Leadership agrees on the goal and disagrees on why existing approaches fall short. This takes one position on that, from a small amount of direct evidence, and says plainly what it is not claiming. One moment, one reason, one place to start.
Before you decide
three things
The day after
everything
Nine September · one day of field work · six guesses written down before I spoke to anyone
I went where the decision leaves a trace, not where it is visible.
You cannot watch someone buy insurance. The decision happens in a household, in minutes, and the people in it cannot reliably tell you afterwards what moved them. So I went after the places the decision leaves a record: the screens it is made against, the accounts that logged it, and the people who had just been through it.
Click a row for what it was chosen to answer.
The only person who can say what the relationship looks like at the moment it is actually tested. Everyone else in this record describes buying; he describes what buying turns out to have bought.
Someone who actually moves. The question I wanted answered was the threshold: what does the money have to be before she will? The answer came back with a trigger attached that was not money at all.
Looks every six months and almost never buys. This is exactly the population the brief is about, and he is the person §02 either serves or fails.
The one channel none of the screens cover. I wanted to know whether an intermediary dissolves the problem or relocates it. He turned out to be the strongest counter-case in the record, and he is quoted against me in §02.
What the record says, set against what the customer remembers. These two do not agree, and the disagreement is where the finding came from.
One subject appears here twice — by recollection and by record. That is one source, not two, and it is counted once.
The three insurance websites were not just read. They were driven end to end by an agent, running scenarios, pulling quotes, with every screen captured. That is why the record of what a buyer is shown is complete rather than remembered — and it is repeatable, which a recollection is not.
It also went against me, which is the reason it is worth telling. I had written that the pages you see before buying keep the important numbers vague and the trivial ones exact. Then I got inside an authenticated account and the coverages screen was genuinely good: every coverage, its limit, its own price. My assumption was wrong, and correcting it is what turned a vague claim about opacity into the actual finding in §02 — which is not about quality at all.
Written 5–9 September, before any contact. Each carries the condition that would kill it. Reproduced unedited, including the ones that did not survive.
| The guess | Outcome | |
|---|---|---|
| H1 | People shop for what they want, not what they need. Insurance is bought under obligation, so normal shopping economics may not apply. | — |
| H2 | Abandoned quotes are not undecided people. They already bought — elsewhere. | CONFIRMED |
| H3 | Switching risks coverage you cannot verify. The fear is not price; it is a gap you discover at claim time. | CONFIRMED |
| H4 | Loyalty is believed to pay, and history does not port. | DEAD |
| H5 | Below a savings threshold, intangibles win. | CONFIRMED · $100/mo |
| H6 | Shopping is event-triggered, never calendar-triggered. | SPLIT |
The point of writing kill conditions first is that a guess you can talk yourself out of afterwards was never a guess. H4 died because I checked my own facts before using them — I believed a household had run driving-tracker-priced insurance and left anyway. They never had it. They were quoted, it came in higher, and they stayed put.
One day of field work, and every line of it is counted and tagged in the record — including the two I contaminated myself, and three errors I found in this brief the night before presenting it.
The insight I believe matters most · evidence under every line
They tell you everything. Just not until after you have bought.
Insurers explain things thoroughly, and almost entirely to people who have already bought — so the buyer has strictly less information at the moment of decision than they will have five minutes afterwards. They are not hiding it. Once you are a customer it is close to excellent: every coverage priced line by line, every email, fee and price change logged and dated. All of that is switched on the day after you decide.
Before you decide: “up to your policy’s limits,” only the obvious exclusions, and the nine-question coverage tool parked on a marketing page outside the quote. The decision is made in the one place the information is absent.
By the time the information exists, the decision has stopped being asked. A renewal arrives as an invoice, not a question. The insurer’s own forward calendar says “nothing scheduled” eighty-five days out. The policy summary says “this coverage summary replaces your prior one” and shows no diff. A premium moves eleven hundred dollars and is logged as “Policy change.”
Everything is published. Nothing is explained.
Source: two authenticated insurer accounts read end to end, one household, plus one paper renewal invoice. Full reads in the record.
A second insurer’s log shows seven escalating, timed, multi-channel contacts to recover a few hundred dollars, followed by referral to a collections vendor. In the same account, over the same months: nothing explaining coverage, nothing explaining a premium change, nothing announcing a renewal decision.
They explain brilliantly where explaining changes what you do for them, and not at all where it would change what you buy.
Those counts are the whole of the evidence for this claim. A driving-tracker comparison stood here and was withdrawn because it traced to no observation; it is in the list of things I got wrong.
The obvious answer is friction — make quoting easier. But the people I spoke to did not describe friction. One switched insurers in a day. One has looked every six months for years without moving.
