Engineering and architecture
Linking RFIs, submittals and change orders into one project record
By SourceX Editorial · Updated
Short answer
To link RFIs, submittals and change orders into one project record, join them on shared keys that already exist in the logs: project number, spec section, sheet and detail reference, and the RFI or change numbers cited in each document. Store the links as a separate table with a confidence level, and never edit the source logs.
Key takeaways
- Most links already exist as references typed into RFI responses, ASIs and change order descriptions; the job is to extract them.
- Spec section and sheet number are the strongest shared keys, once their formats are normalized across systems.
- Every link should carry its type and confidence, so explicit references are never confused with inferred matches.
- A linked chain from question to design response to cost outcome is worth more to AI developers than any single log.
Why connect the three logs at all?#
Connecting RFIs, submittals and change orders turns three lists of events into a record of how a design question became a decision and then a cost or schedule outcome. On its own, an RFI log shows questions and answers. Linked to the supplemental instruction it produced and the change order that followed, it shows consequence.
That chain matters for internal analysis, such as finding which details generate repeat questions or which consultants' sheets draw the most change cost, and it matters for licensing. The SourceX Enterprise Data Value Framework rates linked workflows higher through drivers such as AI utility, human-generated signal and domain expertise, because a model can learn from cause and effect rather than isolated text.
Which join keys connect RFIs, submittals and change orders?#
The join keys that connect project logs are identifiers your teams already type every day: project numbers, spec sections, sheet and detail references, and document numbers cited in other documents. None of them are perfect, so the table below notes where each one breaks.
Use more than one key wherever possible. A spec section match plus a date window is far more reliable than either alone.
| Join key | Where it lives | Typical break |
|---|---|---|
| Project number | Deltek or BQE project list, Procore or Newforma project settings | Accounting and project systems use different numbering; build a crosswalk |
| Spec section | Submittal log, RFI spec field, change narratives | Formats vary with and without spaces, and older projects use legacy section numbers |
| Sheet and detail reference | RFI drawing field, ASI sheet lists, Bluebeam markups | Sheets are renumbered between issue sets |
| RFI number | RFI log, cited in ASIs, PCOs and change order descriptions | Contractor and architect may keep separate RFI sequences |
| Submittal number | Submittal log, cited in RFIs about substitutions | Resubmittals get suffixes that differ by system |
| Change event numbers | PCO, change order request and executed change order logs | One change order often bundles several PCOs |
| Dates | Every log | Useful only as a tiebreaker inside a narrow window |
Explicit, referenced and inferred links#
Links between project records fall into three confidence levels, and keeping them apart is what makes the result trustworthy. Explicit links come from a system field, such as a change event or related-items field in the project management system that records its source RFI. Referenced links come from text, such as an ASI narrative that says it responds to an RFI.
Inferred links come from matching keys without any stated reference, such as an RFI and a change order on the same spec section within a short window. Inferred links are useful for analysis but should be labeled, sampled and confirmed by a project manager before anyone relies on them.
| Link type | How it is found | How to treat it |
|---|---|---|
| Explicit | Linked-item fields in the project management system | Accept, spot-check a sample |
| Referenced | Parsing RFI, submittal and change numbers out of descriptions and responses | Accept after checking the number exists in the same project |
| Inferred | Same spec section or sheet within a date window | Label as inferred and confirm with project staff |
Step-by-step: building the link table#
A link table is a simple list of pairs, each saying record A relates to record B, with the link type, the evidence and who confirmed it. Keeping links outside the source logs means the original exports stay untouched and auditable.
Export method shapes how much parsing you will do. Procore's Extracts documentation, for example, says most items download as PDFs while documents and photos keep their original file type, so a firm relying on that route needs extra processing before linking; tabular log exports are easier to work with where the system offers them.
