Industries / Construction and property / Commercial and residential construction / ConTech
Scheduling
Scheduling software lets a builder list every task on a job, set the order the tasks must follow and how long each takes, work out the finish date, and then track what really happens on site against that plan.
6
Milestones achieved, from 1959 to 2019.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1959
Scheduling sold as a service
John Mauchly left Sperry Rand in 1959 and founded Mauchly Associates, Inc. in Ambler, Pennsylvania. That December, James Kelley and Morgan Walker presented their paper on the critical path method, the calculation of the longest chain of linked tasks that sets a job's finish date, at the Eastern Joint Computer Conference under the firm's name. The firm made the method something a customer could buy: by the account of the scheduling historian Patrick Weaver, it "commercialised CPM", ran training courses and sold consulting, and Mauchly's papers at Penn record "proprietary computer-based programs". Kelley and Walker's first public presentation was a five-day workshop in Philadelphia on 16 to 20 November 1959, by Weaver's count. Weaver names Catalytic Construction of Philadelphia as "the first large construction company" to adopt the method, in 1961.
Credit
John Mauchly, James E. Kelley Jr. and Morgan R. Walker, Mauchly Associates
1964
IBM ships precedence scheduling
In early 1964 IBM announced its Project Control System (PCS) for the IBM 1440 computer. PCS came out of a March 1963 joint venture with H.B. Zachry Company, a San Antonio builder that had tried critical path scheduling since 1959, at first on rented IBM computer time. PCS used precedence diagramming, in which each task is a box and lines between boxes show which must come first. By Weaver's account it was the first computer scheduling system to offer lags, set waiting times between linked tasks that let work overlap.
LessonThe method built into the most widely used software became the industry's method; Weaver reports that by 2004 precedence was "virtually the only" method in general commercial use.
Credit
IBM and H.B. Zachry Company, whose research director was W.D. Tiner Jr
1980
Scheduling moves to the desk
Micro Planning Services, a UK firm, released Micro Planner for the Apple II in 1980. Weaver calls it the first commercial scheduling software for that class of computer. Through the late 1980s, dozens of desktop scheduling programs followed, Microsoft Project among them, and by Weaver's account Microsoft Project sold for under US$100 a copy for many years. Building a schedule no longer needed a mainframe or minicomputer.
LessonCheap tools spread faster than skill; Weaver reports that many people learned scheduling from the software alone, "There was no training or oversight", and the typical schedule filled up with shortcuts the tools allowed.
Credit
Micro Planning Services; Microsoft
1998
The schedule gets a picture
At Stanford's Center for Integrated Facility Engineering (CIFE), Bonsang Koo and Martin Fischer linked each part of a 3D model of an office building to the scheduled task that builds it, so the model could be played forward in time. This is called a 4D model. Their 1998 report found it helped teams spot problems before work began and communicate across the project, but the models were hard to build and analyse. Firms carried the idea into products. London-based Synchro sold 4D software used on Crossrail and by BAM, Skanska and Larsen & Toubro, and Bentley Systems bought Synchro in June 2018.
Credit
Bonsang Koo and Martin Fischer, Stanford CIFE; Synchro Software; Bentley Systems
2015
Schedules get an audit rulebook
The US Government Accountability Office (GAO), the auditor that reports to Congress, issued its Schedule Assessment Guide on 22 December 2015. It sets out 10 best practices that GAO uses to judge agency schedules in its audits. The guide sits beside the US Defense Contract Management Agency's 14-Point Assessment, a set of counts on a schedule file; one says no more than 5% of unfinished tasks should lack a link to a task before or after them. Software now runs these counts. Oracle's Primavera P6 Professional has a Check Schedule tool that runs most of the 14 points, and a free browser tool runs all 14 on P6 files.
LessonAutomatic checks need checking too; consultants at Ten Six found P6's tool reported a baseline execution index (tasks finished divided by tasks due to be finished by now) of 1.0 where the right value was 0.55.
Credit
US GAO; US Defense Contract Management Agency; Oracle
2019
A schedule built to push
On 3 May 2019 the UK National Audit Office (NAO) reported on Crossrail, the new east-west railway under London. Crossrail Ltd had steered toward a December 2018 opening with a "Master Operational Handover Schedule". The NAO found this was "a set of more or less separate plans" for each part of the work, not a detailed plan linking all the work that remained. Crossrail Ltd reset it in 2016, 2017 and 2018 with harder milestones, and progress fell behind each new version soon after. Its programme director told sponsors in February 2018 that the schedule was "intended to drive behaviour", and that the contractors' own programmes, added together, would end in 2020. The NAO named "the compressed schedule" and "the absence of a realistic plan" among the causes of unnecessary cost.
