Audiences AIWhich Five RFIs Matter
Picture a project trailer at half past six on a Tuesday morning. The RFI log on the screen shows forty-three open items. The number is illustrative, but anyone who has run a job of any size will recognize the shape of the moment. Every row carries a number, a ball-in-court party, a date opened, a count of days ageing. Nothing in those fields says which five of the forty-three could move the completion date, or cost meaningful money, if they sit unanswered another week.
That gap has a shape we keep meeting in our research. The record itself is in good order. Procore has captured each question cleanly, routed it to the right court, and tracked every day it has stayed open, and Procore's own material on the subject is frank about how approval delays ripple through construction schedules. What each open question means for this particular project in this particular week still gets assembled by hand, usually by a superintendent or a project manager who holds the sequence of work in their head and knows which trades are stacked behind which answers.
Sort the log by days open and an RFI confirming a paint sheen sits next to an RFI asking whether a steel embed clashes with a post-tension tendon. Both examples are invented; the adjacency is the point. One of them can stay open for a month without consequence. The other is quietly holding a pour, and behind the pour, a floor cycle. Same document type, same workflow, same ageing counter, very different stakes. The judgment that separates them is real work, performed every day on every serious project, and almost nowhere is it written down as logic a system could apply and a team could audit.
A working vocabulary for an open RFI
Our research on this pattern starts with a vocabulary. Read an open RFI against the schedule, the procurement log and the drawings, and it resolves into one of a small set of states:
| State | What it says about this morning |
|---|---|
| INFORMATION_ONLY | Someone wants an answer; no work is waiting on it |
| FIELD_BLOCKING | A crew cannot proceed at that location until it closes |
| CRITICAL_PATH | The answer sits on the sequence that sets the completion date |
| PROCUREMENT_LONG_LEAD | The answer gates an order whose lead time is already long |
| OWNER_DECISION_REQUIRED | The design team cannot close it; the owner has to choose |
| LIKELY_CHANGE_ORDER | The answer will probably arrive with cost or time attached |
| DESIGN_COORDINATION | The documents disagree; consultants need to reconcile them |
| NO_CURRENT_IMPACT | Genuinely fine to leave open this week, and worth saying so |
Assigning these states takes inputs the project already holds in digital form: the current schedule and its critical path, the procurement log with lead times, which locations each trade is working, who is waiting on what. The schedule, the procurement log and the drawing register already hold every input this reading needs, and the rules doing the reading are ones the project team can see and argue with.
That matters because the states lead to materially different actions, and one legitimate action is doing nothing this week. A log where NO_CURRENT_IMPACT is an honest, recorded conclusion is a calmer log than one where every item ages with equal menace. The cost of misreading runs in both directions: chase an answer no crew is waiting on and you spend design-team goodwill; treat a PROCUREMENT_LONG_LEAD item as routine and a two-week question quietly becomes a two-month delay. Both costs are illustrative in the particulars and familiar in the pattern.
The natural question is where such logic should live, and our answer is beside Procore, never instead of it. Procore records the RFI, moves it through its workflow, and closes it, and it does that job well. The decision layer we are studying would read what Procore already holds and answer a smaller, sharper question: of everything open this morning, what deserves attention first, and why. The why travels with the answer, as attached reasoning a project engineer can inspect, and a person keeps the final call on every item.
To be plain about where this stands: this is research territory, and the only systems available for pilot today are Interpret and Triage.
In the meantime the pattern can be run by hand, and probably should be. Tomorrow morning, before anyone sorts the log by days open, a superintendent could ask one question of the team: which of these items, left open one more week, moves a date or a dollar amount the owner would notice? If assembling that answer takes an hour of cross-checking the schedule and the procurement log, that hour is the manual version of the decision logic this research is trying to describe.
If a morning like this one looks familiar, talk to us about your RFI log.