Why Delivery Improvement Stalls, and the Audit That Fixes It

Why Delivery Improvement Stalls, and the Audit That Fixes It

Here’s a pattern you have probably lived through. Every retrospective ends with action items. Every quarter opens with an improvement initiative — better testing this time, or faster releases, or “finally fixing the pipeline”. Everyone means it. And yet, a year later, delivery feels exactly the same: same firefights, same bottlenecks, same three people carrying everything. If that sounds familiar, you have a case of stalled delivery improvement — and the frustrating part is that effort was never the missing ingredient. Let’s diagnose it properly, because the cure turns out to be surprisingly specific.

The symptoms, written down honestly

  • Action items from retros get done — and nothing measurable changes.
  • Improvement work loses every priority fight against feature work by week three.
  • Nobody can say whether the team ships more safely than it did last year — opinions differ sharply depending on who had a bad on-call week.
  • The same category of incident keeps returning wearing a new hat.
  • Each new initiative starts from scratch, as if the previous ones never happened.

Notice what unites these: not laziness, not low skill. The missing thing is a stable reference point. Nothing connects January’s effort to June’s reality, so effort evaporates without leaving a record.

The usual misdiagnoses

“We need better tools.” Sometimes true, rarely first. Tools amplify a process; they cannot substitute for knowing which process is weakest. Teams that buy first and diagnose later end up with excellent instrumentation of an unexamined mess.

“We need more discipline.” This one blames people for a structural problem. Discipline decays precisely when nobody can see its payoff — and without measurement, the payoff of improvement work is invisible by construction.

“We need a transformation.” The most expensive misdiagnosis. A grand programme replaces the missing reference point with a slide deck, and eighteen months later you are diagnosing why the transformation stalled too.

The actual root cause

Stalled improvement is almost always a measurement problem wearing a motivation costume. Improvement requires three things effort alone cannot supply: an agreed picture of the current state, a way to tell whether a change moved that state, and memory between attempts. Retros supply none of the three — they capture last sprint’s irritations, which is weather, not climate. What supplies all three is an old mechanism from outside software: the periodic audit. Score the operation against a fixed set of questions, fix the worst findings, then score again on a known date. The second score is the part your current approach is missing — it is what turns effort into a recorded result and gives the next initiative somewhere to stand.

This is no longer an artisanal practice you have to invent. SCMGalaxy — a consulting and training firm whose SCMGalaxy OS product audits enterprise software delivery — has industrialised it: a hundred-question deep-dive across the lifecycle, scoring built to be repeatable so the follow-up sitting measures the team rather than the mood, surfaced gaps across delivery, governance and reliability, and recommendations ranked so you know what to fund first. Whether you adopt a platform like that or build a forty-question house version, the mechanism is the medicine.

Why this fixes each symptom

Action items that change nothing — because they were chosen by recency, not by severity. An audit ranks the gaps; you fund the two lowest scores rather than the loudest complaint. A team whose weakest area is, say, how releases are coordinated and rolled back will discover that no amount of test-coverage effort touches the thing actually hurting them.

Improvement losing priority fights — because unmeasured work is politically weightless. A written score of 1/5 with the re-audit booked turns “we should improve releases” into “this number must be a 3 by the June sitting”, which survives contact with a roadmap discussion.

Nobody agreeing on the trend — because there was no instrument. Fixed questions plus mechanical grading produce a number two sittings can share. That’s the whole trick, and it is why determinism matters more than sophistication: a clever score you cannot reproduce settles nothing.

Recurring incidents — because the category was never named. Audits grade categories, not events; a repeating incident shows up as a stubbornly low area score that refuses to rise until the systemic cause is funded. If the low area is the pipeline itself, something like a ground-up treatment of CI/CD for software delivery teams becomes the remediation plan for a specific scored gap instead of another generic initiative.

Initiatives with amnesia — because the audit sheet is the memory. Every cycle inherits the previous cycle’s scores, evidence citations, and unfinished items. Nothing starts from scratch again.

The prescription, dosed by month

Days 1–14: pick one value stream, adopt an instrument you did not write (or commit to writing one in a single week — perfect is the enemy here), and sit the first assessment. Days 15–30: do only the mechanical fixes among your two worst areas; book the re-sit for day 180 before you fix anything. Days 31–90: install the routines — the gates, checklists, and rehearsals the low scores demanded — and watch them survive a deadline crunch. Days 91–180: stop adding, let the routines bed in, and gather the evidence the re-sit will ask for. If an area needs sustained capability building rather than a quick fix — operations intelligence is a common one — slot an existing staged programme into the plan, the way a practical phased route into AIOps for DevOps and SRE teams slots under a weak operations score.

Then re-sit, compare per-area, and publish both sheets. One of two good things happens: the scores moved, and your improvement programme has its first provable win in years — or they didn’t, and you have learned in six months what the old approach would have hidden for three years.

Objections from the back of the room

“Our context changes too fast for fixed questions.” Your context changes; the fundamentals being graded — can you release safely, can you recover, does review actually happen — do not. Teams say this most often in the exact quarter they need the baseline most.

“We tried a maturity model once and it became a box-ticking game.” Then the scoring allowed testimony instead of evidence. A question that can be satisfied by a confident sentence will be; a question that demands a log, a date, or a rehearsal record cannot be gamed without doing the actual work — at which point the game and the improvement are the same thing.

“Leadership will weaponise a low score.” Pre-empt it: agree in writing that only deltas between sittings are reportable upward. It is a one-paragraph charter, and SCMGalaxy-style platforms implicitly assume it — progress tracking is the product; the baseline is just the anchor.

The sign you’re actually cured

It isn’t a high score. It’s the day a new initiative gets proposed and someone asks, reflexively, “which audit finding does this address, and what number should it move by the next sitting?” When that question feels normal, improvement has stopped being a mood in your team and become a loop — and loops, unlike moods, compound.