An Australian digital agency operating across multiple locations reversed from -20% to -60% quarterly losses into +40% to +60% quarterly profit — a 100-point swing — through a sixteen-week delivery governance redesign. Throughput improved 45% on the same team serving the same clients.
The starting state: the agency had talent, had demand, and was still hemorrhaging money on every project it delivered. The sales team was closing work. The delivery side was losing money on most of what was closed. The gap was structural, not commercial.
The challenge: diagnose the systemic failures and redesign operations to achieve profitability without cutting the team, without turning away work, and without disrupting the client relationships that were keeping the agency alive.
Starting Conditions
The agency operated across multiple Australian offices, coordinating development work with partner agencies for larger client engagements. Revenue was steady — the sales team consistently closed work. The problem was that every project cost more to deliver than the revenue it generated. The P&L arrived monthly and described damage that had already happened.
Scope creep was endemic. Not because clients were unreasonable, but because no structural mechanism existed to identify scope changes early. By the time overruns were visible in the monthly project accounting, recovery was impossible — the project was already underwater.
Multi-agency coordination was the primary cost driver. Development operations spanned partner agencies. Each partner had its own processes, tools, and quality standards. Integration points were where projects went to die — handoffs lost context, timelines slipped, accountability dissolved across organizational boundaries because no single party owned the seam.
Utilization was invisible. The agency had no reliable visibility into actual utilization rates. Resources were allocated based on gut feel rather than data. Some teams were overloaded while others were underused. The mismatch did not surface until the monthly P&L exposed the damage, at which point the quarter was already in motion.
What had been tried: the agency had invested in project management tooling — Jira, Confluence, time-tracking software. The tools existed and were populated with data. Nobody had designed the governance layer that connected the data to decisions. A project manager could see that a project was at 80% of budget with 50% of work complete, but no structural mechanism existed to trigger intervention at that point — no gate, no escalation, no automatic reallocation. The data was available. The governance to act on it was not. The leadership team's own diagnosis was that they needed more senior delivery managers or different clients. Neither diagnosis would have changed the outcome, because both left the structural layer untouched.
Structural Diagnosis
The losses were not caused by a single failure. They were the product of three compounding structural issues operating simultaneously — each of which was defensible in isolation and devastating in combination.
No delivery governance. Projects ran without standardized estimation, without consistent progress tracking, and without defined quality gates. Each project manager had their own approach to estimation, their own tolerance for scope creep, and their own private definition of "on track." Consistency was a function of which project manager you got, not which process the agency followed. The absence of governance meant scope creep was detected through accounting, not through project controls. By the time the CFO flagged an overrun, the project team had been working at a loss for weeks — sometimes longer — and the feedback loop was too slow to enable any correction that did not involve absorbing the damage. This persisted because the agency had grown by hiring experienced project managers and trusting their judgment. That works at ten people. It does not work at multi-office scale, because judgment varies, and unbounded variance in estimation is indistinguishable from randomness in the P&L. Conventional fixes — mandating a new PM methodology, running more training — do not hold, because they add activity without changing the structure. The managers were not lacking effort. They were operating inside a system that did not connect their activity to consequence.
Multi-agency coordination failure. When the agency partnered with external development shops, the integration points became black boxes. Work was handed off with a specification. What came back bore a variable relationship to that specification. Code quality, testing standards, and documentation practices differed across agencies. The receiving team spent 20-30% of their time on integration work that was not scoped, not budgeted, and not visible in the project plan. The coordination overhead was measured post-hoc during the engagement: 45% of total effort across multi-agency projects was consumed by coordination — meetings, status updates, re-explaining context across agency boundaries, debugging integration issues, reconciling different testing approaches. Nearly half the budget was being spent on making the agencies work together rather than on building the product. This persisted because coordination work is distributed across dozens of small interactions, each of which feels individually reasonable. No single meeting is obviously wasted. Only the aggregate tells the story. Conventional fixes — appointing a coordination lead, tightening the partner contract — do not touch the structure, because they add a person to the same broken seam.
Utilization blindness. Without reliable visibility into actual utilization rates, the agency was making staffing decisions based on feel. Senior developers were being allocated to maintenance tasks while junior developers struggled with complex architecture work — not out of deliberate misallocation, but because no data existed to inform better decisions. The utilization imbalance meant the agency was paying senior rates for junior-appropriate work and paying for rework when junior developers were assigned work beyond their capability. This persisted because the time-tracking tool was already in place and everyone assumed the utilization data was being used. Nobody had built the reporting layer that turned that data into a staffing decision input, and without that layer, the data was decorative. Conventional fixes — hiring a resourcing manager, running more detailed time-tracking — do not change the outcome, because the data was never the bottleneck. The connection between data and decision was.
