Insurance · Demo Call
Insurance Product Demo Script: Running a Claims Demo for Carrier Buyers Who Are Looking for Where It Breaks
Your Chief Claims Officer already said yes to the premise on the first call. That is not who you're demoing to. You're demoing to the VP of IT / Head of Core Systems she dragged onto the invite, who has one question and will ask it in the first eight minutes: where does this touch the core? Behind him is the SIU Director who turned off a fraud analytics tool eighteen months ago because it buried two investigators in false positives, and a VP of Claims Operations who can quote her average days from FNOL to final payment on auto physical damage from memory — and can also quote the number the board asked about last quarter.
They are not here to learn what your product does. They are here to find out where it breaks: at the integration surface against a 20-year-old policy admin system, during a hail event that puts 4,000 claims into a queue built for 400 a week, and in front of a market conduct examiner asking who made the coverage decision. Every interruption you get is either a buying question or a disqualification test. If you run the rehearsed tour — here's the dashboard, here's the admin console — you'll hear typing, then "send me something and I'll route it to our innovation team," and the deal dies in a queue with every other vendor that ever pitched claims.
This insurance product demo script is built for that room. It assumes you're selling something that scores, triages, or routes claim files — sitting beside the PAS, not inside it. Three workflows, fully, in their vocabulary, with the gaps said out loud. Rehearse the detours, not the tour.
The demo call script
Say it in your own words. The structure is the part that matters.
- 1
Confirmation email, 48 hours out — build the demo you're actually going to give
"Two things so I don't waste the hour showing you screens that don't apply: 1. What are you running for policy admin and claims — Guidewire ClaimCenter, Duck Creek, Sapiens, a homegrown core? Which version, and is it hosted or on-prem? If it's customized, roughly how heavily? 2. Which line do you want me to build this around — personal auto, homeowners, commercial property, workers' comp? I'll load claim types, coverage codes, and loss descriptions that look like yours instead of demoing on 'Test Claimant 1.' And who's joining? If your Head of Core Systems is on, I'll have the integration diagram and the port list open and we'll do that piece first rather than making him wait forty minutes for it." [If they won't share a data extract, at minimum get: three real loss descriptions, their coverage code naming convention, and the names of two of their claim offices. Demoing on ACME Insurance Co. after they told you they're a Midwest mutual writing farm and homeowners tells them you didn't prepare.]
- 2
Opening — 90 seconds, set the interruption contract
"Last time you told me three things. Auto files that should close in twelve days are sitting at twenty-eight because the adjuster is chasing a photo estimate in one queue, an appraiser assignment in another, and a coverage question with a supervisor who's out. Your DOI complaint count is up and you can trace most of them to files over thirty days. And your leakage number is a range, not a number — somewhere in the 2 to 5 percent of paid losses you'd admit to. Still accurate, or has something moved?" [Pause. Let them correct you. The correction is the most useful thing you'll get.] "Anything change since we spoke — budget, the migration timeline, anybody new in the chair?" "Here's how I'd like to run it. I'd rather you stop me than sit through something irrelevant. If you're thinking 'that will never survive our core,' say it in the moment — that's the part of this call worth an hour. And before I share anything: what's the one thing that, if this can't do it, we don't need to keep talking?" [Whatever they name, demo it first. If the VP of IT names it, do the integration block before any workflow.]
- 3
Loop 1 — FNOL triage and cycle time (Problem → Screen → Consequence → Check)
**Problem:** "So a hail loss comes in through the call center at 4:40 on a Thursday. Right now it drops into the general property queue, sits until Monday, gets assigned by workload not by complexity, and an adjuster who handles twelve-thousand-dollar roofs gets a total loss with an ALE question on it." **Screen:** "This is the file thirty seconds after FNOL. Same loss description your intake team typed. It's scored on complexity and routed — this one goes to the desk unit because it's a single-peril roof under your STP threshold, and this one goes to a field adjuster because the loss description mentions interior water and the dwelling limit is over 600K. That's not me picking; that's the rule your claims managers set." **Consequence:** "The Monday-morning triage meeting stops being a triage meeting. The twelve-day file doesn't lose four days waiting to be looked at." **Check:** "Is that how assignment actually works in your shop, or do your supervisors hand-assign everything? Because if they hand-assign, this changes who I'd want in the room next." [Then get the number out loud:] "What's your current average from FNOL to final payment on property? And what's the number the board asked about?"