The obvious answer is also transparency, and transparency is already solved. It is simply delivered to the wrong person at the wrong time. Harbor does not need to say more. It needs to be the first insurer that says it while it still matters.
Four people; three have moved. Every one of those moves had a trigger that was not price: a denied claim · twin teen drivers and a broker changing employers · a new vehicle. And price was named by all three, and was the entire reason for one of them.
So price is not the rival explanation to this brief. It is the thing that makes people look, and the thing they settle on once something else has put them in motion. What the finding explains is the gap between those two — why a man who checks the price every six months, for years, has never once moved on price. He can compare the number. He cannot compare what it buys.
The gap is real, and somebody already sells the answer to it: a broker. One buyer here routes six policies across three households through a single person “because we just trust them,” cannot compare two policies without them, and changed insurers when that person changed employers. That is the insurer you already have answer to everything above. It is human, it is sticky, and it beat a insurer relationship in a straight contest.
Two things follow, and both limit what I can propose. What Harbor builds has to be better than a phone call to someone you trust — or aimed squarely at people who do not have one. And it has to survive that buyer’s own objection, stated plainly:
“You’re missing the interpersonal.”The brokered buyer, on §03
One company · the first move that earns it · the test that would kill it
Insurance has no memory that belongs to you.
Every document you get replaces the one before it. Your policy summary says so in as many words — “this coverage summary replaces your prior one.” What a claim decision actually meant disappears inside the insurer. Change company and the context is gone. You can pay the same insurer for eleven years and own nothing that proves what you were covered for in year three.
That is the company I would build. A record of your own coverage that survives, takes on every change as it happens, and travels with you when you leave. Nobody sells it, because every company in this market is built to keep it — and because the moment you own your record, switching stops being frightening.
And this is the part that changes what Harbor should be. A record that helps you leave cannot be owned by a company that needs you to stay. That is the argument against every incumbent on this page — and it would apply to Harbor too, the day Harbor had a book to protect.
So my answer is not a feature for a carrier. It is that Harbor should not try to be another carrier. It should be the layer that makes coverage portable — which is what the insight actually asks for, and the one position in this market not already held by a company whose incentive is to keep you.
And it is buildable now, which is the part that is new. Consumer-permissioned insurance rails already return structured policy data from most of the US market, off a sign-in the customer performs themselves. The plumbing this would have needed five years ago exists and is sold commercially — named, counted and sourced in the landscape scan, along with who buys it today. What has never been built is the version that points the answer back at the person who signed in.
The questions anyone would ask about a company like that. The last three are not answers — one is a better idea than mine, one is what I would go and find out next, and one I cannot resolve at all.
The reason this is fundable at all is that somebody profits when you leave. In insurance there is always a company on the other side of the switch who wants it to happen — and they are already spending the money.
Not the buyer, though. Nobody in my record asked for a record they could carry, and people do not pay a subscription to hold a document.
The carrier receiving the switch pays — and that money already exists. From the man who has shopped every six months for years:
“Nowadays they can put your name in, they can see what coverages you have. I get texts and emails telling me they can cover my house and my vehicles and I’d save X a year with the same coverage I got now, because they know what your coverage is.”
Insurers are already paying to rebuild your coverage and pitch you with it. This does not invent a new cost. It points one that exists the other way round.
And what they buy today is worse. The comparison sites are paid “when you purchase a policy or get a quote through us” — and asked whether the price they show is the real one, their own terms say “It depends!” Unverified leads, most of which never turn into anything: three separate people in my record never even declined a quote, they simply stopped answering.
Someone who arrives holding a structured record of what they have is cheaper to quote and likelier to bind, and the record is the thing doing the verifying. You would be selling a verified switch instead of an unverified lead.
The shape is not new, and the plumbing is not missing either. Credit Karma hands you your own credit file for nothing and is paid by the lender receiving a pre-qualified applicant. Plaid is paid by the company on the receiving end, never by the person connecting their own bank. Consumer-permissioned insurance rails already do the equivalent across most of the US market.
So the payer is not the innovation, and I want to be exact about that. Those same rails are sold today to agents, lenders, mortgage originators, dealerships and carriers — the same side of the table that would pay here. The difference is which way the answer points. Today the customer signs in and the data goes to the seller, who quotes or underwrites with it; the customer never sees a comparison. Nobody has built the version that hands the answer back — and lets it say nothing here needs to change.
A record you can carry makes leaving easy, and every insurer with customers depends on leaving being hard. Shipping it would hand their own customers the door. That is subtraction, not cowardice — and it is why something this obvious does not exist.
