Bind Desk
Open prototype
Ajaia discovery engagement · Marsh & Colby

The checklist isn’t the problem. Everything around it is.

Owen asked for a faster, better-looking checklist. The interviews and the Bindwise log show the time actually goes into finding documents, catching mismatches and waiting on brokers. Bind Desk is designed around that, with a person signing off at the two moments that matter.

What the evidence shows

Where a binding's time actually goes

11 min
to locate a submission's documents
M4 · Colin
3 min
to complete the checklist once they're found
M4 · Colin
42 / 115
submissions with a coverage-limit mismatch (≈37%)
M4 · both
2.1 days
median wait for a broker reply after a flag · 9 took over 4 days
M4 · Anika
One submission, end to end
Find docs · 11 min
3m
If mismatched (~37%): waiting on the broker, median 2.1 days

Not to scale: the broker wait is days while the work is minutes. Speeding up the checklist touches the small grey slice.

Say vs. need

What people asked for isn't what they described

Anika, asked for a wish

“Probably just make the checklist faster to fill out. Some kind of smart autofill would be nice.”

Anika, two answers earlier, same interview

“Honestly the checklist itself is fine… What I actually hate is when I have to go back and forth with a broker… That can take days.”

M1
Colin says mismatches are

“not really the bottleneck for me”

The log shows

A mismatch on 19 of 54 of his submissions. For Anika, each flag starts a wait with a median of 2.1 days.

The assistants rate pain by their own minutes of effort. The log measures how long the policy waits. Both are true, and the design has to cover both.

M2 · M4
Synthesis

What the underwriting assistants actually need

N1

Every document in one place, whatever channel it came through

  • “The slow part is finding the actual documents. They come in from like four different places.” (M2)
  • 11 min locating vs 3 min checklist; 17 of 54 submissions had email or fax documents that took longer (M4)
N2

Coverage limits compared across documents before checklist work starts

  • “the coverage limit on the cert doesn't match what's on the application” (M1)
  • “maybe a third… I have to catch it or it comes back to bite us later” (M2). Log: 23/61 and 19/54 (M4)
N3

A shorter, visible loop for resolving mismatches with brokers

  • “back and forth with a broker over email… That can take days waiting on a reply.” (M1)
  • Median 2.1 days, nine over 4 days, recorded only “when logged” (M4)
N4

Keep the checklist they already know

  • “Honestly the checklist itself is fine… I know it cold.” (M1)
  • “the checklist part goes fast, I basically know it by memory” (M2)
User journey

AI prepares. The assistant decides.

  1. 01System
    Documents arrive
    Portal, broker email, fax: filed to the right submission
  2. 02System
    Extract & compare
    Coverage limits read and compared across app, certificate and limits page
  3. 03Assistant
    Queue
    Each submission's state: mismatch, missing, waiting, ready
  4. 04Human
    Checkpoint 1
    Assistant confirms, dismisses (with a reason) or corrects each mismatch
  5. 05Assistant
    Broker request
    Pre-drafted and precise. Sent from the assistant's own mailbox
  6. 06System
    Track
    Nudge at 48h, escalate at 4 days. Reply auto-attached and re-compared
  7. 07Human
    Checkpoint 2
    Bind gate: everything sourced, nothing open. A person attests and binds
Checkpoint 1 · confirm the mismatch

Before anything reaches a broker. A false flag costs a round trip with a median of 2.1 days and the broker’s goodwill. A missed one “comes back to bite us later.” Telling “small stuff” from what matters is the assistant’s judgment, and fax and email scans are where machine reading is weakest.

Dismiss needs a reason, which tunes the rules. Ignored flags stay at the top of the queue and are never auto-sent.

Checkpoint 2 · bind gate

Binding makes the policy active, and it can’t happen with a missing document or an unmatched limit. Bind stays disabled until every item is sourced and nothing is open. A person attests and binds. The tool never binds on its own.

No checkpoint on confident filing or extraction: that would bring back the 11 minutes. Those values are still visible, with sources, at the gate.

Owen's wishlist

What happened to each item, and why

Smart autofill for the checklist

Reshaped

The checklist fills itself from extracted, source-linked values: a by-product of the comparison, not a typing shortcut for a 3-minute step.

Cleaner interface than Bindwise

Baseline

Comes with any new build. No user mentioned how it looks, so it doesn't drive scope.

AI chat assistant

Not in v1

Nobody described questions they can't answer. The AI goes into filing, extraction, comparison and drafting, where the time is.

Mobile access

Deferred

No evidence of the need yet. If it means “did the broker reply?”, a read-only status view is cheap to add later.

See it working

Queue → mismatch review → broker request → bind gate. Toggle design notes for the reasoning.

Open prototype