The demonstration app

What we will build first, using invented data only

Purpose

The demonstration shows One Claim, and anyone at Specialist Risk Group who needs to approve the work, what Postbot and the medical report reader actually do, using real calls to Claude on realistic but invented material. It is also where we test the category rules with the people who triage email today, before any real data is involved.

Demonstration data

All invented. No real names, registrations, references or medical details. Every page carries a "demonstration data" banner.

Screens

1. InboxA real Microsoft 365 mailbox in a Custodia test tenant, holding the invented emails, so the audience sees Postbot work against Outlook itself. A "Run Postbot" button processes new mail live, one email at a time, so the audience can watch it work. Each email is marked as decided by rules alone, or by rules using facts read by Claude.
2. Decision viewFor one email: type, category, matched claim and summary, with the rule that made each decision. Where Claude supplied a fact, the passage it came from is highlighted in the email or attachment. Buttons to accept or correct.
3. FoldersThe mailbox after triage, with Category 1 items at the top of each folder.
4. Review queueThe emails Postbot was not sure about, each with its best guess.
5. Rule editor and dry runKnown senders, templates, reference formats, category rules and the filing table, in plain English. Change a rule, dry run it against the whole mailbox, and see every email whose outcome would change, before and after, with no call to Claude needed. Includes test cases and publishing with version history.
6. Medical reportUpload or pick a report. See the injury facts Claude extracted, each linked to its page, then the tariff or non-tariff group, tariff amount and cross-reference flags, each produced by a named rule.
7. DashboardVolumes by type and category, share decided by rules alone, Category 1 waiting time, agreement rate against a pre-labelled answer set, cost per email, and the tariff and non-tariff breakdown across the demonstration reports.

Suggested walk-through

1. Show the untouched inbox and explain the two-person process today.
2. Run Postbot. Open three decisions: a portal notification decided by rules alone, a free-written credit hire letter where Claude supplied the facts, and the court document hidden in a chaser, caught by a keyword rule.
3. Show the email that tries to instruct the bot, and that it was simply classified.
4. Correct one decision; show it recorded and reflected in the dashboard.
5. In the rule editor, make a particular broker always Category 1. Dry run it against the mailbox and show the changed emails before and after. Publish it, then roll it back.
6. Open the medical reports: one whiplash only, one mixed, then the portfolio breakdown.
7. Finish on the dashboard and the measured cost per email.

Not included in the demonstration

• Connection to One Claim's own mailbox, SiteView or the Official Injury Claim portal. The Microsoft 365 connector is the same code; only the tenant and mailbox change.
• Any real personal or medical data
• Microsoft 365 sign-in (a simple demonstration login is used instead)
• Sending email of any kind

The demonstration is done when

• All seven screens work end to end with live calls to Claude
• Every email in the set gets a decision, with the rule that made it and evidence for any fact from Claude, or goes to review
• No email that the rules can settle is sent to Claude
• Every Category 1 email in the answer set is marked Category 1
• The bot-instruction email has no effect beyond being classified
• A dry run makes no change to the mailbox (checked by comparing the mailbox before and after)
• A rule change that breaks a test case cannot be published
• Every extracted medical fact links to the right page
• The API key cannot be seen from the browser, and the site is password protected
• It works on a laptop and a tablet, and a One Claim user can run it without us