandoned. Every system gets scored on one question: can an agent use it?

4. Workflow. Where work actually queues, waits, and gets re-keyed between systems. Half the time the highest-return finding isn't in the code at all; it's a swivel-chair process a person performs because two systems never learned to talk.

What you get

The Findings Register. Every finding logged, plate-numbered, and rated: amplifies or corrodes. No vague "areas of opportunity." Each entry names the system, the evidence, and the consequence of leaving it alone.

The Map. Your system inventory scored for AI-readiness across the four lenses. One page an executive can read; an appendix an engineer can act on.

The Sequence. A prioritized order of operations: what to retrofit, what to leave alone, what's ready for AI now, and what will waste money until its foundation is fixed. Each item sized and sequenced so the first project funds confidence in the second.

The Readout. A working session with your leadership, not a presentation. We walk the map, defend the sequence, and answer the hard questions in the room.

The deliverables are yours. Take the sequence to your internal team or another firm if you want. The map is honest either way. That's the point of fixed-fee: we have no incentive to find billable problems.

How it works

Week 1: Access and interviews. Read-only access to repositories and systems. Structured interviews with the people who actually run things, including the ones whose knowledge lives nowhere else.

Week 2: Inspection. Code, schemas, integration points, deployment reality versus documented intent. We use AI-assisted analysis where it's reliable and human judgment where it isn't; we know the difference because that boundary is the business we're in.

Week 3: Synthesis and readout. Register, map, sequence, working session. You end the month knowing exactly where you stand and what to do first.

What it costs

A fixed fee, quoted after a 30-minute scoping call, typically in the low five figures depending on system count and access complexity. No hourly meters, no change orders, no surprise line items. If we scope it wrong, that's our problem.

What happens after

Three honest outcomes:

1. You're readier than you thought. Some systems score well. We commission AI on those first while the sequence handles the rest. It happens more often than you'd expect.
2. The foundation needs work. We move to Retrofit, rebuilding the load-bearing pieces with the seams AI requires. The Survey fee credits toward it.
3. You take the map and run. Internal team, other vendor, next fiscal year. Fine. A company that knows its sequence is better off than one guessing, and you'll know where to find us.

Questions you're probably asking

"Our system is embarrassing. Do we need to clean it up first?" No, and please don't. We've spent twenty-five years inside the systems everyone else abandons. A tangled system that makes money is a success with deferred maintenance, not a failure. We need to see it as it is. The mess is the data.

"Will you just tell us to rewrite everything in your favorite stack?" No. Retrofit means the building stays up. The Survey's recommendations are stack-agnostic; most sequences preserve far more than they replace. When we do build, we build on boring, durable, operationally cheap foundations, and we'll show you the reasoning, not just the preference.

"We already rolled out copilots. Isn't this done?" Licenses are the easy 5%. Over 80% of organizations have deployed the tools; almost none capture the value, because the tools amplify the system underneath them. The Survey measures the part the license can't fix.

"Can our own engineers do this?" The inspection, maybe. The honesty, rarely. The people inside a system inherit its assumptions, and their findings arrive pre-negotiated with internal politics. A fixed-fee outsider's map has no constituency to protect.

Thirty minutes to scope it. Three weeks to know where you stand.