LessonA target schedule used to press contractors is not a forecast, and a project needs a linked, realistic plan of the remaining work alongside it.
Credit
reported by the UK National Audit Office; the schedule was Crossrail Ltd's
Watch for
AI drafts the first schedule
nPlan, a London company, says its Schedule Studio turns scope documents into "a logic-linked first draft" with task breakdown, durations and labour hours, up to 8,000 tasks, ready to export to Primavera P6 or Microsoft Project. It says the model was trained on "780,000+ real construction schedules". Its customer quotes report a tender schedule done in 40 minutes that took 2 days by hand. The page gives no accuracy figure.
Achieved whenan independent group gives the same scopes to Schedule Studio and to experienced planners, and publishes how many logic errors each schedule has and how much time was saved.
Claimed by
nPlan; the tool came out of beta by June 2025, per Project Controls News
Forecasts that save money
nPlan says its probabilistic forecasts of how a schedule will turn out, learned from a dataset of "more than 750,000 projects", have "saved its customers well over $1.2B". It names HS2, Network Rail, Shell and MTR as customers. The release gives no accuracy figures and does not say how the savings were counted.
Achieved whena customer or auditor publishes nPlan's forecast finish dates next to the actual finish dates for a set of projects, and shows how the savings were measured.
Claimed by
nPlan, in its Series B funding release, October 2025
Open
01
Judge each schedule link
A reviewer that reads each task's name and its links in a schedule file and flags links that make no building sense, such as carpet laid before the doors it should follow. It would use a language model to judge each link and plain graph checks to trace what each bad link pushes or hides on the critical path. It would output a ranked list of suspect links, each with a reason.
The briefCredit · unclaimed02
Test AI-drafted schedules
A public test set of 20 building jobs. Each has a one-paragraph written scope and the real schedule used on the job. A scoring script compares any AI-drafted schedule with the real one. Run at least three drafting tools on it and publish the scores.
The briefCredit · unclaimed
Open 01 · the brief
Judge each schedule link
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A reviewer that reads each task's name and its links in a schedule file and flags links that make no building sense, such as carpet laid before the doors it should follow. It would use a language model to judge each link and plain graph checks to trace what each bad link pushes or hides on the critical path. It would output a ranked list of suspect links, each with a reason.
Why it is still open
The checks built into today's software count structural faults, such as missing links, lags and fixed dates. They do not judge whether a link is right. A free 14-point checker warns that "A passing score on all 14 checks does not guarantee schedule viability". GAO calls any "incorrect logic relationship" potentially damaging to the whole schedule, and in its worked example a wrong link, between hanging interior doors and fitting trim, is caught by review, not by a count. A search in November 2026 found no published accuracy figures for a tool that judges whether links make sense.
Built with what exists
MPXJ, a free library that reads Primavera P6 (XER), Microsoft Project and Asta Powerproject files; a current language model such as Claude Fable 5.2 (see the Note "Three working devices"); the Python graph library NetworkX; and real project schedules from DSLIB, a public set of 231 real projects kept by Ghent University's Operations Research and Scheduling group. The advance that makes this possible is language models that know the usual order of building trades from task names alone.
First step
Download DSLIB this month, pick 10 projects whose task names are readable, and write the script that plants wrong links and keeps the answer key.
Done looks like
10 real schedules, with 100 wrong links planted by a third party who keeps the answer key. The tool finds at least 80 of the 100 and flags no more than 20 of the original links, with two independent schedulers settling disputed flags. A standard 14-point check on the same files is run for comparison, and the full tally is published.
Open 02 · the brief
Test AI-drafted schedules
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test set of 20 building jobs. Each has a one-paragraph written scope and the real schedule used on the job. A scoring script compares any AI-drafted schedule with the real one. Run at least three drafting tools on it and publish the scores.
Why it is still open
CONSTRUCTA, a February 2025 study of AI-drafted construction schedules, ran its "Experiments on proprietary data", so no one else can rerun it. nPlan's Schedule Studio page reports customer time savings but no accuracy figures.
Built with what exists
DSLIB's real project schedules as the reference answers; MPXJ to read and write XER and Microsoft Project XML files; language-model drafting tools; Python for scoring; and repeated runs of each tool on the same scope to measure how much its answers change, the kind of consistency test Microsoft describes for its ThinkingBox system (see the Note "All twenty times"). The advance is that general language models can now write a full task list with links and durations from a plain-language scope.
First step
Choose 20 DSLIB projects with complete baseline schedules and write a short scope for each, giving building type, size and main systems, without copying the task list.
Done looks like
20 jobs, at least 3 drafting tools, each run 20 times. A published table gives the share of the real schedule's tasks covered, the share of its links reproduced, the median duration error in working days, the 14-point counts, and run-to-run agreement. The scoring code is public so anyone can rescore.