Analysing construction bids with a ChatGPT plugin
ChatGPT plugin for the ACT stage: map bids to the CCTP and DPGF, recalculate without altering originals and track every round of clarifications.
By Educasium
Install the ChatGPT plugin: Tender analysis →
Fictional example. A practice receives the three bids for trade 02 Structural works (gros œuvre) on a small health centre: A quotes €44,800 excl. VAT, B €47,200 and C €41,900, without the earthworks line. The spreadsheet is filled in and a question goes out to C. Ten days later, C sends back a corrected bid under the same file name, B answers only two of three questions, and the client asks how the comparison is going. Nobody can say any more which version of the table matches which version of the bids.
This article presents the Tender analysis ChatGPT plugin, an Educasium resource designed for this multi-round work: the context of the ACT mission, what the plugin adds to a simple comparison, its cumulative file, its human reviews, how to use it, how it differs from the skill of the same name, how to install it and where its limits lie.
Contents
- ACT: informing the client, not choosing for them
- What the plugin adds to a bid comparison
- The cumulative file: memory from one round to the next
- Human reviews H1 to H4 and the conditional recommendation
- Using the plugin step by step
- Skill or plugin: what is the difference
- Installing the plugin in Codex
- When this plugin is not enough
- Training in bid analysis with AI
- Frequently asked questions
ACT: informing the client, not choosing for them
In a French architectural services contract, ACT (assistance pour la passation des contrats de travaux, i.e. assisting the client in awarding the works contracts) means helping the client compare bids, obtain the necessary clarifications and prepare their decision. The architect informs; the client awards.
The tool must never turn an assumption into a fact: a missing line is not a free line, the word "included" does not price the work, and a price gap alone proves neither an error nor an abnormally low bid. The plugin prepares the analysis; it does not conclude it for you.
What the plugin adds to a bid comparison
Line-by-line mapping to the CCTP and DPGF
Every bid line is kept exactly as the contractor wrote it, with its source (file, page, row or cell actually read), then mapped to the matching article of the CCTP (the trade-by-trade technical specification) and line of the DPGF (the priced breakdown of the lump-sum bid that contractors fill in). Multiple or uncertain matches are kept with their justification. A lump sum covering two items keeps a single amount, with no invented split.
Base, options and variants, stated and recalculated
Base scope, options, variants, discounts and taxes are isolated in separate scenarios. The contractor's stated total is never overwritten: the plugin shows the stated amount, the recalculated amount (quantity × unit price, when both are known) and the difference.
| Situation in the bid | How the plugin handles it |
|---|---|
| Blank cell or missing line | Unknown amount, flagged, never zero |
| "Included" | Cost not isolated |
| Stated total differs from recalculation | Both values and the gap, original preserved |
| Priced option | Separate scenario, inactive without a documented decision |
| Discount with no basis | Discounted total to be clarified |
| No VAT rate given | VAT-inclusive total unavailable, no assumed rate |
Technical scopes (units, quantities, exclusions, timescales) are compared before prices.
Outputs in your practice's format
You can point the plugin to a Word or Excel template, or to a previous report. It derives a style profile (sections, terminology, display conventions) without confusing presentation and method: an old report supplies no prices or criteria, and a "price score 60%" column in a template does not prove that this criterion was adopted. Outputs are produced in XLSX and DOCX depending on the tools in your session, otherwise in JSON and a text note, with the limitation stated.
The cumulative file: memory from one round to the next
This is the main difference from a one-off comparison. The plugin keeps a bids file (dossier-offres.json) that is the memory of the tender; the comparison table and the note are only dated views of it.
In the next round, a few rules prevent silent losses:
- same file name, different content: new version, and the old one stays in the history;
- a partial answer only updates its own items; an omitted line keeps its last known state, marked as not updated, and never becomes a zero price;
- a contractor's answer is a sourced statement; its technical review and its acceptance are separate acts;
- a question sent waits for its answer, and an answer received waits for its review;
- affected results are recalculated, the delta is shown and the reviews concerned go back to "to be reviewed".
The reference remains the tender package actually issued. If the tender CCTP required 72 mm insulation and a later working version requires 98 mm (fictional values), the bids are still compared with the 72 mm version. If you drafted your documents with our ChatGPT method for the CCTP per trade, they serve as a reference, without their approvals becoming analysis decisions.
Human reviews H1 to H4 and the conditional recommendation
The plugin prepares the work without asking for confirmation on every line, then groups decisions into four packages submitted to the people you designate.
| Review | What is presented | Outside the tool's remit |
|---|---|---|
| H1 Comparison framework | Documents issued and read, bids and versions, criteria and method, gaps | Creating criteria after reading prices |
| H2 Scopes and mappings | Line-by-line links, lump sums, inclusions and exclusions | Deciding whether a bid is admissible |
| H3 Analysis and recommendation | Stated, recalculated, gaps, scores, reviewed answers | Awarding the contract |
| H4 Version and communication | Files, versions, recipient, requested action | Signing, notifying or sending |
Decisions are recorded in a register (revue-humaine-offres.json) with their author and scope. Without a real decision, a review stays pending: the assistant that drafts is never the approver.