The Intervention
The turnaround ran sixteen weeks across three phases. Each phase depended on the previous phase's output, and each produced measurable operational improvement before the next began.
Phase 1: Visibility (Weeks 1-4)
What was built: Instrumentation across every delivery stream. Actual hours versus estimates at the task level, not the project level. Scope change velocity — how fast requirements were changing and in which direction. Handoff latency — how long work sat at integration points between agencies. Utilization rates by role, by team, by project.
Why this came first: You cannot redesign a delivery system against a narrative. You have to redesign it against data. Every previous attempt to fix the agency had started with a theory — the clients are difficult, the partners are unreliable, the PMs are inconsistent — and implemented a solution to that theory. Without data, every theory was equally defensible. Phase 1 was about producing the evidence that would make the structural diagnosis non-negotiable, and the reform in Phase 2 implementable without political objection.
The mechanism: The instrumentation was not new tooling. The tools — Jira, Confluence, time-tracking — already existed. The intervention was designing the data flows: which metrics needed to be captured, at what granularity, with what frequency, and connected to which decision points. The work was analytical, not technical. The outputs were dashboards and reports that described a reality the organization had been living inside but never seen clearly.
First-phase outcome: The data exposed what intuition had suspected but could not prove. The 45% coordination overhead figure was the key finding — it meant that for every AUD 100 spent on multi-agency projects, AUD 45 was consumed by agencies learning to work with each other rather than by productive development. This single metric justified the Phase 2 investment on its own. Several additional findings — the integration rework burden, the utilization imbalance between senior and junior developers — came out of the same instrumentation pass and shaped the rest of the reform.
Phase 2: Structural Reform (Weeks 5-10)
What was built: A redesigned delivery framework with three structural components.
First, standardized estimation models that replaced individual PM judgment with calibrated estimation anchored in historical delivery data. The models were not rigid — they included uncertainty ranges scaled to project type and agency combination. What they replaced was the wild variance of individual estimation with bounded variance anchored in how the agency actually delivered, not how individual PMs hoped it would deliver.
Second, quality gates at integration points. Before work crossed an agency boundary, it had to pass a defined check: code review, test coverage threshold, documentation completeness. The gates added time at integration points — approximately two to four hours per handoff. They eliminated the 20-30% rework that had previously consumed the receiving team's capacity.
Third, a shared project intelligence layer — a unified dashboard that all partner agencies could access, showing project status, integration queue, and quality metrics in real time. This replaced the status meetings that had previously consumed eight to twelve hours per week across the partnership.
Why this phase depended on Phase 1: Every component of the reform was a structural response to a specific finding in the Phase 1 data. Without the data, the reform would have been a generic best-practice program, and the partner agencies and internal PMs would have resisted it on the reasonable grounds that generic best practices rarely fit specific contexts. With the data, the reform was visibly an answer to a visibly measured problem. The political resistance dropped because the argument had already been won empirically.
The mechanism: The structural reform did not add new activities to the process. It replaced high-cost coordination activities (meetings, email threads, verbal status updates) with structural mechanisms (gates, dashboards, automated alerts) that accomplished the same coordination at a fraction of the human cost. The gates were the critical piece — a procedure can be bypassed, but a gate that blocks the next step of work cannot be bypassed without making the bypass visible.
Tradeoff introduced: The quality gates cost time at integration boundaries. Projects that had previously thrown work across the seam without friction now had to pause for two to four hours while the gate ran. Some PMs initially read this as added overhead. The rework reduction was larger than the gate cost, but it was lagging — the saving showed up in the next cycle, not the current one — and the behavior change required the leadership team to hold the line while the benefit materialized.
Phase 3: Automation and Governance (Weeks 11-16)
What was built: Automated reporting that replaced manual status compilation. Resource allocation dashboards that made utilization visible and actionable in real time. Early warning systems that triggered at defined thresholds — 70% budget consumed with less than 50% completion, scope velocity exceeding baseline by more than 20%, utilization below 60% for any team for more than two consecutive weeks. And lightweight governance cadences. Not more meetings — the reform had already eliminated most of the coordination meetings. Instead, decision protocols: what happens when an early warning fires, who has the authority to reallocate resources, what constitutes grounds for scope renegotiation with the client.
