Australian construction firms run on spreadsheets. Tender trackers, subcontractor schedules, RFI registers and programme logs mostly live in Google Sheets because the file is shared, cheap and already open on every site laptop. The general-availability release of ChatGPT inside Google Sheets now puts a capable language model one cell reference away from all of that data. For a Sydney or Brisbane head contractor, that is either a quiet productivity win or a new way to put a confidential tender at risk, depending entirely on how you set it up. This guide is a practical look at where it earns its keep on tender trackers and programme logs, what it costs in real dollars, and where you still want a person holding the pen.
What ChatGPT in Google Sheets actually does
The release is an add-on that exposes functions such as =GPT() and =GPT_LIST() directly inside the grid, alongside a side panel for longer instructions. You point it at a range, describe what you want in plain English, and it writes text back into the cells. There are no macros to record and no Apps Script to maintain. For a commercial admin team that has never written a line of code, that drops the barrier from 'we need to hire a developer' to 'we can try this on Friday afternoon'. The functions recalculate like any other formula, so once a prompt works on one row you can fill it down a whole column.
The jobs it does well on a construction sheet are the repetitive reading-and-rewriting tasks that usually eat an estimator's or contract administrator's afternoon:
Summarising long tender addenda into a one-line note per clause
Classifying incoming RFIs by trade so they route to the right foreman
Drafting standard variation-notice wording from a few bullet inputs
Flagging programme rows where the forecast finish sits past a contract milestone
Cleaning up inconsistent supplier names so a cost report stops double-counting the same subbie
A worked example: the tender tracker
Picture a typical tender tracker. One row per trade package, with columns for the subcontractor, the submitted price, their qualifications and a status. The qualifications column is where the hours disappear, because every subcontractor writes their exclusions and inclusions differently. Point =GPT() at that column with an instruction like 'list any scope exclusions mentioned, comma separated' and it drafts a clean, scannable summary column in a fraction of the time it would take to read each submission by hand. You can add a second prompt that flags any package where the lowest price also carries the most exclusions, which is usually the bid that comes back to bite you.
Programme logs work the same way. Give the model the planned and actual dates for each activity and ask for a short variance comment, and you get a first draft of the narrative a project manager would otherwise type out for every monthly report. It is a draft, not a finished document, but starting from a draft is far quicker than starting from a blank cell. Before you point any of this at something that matters, set a few ground rules:
Work on a copy, never the live master tender file
Strip out pricing you are contractually barred from sharing with a third-party processor
Spot-check every cell it writes before that cell informs a submission or a claim
Keep a column noting which rows were AI-drafted, so you have an audit trail later
The cost and accuracy reality
The add-on runs on usage-based pricing on top of your existing Google Workspace plan, so the bill scales with how much text you push through it. A six-person commercial admin team using it across live tenders might spend around $1,800 a month once usage settles, set against the roughly $90,000 a year a full-time tender coordinator costs in Sydney. Framed that way the maths looks easy. The catch is accuracy. The model will, every so often, confidently misread a scanned qualification or invent an exclusion that was never in the document. On a tender worth $2.4 million, one hallucinated exclusion that nobody catches is not a rounding error, it is a margin you do not get back. The right mental model is a fast junior who drafts well and needs everything checked, not an oracle.
Where Claude fits for the higher-stakes work
For anything touching contracts, work health and safety records, or data you would rather not see leave Australia, the question stops being 'which spreadsheet function' and becomes 'where does this data actually go'. That means asking whether the provider trains its models on your inputs, whether there is an audit log of what was sent, and whether data residency meets your obligations under the Privacy Act. Those are the workflows we tend to build on Claude, because the data-handling controls are clearer and the same spreadsheet reasoning still works whether your register lives in Google Sheets or Excel. A tender tracker full of confidential subcontractor pricing is exactly the sort of data that deserves that extra care before it is pasted into any AI tool.
ChatGPT in Google Sheets is a real step forward for the site and office teams that already live in cells. The sensible way in is small: start it on a copy of one tender tracker, measure the hours it saves over a fortnight, and keep a human signature on anything that reaches a client or a certifier. If you want help designing a spreadsheet workflow that stands up to an audit, or a Claude-based version with tighter data controls, book a brainstorm with us.
