Claimants start the online loss report and phone in anyway

You can see digital starts and telephone completions for the same loss, and the conclusion in the room is usually that claimants prefer to talk to a person. That may be true for some losses. It is not what the data says until someone has looked, and the looking is a specific query: a step-level drop table, segmented by device and by whether the reporter is the insured, joined to inbound calls from the same number or policy within the following 60 minutes.

That join is what separates 2 different events wearing the same appearance. A start with no completion and no call is an abandonment — the claimant went away. A start followed by a call 8 minutes later is a fallback — the claimant still wanted to report the loss and your form stopped them. Those need opposite responses, and a funnel report without the call join cannot tell them apart.

Build the drop table before anyone redesigns a screen

1. Instrument every step boundary, not every field. Started, identity passed, loss facts complete, parties complete, photos attempted, photos complete, submitted. 7 events give you a readable funnel; 70 field-level events give you a data project.
2. Segment by device class. Mobile and desktop behave so differently on upload and on long forms that a blended figure hides the entire finding.
3. Segment by reporter type. The insured, a named driver, a broker and a third-party claimant arrive with different information and different rights, and lumping them together produces an average nobody experiences.
4. Join to inbound calls within 60 minutes on the policy identifier or the calling number. Anything unmatched is genuine abandonment; anything matched is fallback, and the step it fell back from is your answer.
5. Compute time-in-step for the sessions that dropped. A step people leave after 4 seconds is a wall. A step people leave after 4 minutes is a task they could not finish, which is a different fix.

One quarter of data is enough. The pattern is stable, because the causes are structural rather than seasonal — with the exception of a catastrophe week, which should be excluded from the baseline and analysed on its own.

Where a first-notice funnel actually leaks

The last row is worth dwelling on because it is invisible in a conventional funnel: the form completed successfully and the claimant rang anyway. That is a confirmation problem, not a form problem. A reference number on screen, an immediate message containing it, and a plain statement of what happens next will remove most of those calls without touching a single field.

Identity before value, and the wall it builds

Putting authentication at the front of a loss report is the most expensive sequencing decision available, and it is usually made for administrative rather than risk reasons. The person reporting a loss is frequently not the person with the login: a spouse, an office manager, a driver, or a third-party claimant who has no relationship with you at all and never will have a portal account. For that last group the identity step is not friction, it is a closed door — which is why third-party notice needs its own path with its own identity and disclosure boundaries, as set out in taking a loss notice from someone who is not your insured.

• Collect the loss facts first, verify after. The facts are perishable and the verification is not; a partially verified notice is worth vastly more than a verified silence.
• Use a tokenised link rather than an account wherever the reporter is not the policyholder. A single-use link to a specific claim gives access to exactly one thing and needs no password reset flow.
• Never require a document to start. Asking for a policy number from someone standing beside a damaged vehicle is asking them to go and find a filing cabinet.
• Let the reporter say who they are. Insured, driver, broker, third party — capture it early, use it to branch, and use it in your segmentation later.

Fields nobody can answer standing on a hard shoulder

Read your own form as somebody who has just had an accident. Deductible amount. Coverage type. Policy effective date. The other driver's insurer. Each is a question the claimant either cannot answer or would have to look up, and each is available to you, from your own records, the moment you know who they are. A form that asks its user for data it already holds is not collecting information; it is transferring work, and the user transfers it back by phoning.

The discipline is to reduce the form to what only the reporter knows: what happened, when, where, who was involved, what condition things are in now, and how to reach them. Everything else is enrichment you perform afterwards. Deciding exactly which fields make a notice workable — rather than which fields would be nice to have — is the substance of what a loss notice must carry to become a file.

A claimant who starts online and finishes on the phone has told you precisely which step is broken. They have done the analysis for you; the only failure would be recording it as a channel preference.

No save and resume means every interruption is a phone call

First notice is reported from places where sessions do not survive: a roadside, a car park, a kitchen with 2 bars of signal and a child in the background. Without save and resume, an interruption destroys the work and the claimant will not start again — they will ring, because ringing has a person who remembers what was already said. Persist server-side at every step boundary, key the draft to the loss rather than to a login, send a resume link, and let a handler pick up the same draft on the phone. Capture that tolerates poor connectivity is a solved problem, treated in field capture that survives no signal.

The losses that should go to voice on purpose

Some losses are the wrong shape for a form and no amount of design fixes that. A house fire, a serious injury, a fatality, a theft discovered on returning home: the reporter is in no state to enumerate structured facts, and asking them to is both ineffective and unkind. The right response is not a better form but a deliberate rule routing those loss types straight to a person, with the digital path offering one obvious control that connects the call and carries whatever was already entered.

1. Name the loss types that route to voice, in writing, with the operational reason for each. This is a policy decision, not a design preference.
2. Express it as a rule the intake system evaluates and logs, so you can see later how often it fired and whether it was right. Whether such decisions should sit in rules or in a model is argued separately in model scoring or written rules for claim triage.
3. Carry the partial notice into the call. The handler should open the file already containing whatever the claimant typed, or the routing has cost the claimant time rather than saved it.
4. Measure those routes separately. Their abandonment figures are not comparable with the rest of the funnel and blending them corrupts your baseline.
5. Review the list annually against actual outcomes, because loss types migrate as the book and the product mix change.

Which channel should take first notice by default is a broader question, turning on line of business, reporter type and whether the loss facts are enumerable at all — argued in guided form or voice capture for first notice. This page settles something narrower: whether your form is failing for reasons you can fix.

Changing a live intake flow without breaking the files in it

Every change here touches claims being reported right now, so the release process is part of the design. Change one step at a time and hold the drop table as the measure; 3 simultaneous changes produce a number nobody can attribute. Keep the old path available while the new one proves itself. And where any part of the flow — routing, triage, prefill — is model-driven, changes to it need a review path of their own, which is approving a change to a model that touches live files.

The work itself is mostly instrumentation and sequencing rather than a rebuild: step events, a call join, resume drafts, on-device image resizing, and a rule that sends the hard losses to a person. That is characteristic internal tools and ops scope, and it usually moves completion further than a visual redesign does. The rest of the ground between a loss happening and a file reaching a queue sits in claims intake, FNOL and triage, inside our insurance and claims work.

• FNOL
• digital intake
• funnel analysis
• claims experience