Include the documents that sit between the logs. ASIs, bulletins, construction change directives and sketch issues are often the bridge that explains why a question turned into a cost, and firms name them inconsistently from studio to studio, so agree on one document-type vocabulary before parsing.
- Export complete RFI, submittal, ASI or bulletin, PCO and change order logs for each project, including voided and superseded items.
- Build a project crosswalk between accounting project numbers and project management system IDs.
- Normalize spec sections to one format and map sheet numbers across issue sets using the drawing index.
- Pull explicit links from linked-item fields.
- Parse document numbers out of free text and keep only those that match a real record in the same project.
- Generate inferred candidates from spec section, sheet and date matches, and flag them.
- Have a project manager confirm a sample of inferred links per project and record the result.
Illustrative: one curtain wall question, traced end to end#
Illustrative: a fictional A/E firm with architecture and structural studios wants to see whether its project records can be linked before a licensing review. Its digital practice leader picks a completed lab building and traces one chain as a test.
The chain reads as follows. A contractor RFI asks how the curtain wall sill meets a revised slab edge, citing the curtain wall spec section and an enlarged detail sheet. The architect answers with a sketch and issues an ASI that cites the RFI. A related curtain wall shop drawing submittal is returned revise and resubmit with a comment pointing to the ASI. The contractor then prices the change as a PCO referencing both the RFI and the ASI, and the PCO is later bundled into an executed change order.
Every link in that chain was either explicit in Procore or written into a description by someone on the team. The firm decides to run the same parser across its completed projects, accept explicit and referenced links, and keep inferred links in a separate review queue.
Mistakes that break the chain#
The most common mistake in linking project logs is matching on titles or subjects, which produces confident but wrong links between unrelated items about the same building element. Titles are written in a hurry and repeat constantly; spec sections and cited numbers do not.
Other mistakes include dropping voided RFIs and superseded submittals, which removes the dead ends that explain later decisions, and mixing the contractor's numbering with the architect's without recording whose sequence each number belongs to. Bundled change orders also need care, because one executed change can trace back to several unrelated questions.
How SourceX uses linked project records#
SourceX looks at linkage during the fit check, using only metadata: which systems hold each log, whether linked-item fields were used, and how many years of projects are accessible. No files are shared at that stage.
If a firm proceeds, linking happens alongside the Preparation step of the SourceX five-step transaction, after rights review has confirmed which projects can be included. The link table, its confidence labels and the reviewer sign-off are documented in the SourceX Evidence Packet so the supplier and buyer see the same description of how records connect.
Frequently asked questions
Do we need a data warehouse to do this?
No. A first pass works with exported spreadsheets, a scripting language or a capable spreadsheet tool, and a single link table. A warehouse helps if you want to repeat the analysis every month, but the link logic is the same either way.
What if our older projects kept logs in Excel?
Excel logs can still be linked if they kept spec sections and cited document numbers. Normalize their columns to match your newer system's exports, mark the source, and expect more inferred links and fewer explicit ones for those projects.
Should the contractor's change order log be included?
Only if you hold it under terms that allow its use. Architects often hold change order requests and executed change orders as part of construction administration, but the contractor's pricing detail can carry confidentiality terms. Link to the change event and its status, and treat cost detail as a rights question.
How does linking interact with de-identification?
Link first, using real identifiers, then de-identify. Pseudonymize project numbers and document numbers consistently across all logs so the link table still works after transformation. Linking after de-identification is harder and loses matches.
Is linked data always more valuable?
Linked data is usually more useful, because outcomes give context to questions. Value still depends on other factors, including rights, recency, data cleanliness and privacy burden, and is only known once a buyer engages with a specific package.
Who should own the link table inside the firm?
A digital practice or BIM lead usually owns the method and the code, while project managers confirm inferred links on their own projects. Keep the table with the exports it was built from, and record the date and system version so the work can be repeated or audited later.
Sources
- Procore's Extracts documentation says most Procore items download as PDFs, while documents and photos keep their original file type. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.