AI email triage that knows your open jobs.
Dataforge reads every email in your shared operations inbox, checks it against the jobs, sites and contacts in your system, and raises the right action with a reason. Your team confirms the work instead of keying it.
A walkthrough of the review console, then a pilot on your own inbox if it fits.
The work behind a shared operations inbox.
Thousands of emails a month and a handful of people to read them. Every one has to be matched to a job, a site or a customer before anything can happen.
- 01
Every email read by a person
Out-of-office replies, bounces and newsletters get the same attention as a burst pipe.
- 02
Searching for the job by hand
“Is this a new job, or chasing the one from last week?”
Each answer means a search in your system before anything can be keyed.
- 03
One issue, three jobs
The same leak reported by three residents becomes three jobs for one visit.
- 04
The urgent one on page three
Urgent emails wait in the same pile as everything else, in the order they arrived.
- 05
No view until something is late
Nobody can see how the inbox is going until a customer chases.
From inbox to job in four steps.
Each email is cleared, matched to your data and decided, then either written to your system or held for a person. Here are three emails from the same morning going through.
- 01
Clear the noise
Out-of-office replies, bounces, alerts and bulk mail are set aside before any AI runs. Old thread history is recognised, so a reply chain isn’t read as a new request.
- 02
Ground it in your system
Job references, site names, postcodes and the sender’s address are matched against your open and closed jobs, sites, accounts and contacts.
- 03
Decide the action
Team, category and priority, then the action: create, update, reopen, close, escalate, or hold for a person, with a plain-English reason.
- 04
Act or ask
Confident decisions go into your system with every mandatory field filled. Anything uncertain lands in the review queue, pre-filled, for a one-click confirm or correct.
Rules guard the edges. AI proposes. Your team sets the limits.
The AI never has the final word on the things that matter. Emergencies, priorities and anything it is unsure about are settled by fixed rules or by a person.
Fixed rules go first and last.
Rules clear the noise before any AI runs, and rules have the final say on priority. They can only move an email up.
AI reads, matches and proposes.
It works out what the email is about, which job it belongs to and what should happen, and a supervisor step checks each proposal.
Your team sets the limits.
You choose which categories it handles on its own. Everything else waits in the review queue with a reason.
A missed collection, matched to the job that already exists.
Ama's email quotes an old reference and says the problem is back. Dataforge checks the reference, finds the closed job and reopens it, so the history stays in one place and nobody raises a duplicate.
- From
- Ama Boateng
- Subject
- Recycling missed again
- Sent
- 12/10/2026 07:42
Hello,
For the third week running the recycling at Quellbrook House hasn't been collected. The bin store is overflowing and it's attracting rats now.
You gave me reference J-47790 last time and told me it was sorted.
Ama, flat 12
- Sender
- ✓Ama Boateng, resident, Flat 12
- Site
- ✓Quellbrook House, QB7 1RN
- Job found
- ✓J-47790 Missed recycling collection, closed 29/09/2026
- Action
- Reopen J-47790
- Routed to
- Estate services · Waste and recycling · High priority
- Reason
- Same site and issue as J-47790, closed 13 days ago, and the sender says it has happened again. Pest risk raises the priority.
Triage that checks your data before it acts.
Labelling an email is the easy part. The value is in knowing which job it belongs to and doing something about it.
It checks before it acts.
It looks up the open jobs, sites and contacts in your system first, so “re: the leak at flat 4” becomes an update to the job that already exists. Every job reference it uses is checked against real data, and it never makes one up.
It picks the action as well as the label.
Create, update, reopen, close, escalate, hold, or split an email that asks for two different things. Each proposal comes with a one-line reason a person can read in two seconds.
It writes straight into your system.
New jobs arrive with the required fields already filled, from the site and account to the contract, instruction type and reporter, with the original email attached as the job note.
Emergencies are never downgraded
Urgent safety language raises priority by fixed rule, and rules can only move an email up. A second check catches likely emergencies the first pass missed and puts them in front of a person.
