# Phased projects: is each phase a real decision?

> Splitting a project into phases only cuts risk when each phase leaves something in use and you can stop there. Three tests to run on your project plan.

- Author: Marlon Trettin
- Published: 2026-11-17 · Updated: 2026-11-17
- Language: en
- Canonical: https://yowpi.com/en/blog/phased-projects-is-each-phase-a-real-decision

---

Splitting a technology project into phases only reduces risk when each phase ends with something the operation already uses and the company is free not to buy the next one. Without those two conditions, a phased plan is a single bet paid in installments, and the board approves the whole project believing it approved only the start.

## TL;DR

- Phases named after activities (discovery, development, testing, rollout) divide the vendor's work. They do not divide the company's decision, because nothing is in use before the last one.
- The US federal acquisition rules describe the kind of increment that reduces risk: it delivers a working solution, does not oblige the buyer to purchase the next one, and has a time limit.
- Slicing a system that way is hard. In 2014, the US Government Accountability Office reviewed 89 IT investments and found only 23 that planned to deliver functionality every six months, although that was already required.
- Three questions test each stage of your plan: what is in use at the end, what the company will know by then, and what is lost if the project stops there.

## Why can a phased plan still be a single bet?

Because most plans divide the project by type of work. Discovery first, then design, development, testing and rollout. Each phase has a date and a price, and the set looks like a sequence of decisions.

Try stopping in the middle. At the end of the development phase, the company holds documents and code nobody uses. The money spent only turns into a result if the next phase is paid for. The question "do we continue?" has one possible answer, and a decision with one possible answer is a formality.

We have argued before that a funding request is stronger when it [reaches the board in phases, with a written condition for continuing](/en/blog/justify-technology-investment-to-the-board). This article is about the step before that. The condition only means something if the phase, on its own, leaves something standing.

## What makes a stage real?

A precise description sits in a public procurement rule. The US Federal Acquisition Regulation has a section on modular contracting for information technology, [FAR 39.103](https://www.acquisition.gov/far/39.103). It says modular contracting is intended to reduce program risk, and it describes what each increment needs.

Three points in the rule are useful to any company. Increments should provide for delivery, implementation and testing of workable systems or solutions in discrete parts. Contracts must be structured so that the government is not required to procure additional increments. And deliveries should, as far as practicable, be scheduled within 18 months of the solicitation being issued. In the FAR overhaul that began in 2025, the model text that agencies such as GSA already follow in place of the codified version keeps the first and third points.

Applied to your plan, and adding what the rule does not spell out, a real stage passes three tests. It ends with something in use. The contract lets you stop there, which in practice means no penalty and ownership of what has been delivered. And it is short enough for the next decision to happen while the assumptions behind the request still hold.

The rule was written for government purchasing, at a scale far larger than a mid-sized company. What carries over is the definition. The time frames belong to government.

## Why is this hard even when it is mandatory?

Because slicing a system into parts that work on their own is design work, and a schedule cannot do it for you. [Report GAO-14-361](https://www.gao.gov/products/gao-14-361), published in May 2014 by the US Government Accountability Office, shows how hard it is.

From 2012, the US government required major IT investments to deliver functionality at least every six months. GAO selected 89 investments at five agencies. Only 23 planned to deliver every six months, and 41 planned to deliver every twelve. In the report's words, almost three-quarters did not plan to deliver functionality every six months, and less than half planned twelve-month cycles.

The agencies' reasons deserve attention. Several said the rule was not feasible for every type of investment, and the Department of Defense answered that 12 to 18 month increments were better aligned with its acquisition framework. The same report lists, among the benefits of incremental development, the ability to terminate poorly performing investments with fewer sunk costs.

What follows is our reading. When an organization says its system does not fit into smaller deliveries, it is often describing less the system than the absence of a design that says which part works before the others. That design is an architecture decision, and it has to exist before the schedule does.