Why this phase came last: Automation without structure automates the wrong behavior. Early warning systems without decision protocols produce alerts that nobody is authorized to act on, which creates alert fatigue and then cynicism. Building the automation before the Phase 2 structural reform was in place would have produced a high-fidelity view of a system nobody had the authority to change. Phase 3 only works after Phase 2.
The mechanism: Each early warning threshold was wired to a specific, named decision-maker and a specific, defined action. The automation did not replace judgment — it forced judgment to be exercised on time rather than after the fact. This is the difference between governance as documentation and governance as structure: the documented version describes what should happen when a project overruns, and the structural version makes sure somebody is required to do something before the overrun gets worse.
Constraint and tradeoff: The governance layer required executive sponsorship to stay honest. The threshold definitions, the named decision-makers, and the defined actions all degraded quietly if leadership stopped using them. The framework was not self-maintaining. This was named explicitly in the handoff — a system like this survives only as long as the people at the top treat the early warnings as binding rather than advisory.
Results
Profitability: The agency reversed from -20% to -60% quarterly losses into +40% to +60% quarterly profit — a 100-point swing. The reversal was driven primarily by two mechanisms: the elimination of coordination overhead as a cost driver (the 45% figure dropped substantially once the shared intelligence layer replaced status meetings), and the prevention of late-detected scope creep (the early warning system fired well before the point at which overruns had historically become unrecoverable).
Throughput: 45% improvement in project throughput. The same team, serving the same clients, delivered more work — not because they worked harder, but because they spent less time on coordination and rework. The throughput gain came from eliminating waste, not from increasing effort. This is the signature of a structural intervention: the capacity was always there, trapped inside the friction.
Multi-agency efficiency: The multi-agency delivery model, which had been the primary source of loss, became a competitive advantage once governed properly. Quality gates at integration points meant that partner agency output was reliable. The shared intelligence layer meant that coordination happened through data rather than meetings. The agency could take on larger, more complex engagements that required multi-agency collaboration — and deliver them profitably — in a market where most of its peers could not.
Structural sustainability: The framework survived the phase of the engagement where consultant attention withdraws and organizational habit tries to reassert itself. It survived because the gates could not be bypassed without making the bypass visible, because the shared dashboard was faster than the old status meetings and so nobody wanted to go back, and because the governance cadence was lightweight enough to keep running under a successor operating rhythm.
Counterfactual: Without the intervention, the agency was on a trajectory that had two visible endpoints, neither of them good. The first was cost-cutting — reducing headcount to match the revenue the system was actually producing, which would have meant turning away work in a market where work was the one thing the agency had in abundance. The second was a partner exit — walking away from the multi-agency model, which would have cut the coordination overhead but also cut the access to large engagements that the model enabled. Both endpoints would have left a smaller, less capable agency. The turnaround preserved the model and fixed the structure, which is a materially different outcome from shrinking into profitability.
The Diagnostic Pattern
The agency did not have a talent problem. It did not have a demand problem. It did not have a client problem. It had a governance problem. The same people, serving the same clients, became profitable when the delivery architecture changed. This is the diagnostic signature of a structural problem: changing the people does not change the outcome. Changing the structure does.
Three questions diagnose the structural constraint in any delivery organization.
Where are decisions being made without data? In this agency, resource allocation and scope management were judgment-based, not data-informed. The data existed inside the tools. No governance connected the data to the decisions, and so the decisions were effectively guesses dressed as experience.
Where is coordination overhead consuming productive capacity? The 45% coordination overhead was the single largest cost driver — larger than any technology problem, any talent gap, any client management issue. Reducing coordination overhead through structural mechanisms (gates, dashboards, protocols) had more impact than any other intervention. Coordination cost is invisible in a line-item budget, because it is distributed across dozens of small interactions, which is why it almost always exceeds what leadership thinks it is.
Where is rework happening because quality was not verified at boundaries? Integration rework consumed 20-30% of receiving-team capacity. Quality gates at handoff points eliminated this rework by catching issues before they crossed boundaries. The gates cost two to four hours per handoff. They saved twenty to thirty hours of rework per integration cycle. That arithmetic transfers: rework downstream is almost always cheaper to prevent upstream, and the only question is whether you are willing to slow down at the handoff to accelerate everything that comes after.
The pattern transfers across delivery organizations. The specific gates and dashboards differ. The structural diagnosis is the same.
Related Service
This engagement falls under my Digital Transformation practice.
View advisory engagement models