The same arithmetic stops the comparison sites, one rung out. They are paid, in their own words, “when you purchase a policy or get a quote through us.” The quote is the product. A comparison allowed to come back nothing here needs to change does not merely fail to earn — it destroys the transaction they are paid for. They would have to build something whose best day is their worst one.
Which is why the version that already exists is the version nobody trusts. The man in this record has been checking those sites every six months for years and has never once moved. If a comparison site were the answer, he would have switched in 2019.
Harbor can build it because Harbor has nothing to lose when you leave. And if Harbor stays the portability layer rather than becoming another carrier, that never changes. An incumbent shipping this would be handing its own customers the door; the day Harbor had a book of its own, the same subtraction would be Harbor’s. So this is not a window that closes — it is the one durable position in this market, and it exists only for a company that does not need you to stay.
Nothing stops the build. Four weeks, two people. A site with real traffic would do it faster than Harbor and reach more people doing it. What they cannot shortcut is what the thing collects.
Every policy that comes in carries words the comparison cannot place — not edge cases, constantly. Loss of Use. OEM Parts Endorsement. Diminishing Deductible. And form numbers with no plain meaning at all: one of them, A264, appears thirty-one times in a single household’s documents and nobody can read it, Harbor included.
Each one gets named to the customer, queued with the insurer it came from, and ruled on. After a year that is a dictionary of what each insurer actually prints and what it means, built from real documents and pointed at the buyer. It is only copyable by doing the same work on the same volume of real policies — and the people best placed to do that are the ones whose incentive is to look complete rather than to publish what they could not read. The list of things it refused to compare is a liability for them. Here it is the asset.
And it compounds. Every change it captures makes the record more complete; a more complete record is worth more per switch; more insurers accepting it makes it more worth carrying. One comparison is an event. A record is a position.
⚠ The honest limit: this is extrapolated from one household’s documents — five unplaceable terms and a form number. I have the shape of the asset and not its size, and I would not present it as more than that.
If you already have a person who explains things to you, the cleaner fix is not a screen. It is to pay them for the answer that loses the sale — pay the broker when they tell you to stay where you are. Existing distribution, no cold start, and it keeps the one thing everyone in my record actually valued: a person.
I do not have a funding model for it. Somebody has to pay for a transaction that does not happen, and the only person who benefits is the customer, who will not pay. But as a mechanism it is cleaner than mine and I did not arrive at it on my own.
What it does not do is reach anybody without a broker. Three of my four buy direct, off a screen, with nobody in the room at all — and for them, fixing a broker’s incentive changes nothing. That is not a better idea than this one. It is a better idea for different people, and both can be true.
Everything above assumes nobody explains things at the buying moment because of how the work is pointed. There is a plainer possibility: nobody explains because explaining is a person’s time, and a person’s time runs out.
The adjuster said it without being asked. Do the promised twenty-four-hour callbacks actually happen?
“Depends how much stuff the adjuster got going on.”
Service quality is capped by how many people one person is carrying. And a broker is only worth having if your premium is big enough to be worth someone’s afternoon. Which would mean buying off a screen with nobody in the room is not a preference. It is being priced out of advice.
So the question is whether the ceiling moves. If one person can carry ten times the households without the service degrading, the floor drops, and people who could never justify a broker get one. That is a larger idea than anything else on this page and I have not tested a single part of it.
And here is the fact that stops me claiming it. The paper bill in my record carries a named local agent, with a photograph and a direct line. That man has a person. He still shops every six months, still cannot compare two policies, and has still never moved. Having someone available did not produce the explaining.
So capacity is not the whole of it. Capacity and reason-to-bother are both holding this shut, and more capacity only opens one of them. Which one actually binds is the thing I would go and find out next — and it is the same conversation as the hour I already owe: ask someone who runs an insurance operation what a person costs, how many households they carry, and what happens to the answer when they carry more.
And a second one, which came out of the session rather than the field work. The adjuster, asked whether the incentives could be better for buyers, proposed an insurer whose members share in the dividend. Checked afterwards: the mechanism is real but the vehicle is mutual insurers, not credit unions — State Farm Mutual paid a $5bn dividend to auto customers this year, about $100 a vehicle. And the check cuts against the idea. State Farm is a mutual, its policyholders are its owners, and it still shows you no comparison before you buy. Aligned ownership did not produce legibility. The dividend is also about $100 a year against the $100 a month this record found as a switching threshold — an order of magnitude short of moving anyone. So the question I would take into that same hour is whether any incentive changes buying behavior, or whether only legibility does.
