5 Problems Precon Software Can Solve That It Couldn't in 2021
Updated: 5 days ago

Preconstruction is mostly reading, measuring and comparing, and those are the jobs software has gotten noticeably better at since 2021. Here are five problems that were manual five years ago and now have digital solutions in precon software. The first four are in commercial use. The fifth, linking field productivity to the schedule, is still emerging, and it is marked that way below.
Catching Changes Between Drawing Sets
A late addendum used to mean flipping through sheets side by side, or running a PDF overlay and hoping the design team had clouded every change. An overlay only shows what moved on the page, and it depends on the revision clouds being complete. Newer tools read the drawing content itself. Nomic says its comparison surfaces changed notes, details, dimensions and references even where no cloud was drawn. Datagrid, now part of Procore, offers an agent that compares a baseline set with a revised one and flags changes that affect scope, cost, schedule, or constructability.
For an estimator, the gain is a short list of what actually moved, so re-pricing can start with those items instead of a full re-read. One caution: this rests on vendor documentation, and
in could find no independent study of time.
Searching Your Own Project History
Past bid tabs, awards and closeout records used to sit in PDFs, shared folders, and personal spreadsheets. A question like "what did we pay for this scope on the last two similar jobs?" meant digging through archives. Plain-language search across project data now answers it in minutes. Procore's agent library includes a Fast Search agent that returns answers from connected Procore data, and a Nomic comparison describes the platform as able to query past RFI patterns, change-order history and subcontractor performance.
This helps most when you need a benchmark for an early budget, or want to sanity-check an outlier subcontractor number. It only works on data that's actually recorded in the system, so cost data that live in email threads or private spreadsheets stay invisible.
Getting a First-Pass Takeoff in Minutes
Manual takeoff meant tracing every area, measuring every run, and counting every symbol before any pricing could begin. Computer-vision tools now produce a first pass automatically. The best-known test is a University of Kansas comparison of Togal.AI and On-Screen Takeoff, published on Togal.AI's own site. On a fire station drawing set, Togal.AI took about 37 minutes (1.2 minutes of automated takeoff plus 36.2 minutes of manual correction) against 2 hours 35 minutes for the manual takeoff. That is roughly 76% less time, with most quantities within 5% of the manual results. On a reflected ceiling plan, the saving was smaller, 22% (16.7 minutes against 21.67).
Those figures need careful handling. The 76% comes from one project, the person doing the takeoff was new to both programs, and the study is hosted by the vendor. The more useful detail is where the time went: about a minute of automation, then over half an hour of checking and correcting. The software supplies the baseline, and the estimator still audits it.
Leveling (comparing, normalizing, and scope-aligning) Subcontractor Bids
Subcontractor quotes arrive in vastly different formats, ranging from unstructured PDFs and scanned bid proposals to customized spreadsheets with non-standard line items, distinct scope exclusions, and varying unit bases. Traditionally, precon teams spent critical hours—often under intense bid-day time pressure—manually retyping line items into Excel bid tabs, manually cross-referencing qualification letters, and attempting to spot buried scope exclusions.
Modern AI document-parsing tools automate this ingestion phase by extracting line items, quantities, and pricing structures directly from diverse PDF quotes and mapping them into a unified, standardized scope matrix. Procore's Bid Leveling agent compares proposals side by side, automatically flagging missing scope items, mismatched quantities, and pricing outliers, while its Bid Analyzer agent verifies submissions against project specifications and drawings. Similarly, Datagrid parses each bidder's inclusions, exclusions, and underlying assumptions, mapping them directly to the defined trade package scope.
From a practical standpoint, this shift provides estimators with a pre-populated bid tab to critique rather than a blank template to build. However, critical verification remains essential: automated extraction is only as accurate as the underlying OCR and language models, meaning subtle qualifications or non-standard contractual language still demand direct review and follow-up subcontractor calls prior to award.
Linking Field Productivity To the Schedule (emerging)
Schedules and labor assumptions have traditionally come from cost books and past job-cost codes. Measuring how much work a crew really produces meant manual observation, which the authors of a recent study describe as labor-intensive, delayed and hard to scale. The lesson reached the next estimate weeks late, if at all.
Two things are changing. On the commercial side, Buildots launched a Delay Forecast feature in May 2024 that flags at-risk activities, estimates the size of the predicted delay and suggests a pace correction. On the research side, a 2026 IEEE Access paper by Malaekeh and colleagues goes down to the individual worker. They combined a wearable motion sensor, a 24 GHz radar and Bluetooth zone beacons on a four-worker masonry crew, with the processing done on a computer at the site rather than in the cloud.
The system separated productive from non-productive work with 96.0% accuracy (± 0.8%) on workers it had not seen, against 93.0% for the motion sensor alone and 82–85% for a raw-feature baseline. Shrinking the model from 2.5 MB to 1.0 MB kept inference under 200 milliseconds, and a zone dashboard predicted completion times within about 6% of what was observed across four zones.
For precon, the appeal is clear: if output per worker can be measured automatically, the labor rates in the next estimate could come from measured productivity instead of a cost book. But this is a prototype, not a product. The data came from one crew, one site and one trade. The 96% is for a two-way split; the three-way result that feeds the dashboard is 83.6%, and idle versus semi-productive work was the hardest to separate. The authors describe the scheduling output as a feasibility illustration rather than a validated forecast, and the version I read is the accepted manuscript, not the final edited paper. Monitoring individual workers also raises consent and privacy questions; the study obtained informed consent from every participant.
Where This Leaves the Estimator
None of this replaces judgment about scope, risk and price. What has changed is how much clerical work stands in front of that judgment. Two limits are worth keeping in mind. Most of the performance claims above come from vendors, and independent measurement is thin. And adoption is partly a people problem: a 2026 interview study of 12 U.S. estimators looks at the workflow frictions and priorities that shape how these tools are used day to day.
Sources
Vendor pages describe product capabilities and are marked as such.
University of Kansas study: Togal.AI vs On-Screen Takeoff (vendor-hosted)
Procore AI agent library (vendor)
Datagrid: bid leveling guide (vendor)
Datagrid: document comparison agent (vendor)
Nomic: AI drawing set comparison (vendor)
Nomic: AI tools for bid preparation (vendor)





Comments