- 4
Loop 2 — leakage and ALAE per claim, shown as a moment not a metric
**Problem:** "It's the 27th. An adjuster has a soft-tissue BI file open at a 14,000 reserve, the claimant attorney wants 22, and closing it this month makes his file count. Nobody's going to audit that settlement." **Screen:** "When he moves to settle above reserve, this fires — not a block, a prompt. It surfaces the two comparable settlements in your own closed-file history for the same injury code and venue, and it flags that there's a police report indicating a third-party rear-end, which is an open subrogation lead nobody's opened a demand on." **Consequence:** "Two things land in numbers your chief actuary recognizes. ALAE per claim, because you're not sending files to defense counsel that shouldn't go. And subrogation recovery rate, because the referral happens while the file is open instead of after it's paid and closed." **Check:** "Where does subro get identified today — is that the adjuster's job or do you have a dedicated recovery unit? And be honest: what's your subro recovery rate on auto liability right now?" [If they say "we don't have a clean leakage number" — good. Say: "Nobody does. That's why I want the pilot run as a back-test on closed claims, so the number comes out of your settlements, not my assumptions."]
- 5
Loop 3 — the SIU workflow, demoed to the SIU Director, not around him
[Address him by role, directly.] "You've been burned on this before, so I want to run this part at your pace. What was the alert volume the last tool put in front of your investigators per week, and what could they actually work?" **Problem:** "By the time a file gets referred to you today, the rental's been paid for eighteen days, the medicals are in, and the adjuster has a relationship with a claimant she now has to accuse." **Screen:** "This scores at FNOL and re-scores on every new document. The reason codes are visible — prior loss at the same garaging address, treatment pattern matching a clinic already on your watch list, a repair facility that shows up on four of your last nine questionable files. It doesn't refer. It queues to you with the reasons attached." **Consequence:** "Your referral rate is 1 to 2 percent. This isn't about making that 10 percent. It's about the same two investigators seeing the file in week one instead of week seven." **Check:** "You set the threshold. If your team can work forty a week, we size the alerts to forty and you never see forty-first. Does that match how you'd want to run it, or do you want everything scored and you'll filter?"
- 6
Pressure line 1 — integration, raised before they raise it
"Let me get ahead of the question your Head of Core Systems is holding. Where does this touch the core? Read-only against the claim file. We take a nightly flat file to SFTP, or we consume off your message queue if you already publish claim events — your choice, and I'd take the flat file first because it's the boring option. We do not write back to the PAS. Ever. No schema change, no new field on the claim record, no touch to check issuance, no touch to your statutory extracts. When we have a recommendation, it lands in a work queue and your adjuster acts on it inside ClaimCenter the same way they do today. That's on purpose. What usually breaks is a vendor trying to write back into the core, and then somebody's Schedule P extract is short on a Friday." [Then ask:] "What actually broke last time? I'd rather build the pilot around that failure than around my architecture slide." [If they're on a heavily customized or 2000s-era core:] "On an on-prem core that old, you'd need a gateway inside your network for the file drop. That's an afternoon with your sysadmin. I'll send the port list and the service account permissions before we talk again so nobody's guessing."
- 7
Pressure line 2 — who owns it after you're gone
"Once we're live, who maintains this? Not IT. It's roughly two to three hours a month and it's claims work — adding adjusters, adjusting the routing rules when you re-org a unit, moving the SIU threshold. Your claims operations analyst does it. The person who maintains your current work queues is the right profile. On our side: [name] runs implementation, [name] is your ongoing contact, and it's the same person after go-live — you don't get handed off to a portal. And the honest version: if your analyst leaves, a replacement needs about a half-day of handover. The rules are in plain language, not code. I'd rather tell you that now than have you find out at renewal." [If it genuinely needs a technical owner in their environment, say it here. Discovering that at contract stage is how deals die at legal.]
- 8
Pressure line 3 — what happens when it breaks, and what happens in cat season
"I'm not going to give you an uptime percentage. Here's the failure mode. If we go dark, nothing in your claims operation stops. Files still land in ClaimCenter, adjusters still work them, checks still issue — you lose the scoring and the routing recommendation for as long as we're down, and when we come back we score the backlog. You degrade to how you operate today. That's the whole blast radius. P1 is a phone number and a human answers it. P3 is email, next business day, and I'm not going to pretend otherwise. Now the question you actually care about: cat. A named storm puts 4,000 claims into a queue built for 400 a week and you're standing up IAs at two to three times internal cost. Everything you buy gets judged on that week, not on a normal Tuesday. Volume scaling is the easy part — the part worth demoing is this: [show cat mode] your IA cohort logs in with a restricted profile, they only see files scored below the complexity threshold, and the field adjusters you trust get the severity tail. You're not paying an IA at 3x to sit on a total loss he'll take four weeks to close." [Check:] "What did your cycle time do during your last event, and how long did the complaint volume take to show up behind it?"
