Our advisers spend a large part of their week sorting evidence, cross-checking documents, writing first drafts and answering enquiries that never become clients. AI agents can take on that groundwork so advisers spend their time on advice, which is what clients pay for. This page covers agents outside the Visa Platform, for casework and for winning new clients.
Sorting, cross-checking and first drafts move to the agent. Advisers review and decide.
Enquiries get a reply in minutes, in the person's language, at any hour, instead of the next working day.
The same team can take on more cases, including the rush after a policy change.
Conflicting facts and missing items are found before INZ finds them, which means fewer follow-up requests.
An agent does a whole job, not a single answer. It reads the files, sorts them and drafts the output, then hands the result to a licensed adviser. Every point links back to the original document, anything uncertain is marked as uncertain, and each adviser approval is recorded. Nothing reaches a client or INZ until an adviser approves it.
We don't start everything at once. Each phase has to prove itself before the next one starts.
| Phase | Agents | What it proves |
|---|---|---|
| Phase 1 (pilot) | Partnership evidence organiser, with the case consistency check Enquiry intake and booking, on the website | Adviser hours saved, and more enquiries turning into consultations |
| Phase 2 | Case briefing and handover Pre-lodgement check INZ letter response assistant Consultation notes and follow-up Employer paperwork assistant | Wider daily use across all case types |
| Later, if needed | Tribunal decision research, interview practice, WhatsApp and WeChat, policy-change outreach, multilingual content | Decided on demand and on what it takes to make each one reliable |
Turns hundreds of chat screenshots, photos, bank statements, tenancy papers and travel records into an indexed evidence bundle.
Example: 300 files in, an indexed bundle out, with a note: "No evidence of a shared address from March to August 2024."
What we gain: the slowest part of a partnership file becomes a review task, and gaps are found before INZ finds them.
Watch out: photo dates can be wrong, so the agent marks uncertain dates. It never edits or crops evidence.
Compares the passport, questionnaire, application form, CV, employment contract, bank statements and any previous applications, and finds facts that don't match.
Example: "Questionnaire says the job started in March 2022; the employer letter (page 1) says June 2022. Which is correct?" Other typical finds: address dates that don't match the tenancy, or work history that differs from a previous application.
What we gain: far less manual cross-checking, and fewer INZ follow-ups and rework after lodging. It reuses the document reading built for the evidence organiser.
Watch out: it never decides which version is right and never labels a document as false. Its job is to show the difference and where it comes from.
Answers new enquiries on the website in the person's language, collects their situation and books a consultation. WhatsApp and WeChat come later.
Example: a Saturday night enquiry in Mandarin is booked for Monday 10am with a Mandarin-speaking adviser, who already has the facts.
What we gain: enquiries don't go cold over the weekend, and consultations are spent on advice instead of fact-finding.
Watch out: it must not tell anyone whether they qualify. That is immigration advice and comes from a licensed adviser.
When an adviser opens a case, or takes one over from a colleague on leave, the agent turns the emails, meeting notes and files into a one-page brief.
Example: "Employer contract received; waiting for the client's bank statements. Open: two documents give different start dates. Next: adviser to call the client by Friday."
What we gain: advisers stop rereading whole cases, and less is lost when a case changes hands.
Watch out: for cases already in the Visa Platform, its own case summary covers platform data. This agent covers emails, notes and files outside it, and can start with a few chosen mailboxes and folders.
A final check that the whole application is ready, against the visa type and checklist the adviser has confirmed. The consistency check asks whether facts conflict; this asks whether anything is missing.
Example: "Submission letter refers to Appendix 7, which isn't in the bundle. Police certificate translation is unsigned. Could not check page 3 of the bank statement (unreadable)."
What we gain: a consistent last check on every file, whoever prepared it.
Watch out: "couldn't check" is always shown, so a clean result means everything was actually checked.
Helps answer potentially prejudicial information (PPI) letters and requests for further information.
Example: "INZ raises 3 concerns. Concern 2 (funds) is covered by the statements uploaded on 12 Sep. Concern 3 (job offer) has no evidence yet. Response due 7 Nov."
What we gain: lower risk of a missed deadline, and advisers start from an organised file instead of a blank page.
Watch out: an adviser confirms every deadline the agent reads, and reminders escalate to a manager if nothing has happened close to the date. The legal argument stays with the adviser.
Records the consultation (with the client's agreement), writes the notes and prepares the next steps.
Example: the quote and agreement are ready an hour after the call, not three days later.
What we gain: faster proposals win more work, and fewer leads are forgotten.
Watch out: recording needs clear consent at the start of each call.
Drafts the paperwork employer clients need for work visas: job adverts for labour market testing, job descriptions matched to the right occupation code, and accreditation and job check packs.
Example: "Offered pay is below the rate set for this role. Advert is missing the pay range."
What we gain: employer work turns around faster, which helps us grow the employer side of the business.
Watch out: thresholds and rules change often, so they come from settings our advisers maintain, not from the AI's memory.
Before the pilot starts, we record how long these tasks take today and agree a pass mark for each measure. If the pilot doesn't reach it, we stop or change course.
Time the task took before, minus the adviser time spent reviewing and fixing the agent's output.
Errors and omissions found, and how much advisers had to change.
Whether advisers keep using it after the first weeks, and which tasks they give up on.
How many enquiries book a consultation, and how many of those become paying clients.
Engineer-weeks for one full-stack engineer, including tests. These are planning ranges, not quotes. They assume we can export client files and emails from our current systems, have enough sample files to test on, and that advisers can spend time reviewing results during the build.
| Piece | Engineer-weeks | Notes |
|---|---|---|
| Shared foundation | 3–4 | Secure document store, access control, consent records, audit trail, cost limits. |
| Partnership evidence organiser | 4–6 | The largest piece, and the one with the most time to save. |
| Case consistency check | 2–3 | Reuses the evidence organiser's document reading. |
| Enquiry intake and booking (website) | 3–4 | Plus 1–2 per messaging channel later. |
| Phase 1 total | 12–17 | |
| Case briefing and handover | 2–3 | Depends on how emails and notes are stored today. |
| Pre-lodgement check | 2–3 | Needs a checklist per visa type, set by advisers. |
| INZ letter response assistant | 3–4 | Includes deadline confirmation and reminders. |
| Consultation notes and follow-up | 2–3 | Depends on how our client records are kept today. |
| Employer paperwork assistant | 2–3 | Plus adviser time to set up current thresholds. |
| Phase 2 total | 11–16 |
Running costs include AI model use, document reading, storage, messaging channel fees, maintenance and the adviser time spent reviewing. The pilot measures each one.