The challenge
Process managers at Heijmans drew up Project Management Plans entirely by hand. Each plan took hours to days of work: gathering information from earlier projects, structuring it along the fixed template and writing it out in the right tone and house style. And since a PMP is typically written by a team of about five people, coordination and version control came on top.
It was repetitive work that still had to be tailored every time. Boilerplate was rewritten again and again, quality depended on whoever happened to be writing, and the knowledge from earlier plans sat scattered across folders and heads. The time spent structuring and writing came at the expense of where process managers make the difference: the substance and strategy of their projects.
The ambition was sharply defined: automate around 80% of the manual writing work, at equal or better document quality. But on the terms of a listed construction group. Everything had to run in Heijmans' own Azure tenant, behind their own access policy, with encryption, audit logging and the source code in their own GitHub organisation. No data leaving the building and no black box.
Our approach
- 01
Generating from their own knowledge
The suite generates a plan section by section, drawing on Heijmans' official templates and a knowledge base of historical plans and writing guidelines. Every section shows which sources were used, so the origin of the text stays traceable.
- 02
Human in the loop, team in the editor
AI drafts, people decide. The whole writing team works in the same editor at once, every version stays traceable and every final version is validated by the project manager. Once the plan is done, the suite exports it to Heijmans' official Word template.
- 03
Entirely within Heijmans' walls
The suite runs in Heijmans' Azure tenant, with sign-in through Entra ID and mandatory multi-factor authentication. Every user action is logged and the source code lives in Heijmans' own GitHub organisation. Bonsai builds, Heijmans owns.
- 04
From pilot to platform
Version 1 stood in four months, in four phases each closed with an explicit go/no-go. After the successful pilot came a second engagement: workspaces per document family, roles and permissions through Entra ID groups, template versioning with a full audit trail and an infrastructure ready for thousands of documents a year.
The result
During the pilot around 80% of the manual writing work was automated, saving 50 to 80 percent of the time per plan. Scaled up to more than 1000 documents a year, that per-document saving adds up substantially. Hours that used to go into structuring and writing go back into the substance of the project. The writer remains final editor of every plan that goes out the door.
What started as a PMP generator is now a writing platform on which Heijmans sets up new document families itself, each in its own workspace with its own templates, permissions and versioning. And perhaps the most important result: trust. Every phase of both engagements closed with an explicit go, and Bonsai is now building the next AI applications within Heijmans along the same lines.