A second opinion on every decision
A supervisor step reviews each proposal and can veto it or send it to a person instead. Hard cases go to a deeper check with strict limits.
It reads the attachments
PDFs, Word documents and spreadsheets are read with the email body, so the quote, the invoice or the form is part of the decision.
A review console built for speed
One queue sorted by day and urgency, with response clocks. Each held email shows the key sentence, the proposed action and why. Confirm, correct, junk or escalate from the keyboard.
A record of the day
The day log and team view show what came in, what was done with it and who did it.
It learns from your team
Every correction is stored with the proposal it corrected, and used as evidence for how similar emails are handled next time.
See the review console, the matching and the audit trail with emails like the ones your team gets.
Book a demoYou decide how much it does on its own.
For operations leads, IT and data protection: what it can do without a person, what it sees, and where the record of it lives.
- Autonomy by category
- Start in shadow mode, where it proposes and your team decides, then let it act on its own for the categories you trust.
- Personal data scrubbed
- Names, phone numbers, addresses and email addresses are replaced with placeholders before any AI model sees an email, and restored afterwards.
- UK and EU data residency
- Your data stays in the UK or EU, whichever model does the reading.
- Full audit trail
- Every email, every proposal and every human decision, timestamped and attributable.
- No emails to your customers
- It never sends a reply. It works the inbox, and your people own every conversation.
- Measured against your experts
- Accuracy is tested against emails labelled by your own team, and every change is checked against that benchmark before it goes live.
Autonomy is set category by category, as the numbers earn it. Emergencies always go to a person, whatever the settings say.
Measured on emails it had never seen.
- 90%+of emails routed to the right team
- 1 in 5emails cleared as noise before any AI runs
- Over halfof all email actioned with no keystroke, the rest held with a reason
- +10 ptscategorisation accuracy on emails with attachments, from reading them
A live pilot on a high-volume operations inbox. Accuracy was measured on emails the system had never seen, and every figure is rounded down.
Connects to your mailbox and the system you already use.
Dataforge Inbox reads from a Microsoft 365 or Google mailbox and works inside your ERP, job management or CRM system. We confirm how it connects to yours before a pilot starts.
A pilot on your own inbox, in four stages.
- 01
Connect the inbox
Read access to one shared mailbox and a connection to your system. Nothing is written until you say so.
- 02
Run in shadow mode
It proposes an action for every email and your team decides. Every proposal and every correction is recorded.
- 03
Review the numbers together
Accuracy by team, category and action, measured against what your own people decided.
- 04
Switch on autonomy where it’s earned
Category by category, only where the numbers support it. Everything else stays with your team.
Stratas has helped more than 150 organisations automate document‑heavy work.
Read customer stories
No. Dataforge Inbox never sends email to your customers or contractors. It reads, routes and raises the work, and your people own every conversation.
It reads from Microsoft 365 or Google mailboxes, and connects to your ERP, job management or CRM system to look up jobs, sites and contacts and to write jobs back. We confirm how it connects to your system before a pilot starts.
The email goes to the review queue with the proposal already filled in and a plain-English reason. A person confirms or corrects it in one click, and the correction is kept as evidence for next time.
It takes away the reading, sorting and retyping. Your team still makes the calls it holds, handles the conversations, and decides how much it does on its own.
We test it against emails labelled by your own experts, measured on emails it has never seen. Every change is checked against that benchmark before it goes live, and the results are shared with you.
In the UK or EU. Names, phone numbers, addresses and email addresses are replaced with placeholders before any AI model sees an email and restored afterwards, and every action is kept in the audit trail.
Yes. PDFs, Word documents and spreadsheets are read alongside the email body, so a quote, an invoice or a form counts towards the decision.
Read access to one shared mailbox and a connection to your system, a shadow period where it proposes and your team decides, then a review of the numbers together before anything runs on its own.
See it on emails like yours.
We'll walk through the review console, the four steps and the audit trail, and talk through what a pilot on your own inbox would look like.
No commitment. We'll set the demo up around how your inbox works after you enquire.