- 9
Pressure line 4 — time to value, in three dates against their calendar
"Three dates, not one. **Live** — first claim scored: that's the back-test, and it needs no production connection at all. You give me 18 months of closed claims on one line, we score files you already know the outcome of, and you see the precision before you've integrated anything. **Useful** — one unit stops doing the old thing: the desk unit stops the Monday triage meeting. That's the first thing your VP of Claims Operations can point at. **Fully rolled out** — all lines, all offices, SIU threshold tuned. Here's what you have to do, in hours: a closed-claims extract from your data warehouse — that's your BI team, not your core team; a user list with unit assignments; and your current routing rules, which in most shops means an hour with two claims managers and a whiteboard because they're not written down anywhere. Ninety minutes of training per adjuster cohort. And we don't cut over during your close or in the middle of cat season. When's your quiet window?" [Reference: name one real carrier from your list — line of business, approximate size, and the two dates. One sentence. Not a logo slide.]
- 10
When engagement drops — stop, don't accelerate
[Shorter answers, a delay before responses, a flat "yeah, no, that makes sense," typing, camera off. Stop.] "I've been talking for a while. Is this the part you care about, or should I jump somewhere else?" Or: "Let me stop the tour. What's the thing you're worried about that I haven't touched?" Or, best and almost never done: "Do you want to drive? Tell me what to click and I'll follow." Or, if you've genuinely built the wrong demo: "I think I'm showing you the wrong things. Can I stop, spend ten minutes understanding what actually matters, and come back Thursday with a demo built for that?" [What does not work: talking faster, adding features, "one more thing I want to show you."]
- 11
Close — say the gaps out loud, book the next thing on the call
"Let me summarize, including what didn't land. Three things worked: FNOL triage against your complexity rules, the subro flag while the file's still open, and scoring for SIU at a volume your two investigators can actually work. Two open. One — you asked whether we can score workers' comp med-only, and the answer is no, not today, and it isn't on the roadmap this year. Two — I don't know yet whether your 2018 on-prem core can publish claim events or whether we're on a nightly file, and I'll have that answered by Thursday with your Head of Core Systems on the line. Where does this sit for you now — worth pursuing, or is there something that's already ruled it out?" [Let them say no. A clean no on the demo beats six weeks of chasing.] "Who else needs to see it? I'd run a twenty-minute version for your Head of Core Systems that's architecture and security only — no workflow screens. And if your Chief Underwriting Officer cares about the severity signal feeding back into rate adequacy, that's a different twenty minutes." "Calendar's open — Thursday the 14th, 10am, or Friday morning?" [Send within 24 hours: the gap list with your answers, the four screens that mattered, the integration diagram and port list if IT was on, and a one-line list of what you need from them. Nothing generic.]
How the call actually sounds
Prospect on the left, the rep on the right.
Rep
Before I share anything — last call you said personal auto is running 26 days FNOL to final payment against a 14-day target, and your DOI complaint count is tracking with it. Still the shape of it?
Buyer
It's 28 now. And I'll save you some time — Dave from core systems is on, and we're eleven months into a Duck Creek migration. There's no integration bandwidth until 2027. So I'm not sure why we're doing this.
Rep
That's fair, and it's exactly why I'd rather have this conversation now than in 2027. Quick question for you: has your cycle time gotten better or worse since the migration kicked off?
Buyer
Worse. It was 22 when we started.
Rep
Right — that's the pattern. Core migrations run two to three years and claims service degrades all the way through, and the complaints don't wait for go-live. So here's the version of this I'd propose: nothing that competes with your migration team. Dave, does that change your posture at all, or do you still want the architecture question first?
Buyer
I want it first. Last vendor told us it was read-only too, then their connector tried to write a status back into the claim record and we spent a weekend rebuilding the statutory extract. So: where do you touch the core?
Rep
We don't. Nightly flat file to SFTP, out of your data warehouse if you'd rather not point at the core at all. No write-back, no schema change, no new field on the claim record, nothing near check issuance or your statutory extracts. When we produce a recommendation it lands in a separate work queue and your adjuster acts on it inside your system the way they do today. And for the pilot there's no production connection at all — 18 months of closed claims, scored offline, files you already know the outcome of.