Worth saying plainly: if the reason nobody explains is cost rather than choice, then the alternative I flagged as most likely to sink this brief is right — and it stops being a threat and becomes the opening. I would rather write that down now than discover it later and pretend I had meant it all along.
If insurers pay for it, whose side is the record on? The whole complaint in this brief is that the industry aims its ability to explain things outbound. A record paid for by insurers could rebuild exactly that with a nicer interface. I do not have an answer to this.
The nearest thing I have is a mechanism I want rather than one I have seen: the record is only worth paying for because the person carrying it trusts it, so the moment it starts serving the payer it stops being worth buying. That is an argument, not evidence.
And I have no sizing. Nothing I did touches what any insurer would actually pay. Any number here would be invented, so there is not one.
One conversation settles this and the biggest open question in §02 at the same time — an insurance operations lead, asked what a verified switch is worth against an unverified lead, and whether the reason nobody explains coverage at the buying moment is cost or appetite. One hour. Cheaper than most of what I did spend, and I did not spend it.
Not one person described wanting a record they could carry. The only thing anybody named as worth having was a person who explains things to them. So the position is a claim, and I am proposing it rather than asserting the demand — which is why what follows is the smallest slice of it that can be built and killed inside thirty days.
Show people the difference before they buy.
Before someone buys, Harbor puts the policy they already have and the quote in front of them side by side, line by line — in their own numbers, and saying plainly what it could not compare. And it is allowed to come back with nothing here needs to change.
The insurer you already have will never show you this, for the same reason nobody else will: a comparison is only worth anything if it can come out against whoever is showing it to you.
See the three working screens →
Including the one that tells you to stay where you are.
The moment of buying: after a price is on screen, before payment. Not the quote request, which is the link Harbor already owns and every competitor is also optimizing.
§02 says everything you only get told afterwards is switched on the day after you decide. This is the narrowest possible correction: move everything you only get told afterwards in front of the decision — applied to the one comparison two independent sources say nobody can perform unaided.
And the capability is already built. Insurers line your coverage up today in order to poach you: “they know what your coverage is.” Nobody points it at the buyer so the buyer can check.
Harbor’s own framing says the goal is customers who feel confident, not just informed — and that distinction is right. More information at the decision moment would be a worse version of this.
A diff is not more information. It is a verdict on information the shopper already has, and it is the only output in the flow that can come back nothing here needs to change. That sentence produces confidence; a longer explanation produces the opposite.
Which is why this is aimed at binding rather than shopping. The man in my record who checks the price every six months, for years, is short of neither information nor intent. He can compare the number. He cannot compare what it buys — so he shops, and he does not bind.
All of it. If buyers already had everything you only get told afterwards at the decision moment, a before-you-buy comparison is redundant. It has no reason to exist except this back-to-front order, and it dies the day §02 does.
Connect the policy you already have. Consumer-permissioned insurance rails already exist: a shopper signs in to their own carrier and structured policy data comes back in under a minute, across the large majority of the US market. Photographing the declarations page is the fallback, not the path. Either way it is data the customer already owns — nothing a new entrant lacks. Line it up against the six standard car coverages; anything outside them is refused by name, in the buyer’s own printed term, never silently dropped. Render the difference and the refusals. The first batch is done by hand before anything is automated.
One engineer, one designer, roughly four weeks, under one named owner — including the times it says nothing needs to change. It is the first testable slice of the platform, not a feature bolted onto somebody else’s quote flow. estimate
First, before the test that would prove me wrong: will a shopper hand over the policy summary at all? Nothing in my record says anyone will, and nothing I did measured it. Measure upload rate before measuring anything else. Below a rate that could reach a usable sample inside the window, the idea is dead regardless of whether explaining things works.
Then the test that would prove me wrong. Among shoppers shown a difference big enough to matter, compare against a no-diff control. Always say how many people it was out of, and how wide the margin is — never a bare number. If the intervals overlap, it did not work. Below roughly sixty-two qualifying shoppers it reports did not run, which is neither pass nor fail and is itself a finding about Harbor’s traffic. There is no threshold, because none is defensible — nothing in the record says what share of shoppers change a selection normally.
This screen succeeds by losing sales, and nothing in the design protects it. The test counts people who walked away as evidence it is working. The quality bar requires it to tell people to stay put. The working rules then forbid following up on the people it told to stay. So its best day looks exactly like a conversion problem, and the first growth review with a quarterly number will find it.
Two defences exist and both are weak: commit to the renewal diff in writing before Harbor has a book, when the promise costs nothing; and count the null results in production so their disappearance is detectable. Those are alarms, not safeguards. They tell you it was killed; they do not stop the killing. I could not resolve this and I am not going to present it as resolved.