If criteria and a method are provided, the plugin applies the formula exactly. A missing technical score makes the final score unavailable, with no zero and no redistribution of weights. Without criteria, the comparison stays factual, with no official ranking. The recommendation is always conditional.
Using the plugin step by step
Step 1: Gather the real file. Documents actually issued (CCTP, CCAP, i.e. the special administrative conditions, DPGF and drawings), dated and versioned bids, announced award criteria and, if possible, your practice's report template.
Step 2: Start the analysis in Codex. Name the plugin explicitly: "Compare the three fictional bids for trade 02 Structural works with the attached CCTP and DPGF. Document gaps, comparability limits and questions for contractors."
Step 3: Confirm the framework (H1). Check the documents read and the method used.
Step 4: Check the mappings (H2). The missing earthworks in C's bid appear as a missing line, and the conclusion "C is cheaper" stays blocked until the scope is confirmed.
Step 5: Send the questions yourself. The plugin prepares them by trade and by line, without revealing competitors' prices, but does not send them.
Step 6: Pick up in the next round. Provide the complete previous file, C's corrected bid and B's partial answer. The plugin adds the new version, recalculates and shows what changed.
Step 7: Finalise (H3 and H4). Approve the analysis, then choose the version to send to the client.
Skill or plugin: what is the difference
Our article on the ChatGPT method for construction tender comparison presents the skill: a method applied within a conversation to build a clean comparison. The plugin takes that method and adds what a tender running over several weeks demands.
| Need | Skill | Plugin |
|---|---|---|
| Comparison with explicit scope | Yes | Yes |
| Memory between clarification rounds | Within the conversation | Versioned cumulative file |
| Approvals | Done by you | Traced H1 to H4 reviews |
| Practice template | Not covered | Style profile |
| Environment | ChatGPT | Codex, after installation |
Installing the plugin in Codex
The download is reserved for participants in Educasium training courses. From the plugin page, download the ZIP and unzip it into a stable folder you will not move. In a terminal, add the catalogue with codex plugin marketplace add "path to the unzipped folder", then install the plugin with codex plugin add archi-analyse-offres@educasium-analyse-offres. Then open a new Codex session and check that the installed version is 0.2.2.
When this plugin is not enough
The plugin organises document work; it replaces neither your tendering tools nor your judgement.
- No bid-opening portal and no submission platform for contractors.
- No proprietary import engine and no OCR: reading depends on the tools in your session.
- No signature, access control or approval lock: the register documents decisions, it does not secure them.
- Award criteria come from the real project file, never from a default scoring grid.
- The recommendation is never an award.
Training in bid analysis with AI
A plugin is only useful if you know how to feed it the right file and review what it produces. Our Architectural Management with AI training works on these uses through concrete project cases, from written documents to bid analysis.
Frequently asked questions
Can the plugin name the contractor to choose?
No. It makes a conditional recommendation, separating facts, risks and points to clarify, and applies a scoring method only if one is provided with its source. A price order is not a multi-criteria ranking, and an approved analysis is not an award. In the ACT mission, the architect informs the client, who decides. The plugin does not sign, notify or contact contractors without an explicit request.
What happens when a contractor sends a corrected bid?
Provide the complete previous file together with the new bid. The plugin records the new version without deleting the old one, even under an identical file name. It links lines whose identity is established, keeps the last known state of omitted lines and never treats a missing line as included. The affected calculations are redone and the reviews concerned go back to pending, instead of carrying over an approval that applied to a different version.
Can the plugin be used without award criteria?
Yes, but the output changes in nature. Without documented criteria and method, the plugin produces a factual comparison: stated and recalculated amounts, scopes, anomalies and questions. It produces no official ranking, invents no scoring grid and creates no criteria after reading the prices. The H1 review makes this gap visible so that the client or the designated lead can supply the tender's actual method.
Do I need Revit or a BIM model to use the plugin?
No. The plugin works from the tender documents and the bids: spreadsheets, PDFs and text. No model is required and no IFC reader is included. Reading and export capabilities depend on the tools in your session: an unreadable PDF stays flagged as not read rather than guessed. Any instructions that might appear inside a bid are treated as data, never as commands to execute.
For single-pass comparison, read our ChatGPT method for construction tender comparison; upstream, see our ChatGPT method for the CCTP per trade.
Funding possible depending on your eligibility (OPCO, FIF PL). Qualiopi-certified training provider. To structure your bid analyses with AI as part of our Architectural Management with AI training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal. You will find the download and installation tutorial on the dedicated ChatGPT plugin page.