Buyer
Okay. But that's the other problem. We ran a fraud model two years ago. My SIU Director had two investigators drowning in alerts — the precision was garbage, they stopped opening the queue, and we shut it off. I'm not doing that to him again.
Rep
What was the threshold set at, and what alert volume did they actually get per week?
Buyer
I don't know the threshold. The volume was somewhere around 200 a week for two people. It was absurd.
Rep
Then that was a workflow failure before it was a model failure — nobody sized it to your investigative capacity. So invert the pilot. We score last year's SIU-confirmed files and see how many the model would have caught, and critically, how many clean files it also lit up. And your SIU Director sets the volume he can work — if it's forty a week, we deploy at forty a week and he never sees the forty-first. Is he on this call?
Buyer
He is. And before he asks — we're a mutual. Surplus is policyholder money. There is no innovation budget line here, and anything with 'AI' in the name is going in front of legal and compliance because we've got a market conduct exam next year and I am not putting a black box in front of an examiner.
Rep
Then let's take the exam question first, because it's the one that ends the conversation if I get it wrong. The model recommends; the adjuster decides. It never makes a coverage determination and it never touches a denial. Every recommendation writes a reason code into the file notes, so what the examiner pulls is a claim file with a documented human decision and a visible rationale. Do your compliance and legal teams already have a model governance standard written?
Buyer
There's a draft. It came out of the underwriting side when we looked at a pricing model.
Rep
Send it to me and I'll answer against it clause by clause rather than against a generic AI objection. On the mutual point — I wouldn't frame this as a technology spend at all. It's a combined ratio conversation. Every day of cycle time shows up as a policyholder call, then a complaint, then a non-renewal your agents hear about at the next council meeting. What's the board's expense ratio target for next year, and when does your planning cycle close?
Buyer
Planning closes in October. Look — send me something and I'll get it to our innovation committee.
Rep
Happy to. Before I do — honest question. Is cycle time actually a priority for you this year, or is the real pressure coming from somewhere else, like severity on the property book or agent retention? Because if it's severity, I'm demoing the wrong three screens and I'd rather find that out now.
Buyer
It's retention. Two of our biggest independent agencies moved commercial accounts last year and both cited claims service. That's the thing my CEO asks me about.
Rep
Then let me stop the demo I built. Give me ten minutes on those two agency defections — which files, which line, how long they sat — and I'll come back Thursday with a demo built around commercial property cycle time and what your producers see, not around the SIU queue. Does Thursday at 10 work, and can we get your SIU Director back on it anyway so he's not hearing about it secondhand?
Objections you will hear
What they say, and what you say back.
| Objection | How to answer it |
|---|---|
| “We run on a 20-year-old core. Integrations die there — the last vendor broke our statutory extract.” | Agree first, because they're right and they've been burned. Then get specific about the surface: read-only against the claim file, flat file to SFTP or a message queue rather than writing to the PAS, no schema change, no new fields, nothing touching check issuance or statutory reporting. Then ask what actually broke last time — nine times out of ten a vendor tried to write back into the core. Offer the smallest possible proof: one line of business, historical closed claims, no production connection at all until they've watched it score files whose outcomes they already know. |
| “Every dollar goes through actuarial scrutiny. The ROI has to be bulletproof.” | Do not open a calculator pre-filled with your own assumptions — a chief actuary will find the assumption in ninety seconds and stop listening. Bring the model empty and let them populate it with their paid losses, their LAE ratio, and their own leakage estimate. Then ask which number their chief actuary would actually book: cycle time reduction is real but hard to reserve against, whereas a measurable drop in ALAE per claim or a documented subrogation recovery shows up in a reserve review. Offer the pilot as a back-test on closed claims so any savings are calculated against actual settlements, not projections. |
| “We tried a fraud analytics vendor two years ago. SIU drowned in false positives and we turned it off.” | Ask two questions before defending anything: what was the precision, and where was the referral threshold set? Most of these failures are threshold and workflow, not model. Then invert the pilot — don't score new claims, score last year's SIU-confirmed files and show whether it would have flagged them, and how many clean files it lit up alongside. Let the SIU Director set the alert volume his investigators can genuinely work and size the deployment to that capacity, not to the model's output. If he can work forty a week, he sees forty. |
| “We're mid-migration to a new policy admin platform. There's no IT bandwidth until 2027.” | That's a reason to talk now. Ask what their cycle time has done since the project kicked off — it has almost certainly gone backwards, and the complaints and non-renewals don't wait for go-live. Position anything you sell as sitting outside the migration path so it never competes for the same integration team, and scope the pilot so the claims org owns it end to end. If it genuinely requires their IT team, say so plainly and ask to be on the roadmap conversation for the quarter after go-live rather than pretending you're weightless. |