It cannot reach exclusions or endorsements. Those live in the policy form, not the policy summary — “modified by forms 4884, Z357, A239 AR, A264 and A331.” The tree question is a form question and this does not answer it. So the diff ends by naming the endorsements it could not read, which turns a form number from noise into a countable gap. Saying plainly what it could not compare is the useful part. Pretending it compared everything is what the industry does now.
And it does not beat a broker. This is for people who do not have one. A scope, stated — not a market I am pretending to win.
Four explanations survive my own evidence. Each one, if true, costs this brief something specific — and each is a thing that could be checked rather than a caveat.
Chasing money owed is off-the-shelf: standard, bought in, an outside firm at the end of it, and no license needed. Explaining coverage has to be done case by case, changes from state to state, and sits right next to a line you need a license to cross. That asymmetry predicts exactly what I observed, without anyone choosing anything. If that is right, “being new is the asset” is much weaker.
What the record offers against it: the ladder is not only bought collections. The same log shows automated cancellation notices, instant rescission on payment, and a payment restriction placed on a bank — machinery reacting to policy state within minutes, bespoke to the policy, and not a vendor product. Weakened, not eliminated.
⚠ This is the one I would bet against myself on. One conversation settles it — an insurance operations or customer lead, asked what a comparison at the moment of buying would cost them and whether the reason they have not shipped one is cost or appetite. One hour, cheaper than most of what I did spend, and I did not spend it.
Recommending a coverage level to a named person is a licensed activity. If asking those nine questions inside the buying flow crosses that line, both insurers were constrained, not coy.
What the record offers against it: both insurers publish the tool openly to anyone. Only its placement differs. A licensing constraint on the questions themselves would bar the tool, not relocate it — which fits a conversion decision better than a compliance one. Still open, and it is one call to an insurance lawyer. I have not made that call and will not assert the strong version until someone does.
Everything here shows that nobody explains it. Nothing here shows that explaining it would move anyone. This is the one thing everything else rests on, and nothing I did tests it. It is not a gap in the evidence — it is the thing this whole idea exists to put at risk. Anything you build whose test does not go after that possibility is testing something easier than the claim.
People do not fail to switch because they cannot compare. They decline to spend the effort comparing, because the expected gain does not justify it — and they hold just enough confidence to stop looking. Put to me plainly:
“Maybe it’s not 100% trust, but it’s not worth the effort… I trust my agent enough to think, okay, I’m probably covered.”
Why this one is more dangerous than the last. The one above says the idea might not work. This one says it is pointed the wrong way — because §03 asks a person to go and find a document and upload it. If the reason they stay is that it is not worth the effort, I have added effort to the moment I am trying to unblock.
And this is where the rails matter most. The objection assumes the ask is an afternoon. It is not: consumer-permissioned insurance APIs already return structured policy data from most of the US market in under a minute, off a sign-in the shopper does once. The diff does not add effort — it replaces an afternoon nobody was going to spend with a login. Whether anyone accepts even that trade is what the gate measures, and it is why connect rate comes before everything else.
Choosing one moment means declining several — so the choice is visible rather than implied. Pricing, regulation and advertising are not on this list: the brief ruled those out. These three are mine.
| Not addressed | Why not |
|---|---|
| The quote experience itself | The link Harbor already owns and the one every competitor is also optimizing. Improving it does not address what happens after. |
| Home insurance, if this turns out to be about cars only | Home insurance is usually paid through your mortgage, which makes it work differently. A finding that only holds for one line should say so rather than be stretched to cover both. |
| The agent and broker channel | Every screen I measured is direct-to-consumer. One of four people sees none of them. A scope, not a universal. |
Of the people who run a comparison and do not buy, how many come back to it on their own? A comparison people return to is already behaving like a record they own.
If nobody comes back, then a record you can carry is a thing I find appealing and customers do not — and it stays out of the product. That is the whole test, and it runs for nothing alongside the first move. I would rather find out that way than argue for it.
The natural experiment already exists, and I have one person standing in it. Brokered buyers already have the thing §02 says is missing — a person who performs the comparison, at the decision moment, in their interest. If people who have someone to explain it behave the same as people who do not, then explaining is not what changes buying.
The test: ten brokered buyers and ten direct buyers, same question set. Do the brokered ones decline the quotes they do not take? Can they say what their policy excludes? Did they move for a reason other than price? If the two groups look alike, this brief is describing something real and inert.
⚠ And the record already holds one point against me. The one person here who uses a broker already has someone to explain it — and still names price first, still never declined the quote he did not take, and switched insurers because his agent changed employers rather than because anything was explained to him. One person is not a refutation. It is exactly where I would pull.