## How big should the first stage be?

As big as what it has to prove. The [software budgeting handbook published by 18F](https://guides.18f.org/derisking/state-field-guide/budgeting-tech/), a US government digital services team shut down in March 2025, was written for the people who approve funding. Some of its recommendations apply outside the public sector.

The first is about the size of the commitment. The handbook proposes that funding ramp up over several budget cycles, as the agency demonstrates that it is reducing risk. A request at high risk of failure gets little in the first year, and the amount grows as the project delivers.

The second is about time. Software has to reach end users no more than six months after the contract is awarded, and keep reaching them from then on. The third is about reporting. Those providing oversight should accept demonstrations of functioning software as their reports, and the handbook states that the only measure of success that matters is the value delivered to end users.

The dollar figures in the handbook are government figures and are no reference for your company. The criterion is. The first stage should be the smallest part of the system that someone in the operation uses with real data, and it should carry the project's biggest unknown. If the unknown is the integration with the ERP, the integration goes into the first stage, even if only for one type of record.

## An example: the same project sliced two ways

The scenario below is hypothetical, and the time frames are there only to illustrate the reasoning. An industrial maintenance company with 120 employees tracks work orders in spreadsheets and wants a system connected to invoicing.

In the first proposal, the plan has three phases: six weeks of discovery and design, sixteen of development, six of testing and rollout. The board approves phase 1 and books a decision meeting for the end of it. At that meeting, what exists is a document. At week 22, what exists is a system no technician has opened. At neither point does stopping preserve anything.

In the second proposal, stage 1 takes eight weeks and delivers opening and closing work orders for a single field team, with real data and invoicing still handled by the current process. Stage 2 connects work orders to invoicing. Stage 3 takes the system to the other teams and adds the management dashboard.

At the end of stage 1, one team works in the system and the board decides by looking at usage. If the company stops there, it keeps work order control for one team, which is little and still works. The second proposal has a cost that should be stated: running the old process and the new one side by side for a while takes effort, and the total may come out higher than the first. In exchange, the amount at risk at each decision drops to the size of one stage.

A real case shows the same order. At [UniTrust](/en/cases/unitrust), a US insurance IMO, the first system went live in two weeks in 2022, and the current platform was built on that base over the following years.

## What should you take to the board?

A table with one row per stage and three columns. What is in use at the end. What the company will know that it does not know today. What is lost if the project stops there. A stage where the first column is empty and the third says "everything" is an installment, and should be presented as one.

That table connects to two arguments the request should already contain. The [cost range in place of a single number](/en/blog/tech-project-estimates-give-the-board-a-range) gets narrower with each completed stage, because the estimate for the next one uses data from the last. And the [shutdown criterion agreed before starting](/en/blog/when-to-shut-down-an-innovation-project) gains a place to be applied, since each stage ends on a date when stopping is a real option.

## Frequently asked questions

**Doesn't splitting the project into stages that work on their own make it more expensive?**

It can raise the total, yes. Independent stages need temporary bridges between the old process and the new one, and part of that work is thrown away later. What the board compares is something else: the amount committed before seeing any result. In a plan organized by activity that amount is the whole project, and in a plan organized by stages it is the cost of the first one.

**Don't short stages take too much of my team's time?**

They take time earlier, and more predictably. Instead of one concentrated acceptance effort at the end, the team uses and assesses a small part at each stage. The total hours depend on the project. What changes is when problems show up: earlier, when fixing them tends to cost less.

**What if the first stage does not remove the bottleneck?**

Then the company found that out by spending one stage, not the whole project. That is the stage's job. If the first stage was designed to carry the biggest unknown, a poor result is information for redesigning the next one or for stopping, with what was delivered still in use.

## The next step

If the plan you intend to take to the board has phases, but none of them leaves anything in use before the last, the Operational Architecture Diagnostic is a 30-minute conversation to find the first stage that works on its own. [Talk to an architect](/en/contact).
