WhatsApp Issue Tracking for Construction Teams

The problem

The chat knows first. The register finds out weeks later.

The fastest, most truthful record of what happened on site lives in WhatsApp, walk-downs and huddles. It is also the least durable place it could live.

history_toggle_off

Decisions evaporate

A supplier confirms a two week slip at 10:42 on a Tuesday. Three hundred messages later it is gone, and the schedule still says Friday.

photo_library

Photos die in camera rolls

The best evidence on the project is taken on phones and shared once in chat. It never gets attached to the issue it proves.

edit_note

The register is rebuilt from memory

Weeks later someone reconstructs an issues log by scrolling the group. What was raised, by whom, and when becomes a best guess.

How it works

Chat in. Issues register out.

Three steps between a site conversation and an evidence-linked commissioning record. Nothing changes for the people on site.

link

Link the group, or upload an export

Connect a live project group, or hand over a WhatsApp chat export backup from any project, including a finished one.

neurology

AI reads messages, photos and voice notes

Every message is read in the context of the whole conversation. Issues, actions, risks, open questions and decisions are proposed as typed suggestions, each carrying the messages that raised it.

how_to_reg

A person confirms, and only then is it real

Suggestions wait in a review rail. Confirming one creates the record with its evidence attached. Nothing enters the register without a human decision.

cxstat.com/field-intel · detected items awaiting review

radarField IntelAuto-scan · 2 min ago · NG1 DH2 L4 groupLive

2Issues

4Actions

2Risks

2Open Qs

reportIssue detectedHigh

Damper actuator on CRAH-02-01 not reaching commanded position (stalls ~80%)

“It’s stalling around 80% every time, even after we recalibrated.”

MBM. BianchiCRAH-02-01Mechanicalopen_in_newSource

addCreate issuecloseDismiss

task_altConfirmed by D. WhelanNG1-03

Created in the issues register with severity, system and 2 evidence photos

linkTraceable to the 09:38 message that raised it

The review rail: AI proposes, a person confirms or dismisses. Only confirmed items become records.

cxstat.com/issues · register built from a WhatsApp export

CxSTAT issues register generated from a WhatsApp chat export, with severity, testing impact and evidence links

swipeSwipe to explore the register

Confirmed items land in the issues register, filterable and exportable.

From one real pilot: a single L4 testing group export produced a register of 20 issues, nine of which had halted testing, with 34 evidence photos, each entry linked to the message that raised it. The screens on this page carry the same structure with fictional demonstration data.

Context-aware reading

It reads today’s message with the whole conversation in mind.

Site chat is full of pronouns, shorthand and follow-ups that only make sense with history. Extraction here is thread-aware: each message is interpreted against the running conversation and the project graph, not in isolation.

history

Thread-aware extraction

The model carries the conversation history as context. “Still stalling at 80%” three days later is understood as the damper fault from Tuesday, not a new report.

hub

Entity resolution against the project graph

“It”, “that unit” and “same fault again” resolve to the tagged equipment, system and open issue they refer to, using the Turii project graph as the reference.

join_inner

One fault, one record

Repeated mentions deduplicate against the existing register. Follow-ups land on the open issue as status updates and evidence, not as five copies of the same fault.

timeline

The full story, in order

Because updates accumulate on one record, each issue carries its whole timeline: raised, investigated, parts ordered, fixed, verified, with the source message behind every step.

cxstat.com/field-intel · thread-aware review

Tuesday

M. Bianchi · 09:38

Damper actuator on CRAH-02-01 stalls around 80% open on repeated commands.

reportNew issue proposed · NG1-03

Friday

M. Bianchi · 08:12

Recalibrated twice now, it’s still stalling at the same point.

linkRecognised as NG1-03, the open CRAH-02-01 damper issueStatus updated · no duplicate created · message linked as evidence

The following Wednesday

L. O’Brien · 14:20

Replacement actuator fitted and stroking full range. Photo attached.

Site photo

task_altNG1-03 resolution proposed with photo evidenceAwaiting human confirmation to close

Two ways in

Live on today’s project, or retrospective on any chat.

bolt

Live: linked groups

Field Intelligence watches linked project groups and proposes records as the conversation happens. The register stays current with the site, not with the paperwork backlog.

upload_file

Retrospective: upload an export

Export any WhatsApp group and upload the backup. CxSTAT reads the whole history into a register, which works on finished projects and on disputes about what was raised when.

graphic_eq

Beyond chat: voice notes and meetings

Walk-down voice notes and meeting capture are transcribed and read the same way, so the decisions spoken at the pump reach the record too.

verified_user

Always: a person decides

In every mode the AI only proposes. Confirmation is human, severity is human, and sign-off is human. Humans sign, the system proves.

Common questions

Frequently asked.

Does the site team have to change how they work?add

No. Teams keep the WhatsApp groups they already run the job in. There is nothing new to install on site and no forms to fill. The record is built from the conversation the team is already having, and the messages stay in WhatsApp.

Can it build an issues register from an old WhatsApp chat export?add

Yes. Upload a WhatsApp chat export backup and CxSTAT reads it retrospectively into an evidence-linked issues register with severity, system and testing impact. In one pilot, a single L4 testing group export produced a register of 20 issues with 34 evidence photos. It works on live and finished projects.

Does the AI create issues on its own?add

No. The AI proposes issues, actions and risks, and every suggestion carries the source messages that raised it. A person reviews and confirms before anything enters the commissioning record. Humans sign, the system proves.

How does it avoid duplicate issues when the same fault is discussed for days?add

Extraction is thread-aware: every new message is read against the running conversation history and the project graph, not in isolation. References like “it” or “the same fault” resolve to the tagged equipment and the open issue they belong to, so follow-ups attach to that record as updates and evidence instead of creating duplicates.

What happens to photos and voice notes in the chat?add

They are read alongside the text. A photo of a nameplate, a leak or a tripped breaker is linked as evidence to the issue it supports, and voice notes are transcribed into the same review. Media stops dying in camera rolls and becomes part of the record.

Still have questions? We’re here to help.

Book a Demo arrow_forward