Architecture That Persuades: Still Running

The rationalization deck looks its best the day it clears the room. The slide that retires an application is the cleanest slide in the whole portfolio review: a box, a number under it, an arrow pointing to “decommission.” The savings are summed at the bottom where everyone can read them. And everyone can read a subtraction. It needs no new money, no new headcount, no new platform. You are taking something away. What feels safer than less.

So the room approves it. The number goes into a deck that goes up a level, and somewhere above the portfolio review a leader reports a savings figure to someone they answer to. The rationalization has now succeeded. Not the removal. The rationalization.

The trouble is the box. The slide draws the application as a box with a number under it, standing on its own, ready to lift out. Nothing that has run in an organization for any length of time is a box. It has grown wiring. A report someone pulls every Monday off a database no one ever mapped. An integration feeding a downstream system that nobody on the rationalization team has met. A handoff built around the system’s quirks until the quirks became the process. A single power user whose workaround quietly turned into how the work gets done. The box on the slide shows none of it, and the savings number assumes the box comes out clean. The blindness is built into how the inventory gets made. A rationalization captures what is deployed and what it costs to run, because that is what the tooling reports. It says nothing about what each system is wired to, because the connections between systems are nobody’s system of record. So the analysis comes out precise about the box and silent about the wiring, and the silence reads as cleanliness.

I have watched the cleanest slide in the review become the longest-running project on the roadmap more than once.

Removal is where the wiring shows up. You announce the sunset, set a date, send the note. Then someone goes to actually unplug the thing, and the box turns out to have been holding a dozen threads you could not see from the slide.

Each thread is small. None of them is the headline. The Monday report breaks, and it turns out a regional team was running their forecast off it. The downstream integration starts throwing errors, and the team that owns the downstream system did not know they were consuming from a system marked for death. The handoff stops working, and the workaround that was holding it together walks out with the one person who understood it. None of these is a crisis on its own. Together they are a project, and the slide funded a savings number, not a teardown.

None of this is rare enough to deserve its own war story. The everyday tax of fragile point-to-point integrations and the workarounds layered on top of them is what a survey of CIOs at large companies called the interest on technical debt, the routine cost that diverts ten to twenty percent of the budget meant for new work (McKinsey, 2020). For any system old enough to be worth retiring, that wiring is the standing condition of the estate, the thing every clean box on every slide is hiding.

And then there is the one thread that does not come loose. It is usually a single integration sitting between two teams that report to different leaders. The system marked for death is writing into something the second team depends on, and no one ever wrote down why. Cutting it is a risk the first team will not take, because they do not own what it feeds. Funding the work to untangle it is a cost the second team will not carry, because their system works fine and the problem arrived from someone else’s cleanup.

So it goes to a meeting. Everyone in the room agrees the old system should go. Everyone also agrees the dependency has to be understood before anyone touches it. Someone volunteers to scope it. The scoping needs real time from a person who is already fully booked on funded work, and decommissioning is not funded work. The action item is genuine and it does not move. Three months later the same integration is on the same risk list, and the same room has the same conversation. It was never anyone’s actual job to make it go away.

The pause becomes the resting state. Users move off the front end. The back end keeps feeding the report. The license keeps renewing on a quiet line item. It still has to be patched, because a system on the network is a risk whether or not anyone owns it, so a security team spends hours every quarter hardening a thing that was supposed to be gone, and eventually an audit finds it and someone writes a remediation plan for an application three slides ago marked retired. The system is half-decommissioned and fully alive: too dead to invest in, too wired-in to remove. Everyone agreed to kill it. It is still running, because the decision named a target and never named who does the cutting or what the cutting costs.

The savings rarely survive this part, and the reason is structural, not careless.

The savings were booked at approval. The portfolio review counted them the day the slide cleared, and the number went up a level into a deck where it became part of someone’s reported results. Savings get counted once, in public, by the person who proposed them. The cost lands the opposite way: slowly, in private, spread across the teams who inherited the threads, none of whom proposed anything. The two halves of the same decision are never weighed in the same room. The win is announced before the work starts, and the work, when it stalls, is not a story anyone has a reason to tell. Reopening the case means walking back a figure a leader has already reported. It is cheaper, politically, to let the zombie run.

It is also why the savings on the slide so rarely show up in next year’s budget. The money that was supposed to be freed is still going to keep the old thing alive: the license, the patching, the audit remediation, the slice of the run budget that never came down because the run never actually stopped. A rationalization that does not finish frees nothing, while adding one more half-dead system to the base it was supposed to shrink.

And the base only grows. Every rationalization that stops halfway leaves a system that is no longer used enough to defend and not gone enough to remove, and those accumulate. A year later the portfolio review faces a larger run cost than the last round of savings was supposed to leave behind, so the new targets have to be more aggressive to show the same number, which lays a fresh layer of half-dead systems under the last one. Nobody plans the sediment. It is just what is left when removals keep getting approved and never finished.

The rationalization that actually executes is the one that told the removal story before the room voted. It named the wiring instead of pretending the box was clean. It scoped the teardown as real work, with an owner and a line in the budget, instead of as a footnote to the savings. It showed three things, not one: what comes out, what it is wired to, and who does the cutting. In practice that is a second number on the slide, sitting next to the savings: the cost to remove, with a name attached to the work. A room that sees both numbers is approving a removal. A room that sees only the first is approving a hope. That version is harder to get approved, because it is honest about cost, and a room that has only ever been shown subtractions does not enjoy being shown the bill for them.

The honest objection is that some of those rationalizations will not get approved at all. A target that looked like free money with the teardown left off looks like a real project once the teardown is on the slide, and real projects compete for funding and sometimes lose. That is the trade worth making. A rationalization that loses because the room saw its true cost is cheaper than one that wins and then sits half-finished for three years, drawing budget the whole time. Fewer approvals and more actual removals is the better outcome, and a room that has been told the truth once will trust the next number faster.

That honesty is the whole persuasion move. A savings number is easy to present and easy to approve, and easy things rarely survive contact with the operation. The removal story is the part that decides whether anything is actually removed, and a rationalization that carries it is doing the architect’s real job: not proving the cut is worth it, but making the room see the cut all the way through to the line item that finally goes to zero.

A rationalization that shows only the savings is a slide. A rationalization that prices the teardown is a decision. Both can be perfectly rigorous. What separates them is whether the story includes the part where someone has to do the work.

So the cleanest slide in the portfolio review turns out to be the one that hides the most. The savings on it are real. What it leaves off is who does the taking-away, what the wiring will cost them, and whether anyone has agreed to carry it. A rationalization that cannot put those on the same slide has drawn a very tidy box around work that nobody has started.

Sources

McKinsey & Company, “Tech debt: Reclaiming tech equity,” 2020.