| “We can't have a black box making claim decisions. We'd get destroyed in a market conduct exam.” | Concede the frame immediately: the model recommends, the adjuster decides, and every recommendation writes a reason code into the file notes. Then walk them through what an examiner physically sees — the audit trail, the disparate impact testing, and the fact that it touches routing and prioritization, not coverage determinations or denials. Ask whether compliance and legal already have a model governance standard; if they do, get the document and answer against it clause by clause instead of arguing with a generic AI objection. |
| “We're a mutual. We don't have a technology budget line — surplus is policyholder money.” | Reframe it as a combined ratio conversation rather than a tech spend. Expense ratio and loss ratio come out of the same surplus, and slow claims cost them in complaint handling, agent defection, and rate adequacy they then have to claw back through a filing. Ask when the planning cycle closes and what the board's expense ratio target is for next year, then size phase one to sit under whatever threshold doesn't require board approval. |
| “Send me something and I'll route it to our innovation team.” | The innovation committee is where claims vendors go to die. Acknowledge it politely and anchor straight back to an operational number: 'Happy to — before I do, is cycle time actually a priority this year, or is the pressure coming from somewhere else, like severity or agent retention?' If they name a real pressure, book time with them and whoever owns that number. If they can't name one, you've learned this is a stall, and the right next question is who in the org actually owns claims cycle time. |
Questions reps ask about this call
- How is an insurance product demo script different from a normal SaaS demo?
Two things. First, the integration question comes first, not last — the VP of IT / Head of Core Systems will ask where you touch the policy admin system inside the first ten minutes, and if your answer is vague the rest of the demo doesn't matter. Second, the buyers have a regulatory audience you don't. A Chief Claims Officer is thinking about what a market conduct examiner sees in the file, and a Chief Underwriting Officer is thinking about rate adequacy. Build the script so the compliance answer and the architecture answer are ready before the workflow screens.
- Which metrics should I put in the demo itself?
Use theirs, not yours, and get them said out loud on the call. Average cycle time from FNOL to final payment by line is the one most claims leaders can quote from memory. ALAE per claim and claims leakage as a percentage of paid losses are the two that land in a reserve review, which is what makes them survivable in front of a chief actuary. Subrogation recovery rate, SIU referral rate, and DOI complaint count are the supporting cast. Never quote an industry average back at them — ask for their number and demo against it.
- What do I do when the SIU Director says a previous fraud vendor buried them in false positives?
Don't defend your model. Ask what the precision was and where the referral threshold was set — most of these failures are threshold and workflow, not the algorithm. Then invert the pilot: score last year's closed SIU-confirmed files and show both the catch rate and how many clean files got flagged alongside them. Let the SIU Director set the weekly alert volume his investigators can actually work and size the deployment to that number. A tool that produces forty workable alerts a week beats one that produces two hundred ignored ones.
- Should I demo at all if the carrier is mid-migration to a new core?
Yes, but change the pitch. Ask what their cycle time has done since the migration started — it usually degrades, and the complaints and non-renewals show up during the project, not after. Then scope everything to sit outside the migration path so it never competes for the same integration team, and make the pilot something the claims organisation can own without a ticket to IT. If it genuinely needs their core team, say so and ask for the roadmap conversation the quarter after go-live rather than overpromising.
- How do I show a claims product without generic sample data?
Ask in the confirmation email for a closed-claims extract on one line of business, or failing that, three real loss descriptions, their coverage code naming convention, and two of their claim office names. Demoing on 'Test Claimant 1' and 'ACME Insurance' after a VP of Claims Operations has already told you they're a mutual writing farm and homeowners in the Midwest signals you didn't prepare, and it costs you more credibility than any missing feature.
- How should I answer the reliability question from a carrier?
Not with an uptime percentage. Describe the failure mode in their operation: files still land in the claims system, adjusters still work them, checks still issue — you lose scoring and routing until service returns, then the backlog gets scored. Then tell them who physically answers the phone on a P1 and be honest that a P3 is email and next business day. And answer the cat question separately, because everything they buy is judged on the week a hail event puts 4,000 claims into a queue built for 400, not on a normal Tuesday.