Delivery management, run as a business: the operator's guide

Most definitions stop at "ensuring projects are delivered on time and within budget". That describes a project manager. Delivery management is broader and more commercial. It owns the whole portfolio of engagements a business has promised to its clients, the people and cost required to honour those promises, and the margin left when the work is done. The unit of analysis is not the project. It is the practice: revenue in, cost out, quality and client health in between.

In a services company, the delivery leader is accountable for five numbers every month: revenue recognised, gross margin, utilisation, forecast accuracy and client health (renewals, escalations, NPS or its equivalent). If a delivery leader cannot state those five numbers for last month from memory, they are running projects, not delivery.

Project management is the execution of one scope, one timeline, one budget. Program management coordinates several related projects toward a shared business outcome — a replatform, a market launch — and manages the dependencies between them. Delivery management sits above both: it owns every program and project the business is running, decides which people work on what, and answers for the commercial result of the whole.

A useful test: a project manager is measured on their project, a program manager on the outcome across projects, and a delivery manager on the practice's P&L. The skills overlap; the accountability does not.

Revenue recognised. What was actually earned this month, by engagement, against what was forecast at the start of the month. The gap is your forecasting problem.

Gross margin. Revenue minus the fully loaded cost of the people who delivered it. In digital commerce services, a healthy practice runs 40–55% gross margin; below 30% the business is subsidising the client.

Utilisation. Billable hours as a share of available hours. Target 75–85% for delivery staff; higher means burnout and no time for pre-sales, lower means idle cost.

Forecast accuracy. How close the month-start revenue and margin forecast came to the actual. Within ±5% is disciplined; ±15% means surprises for the CFO every month.

Client health. Renewals on time, escalations count, payment delays. This is the leading indicator for next quarter's revenue.

I review these five weekly in one page, and I own the number. When one moves, I know why before anyone asks.

The practice I run has 33 people organised into pods, each pod a self-contained unit with a lead, engineers, QA and a shared architecture and DevOps layer. Pods own a portfolio of clients so knowledge compounds; a client is never a stranger to the team that serves them. Above the pods sits a small PMO that runs forecasting, staffing, reporting and commercial hygiene (SOWs, change requests, invoicing). The delivery leader owns the whole and the P&L.

The advantage of pods over a flat pool is commercial: margin is visible per pod, staffing decisions are local and fast, and pod leads grow into practice leaders because they see the number.

Concurrency is the real difficulty of delivery management. Any one engagement is manageable; eighty of them, each with its own client, contract and risk, is where practices break. What works:

The practice has shipped 15+ global go-lives, including three back-to-back Adobe Commerce production upgrades that protected live trading throughout. That is not luck and it is not only engineering. It is a management outcome: a go-live runbook that is rehearsed twice, a rollback path that is tested rather than assumed, a hypercare rota that is staffed before the launch date, and a client who has signed off on the acceptance criteria in writing. The delivery leader's job is to refuse a go-live date that does not have those four things.

Put AI where a wrong answer costs a forecast miss, not a customer. In delivery operations that means utilisation and revenue forecasting, staffing recommendations, risk flagging from ticket and timesheet patterns, and drafting status reports for human review. It does not mean letting a model talk to a client or commit to a date. In client programmes, the same rule applies: the AI supply-chain suite we delivered for a global CPG brand improved forecast accuracy by 22% and cut stock-outs 18% because it advised planners; it never placed an order on its own.

Scope that grows without a change request. Estimates made by sales alone. Utilisation reported without bench cost. Hypercare that becomes free support for months. Senior people doing junior work because the bench is thin. Status reports that describe activity instead of outcome. Each is small; together they take a 45% practice to 25%.

Someone who can read a P&L and a stack trace in the same afternoon. Who asks "what is the margin on this" before "when is it due". Who has shipped something to production and has been on the call when it broke. Who writes clearly enough that a client and a CFO both understand the same status note. Certifications are useful signals; owning a number is the proof.