A National Auto Platform Is a Provincial Rules Engine

“National auto platform” sounds more settled than it is. It usually means one portal. One account. One broker experience. One service model. All of that may be sensible. A fleet manager with units in several provinces shouldn’t learn a new interface every time a truck crosses a boundary. But the interface is the shallowest part of the obligation.

The national auto platform is not one product with provincial labels. It is a set of provincial legal regimes sharing an interface.

It becomes national at the edge only after it becomes provincial at the rule level. That is an awkward premise for an architecture review, because it kills the cleanest diagram. We want a canonical vehicle, a canonical quote, a canonical bind, a canonical claim. We want provinces to show up as a configuration object with a few rate tables and some document templates. Commercial auto does not consent to that arrangement. Coverage wording, vehicle attachment, owner liability, fleet treatment, carrier authority, rate filing, market conduct: none of them varies in the same place. They do not answer to the same regulator either. A thin provincial layer eventually normalises the wrong obligation. The expensive failure will not look like a failed deployment. It will look like a clean transaction that produces the wrong document. Or misses a required filing. Or routes a claim somewhere a province doesn’t recognise. The interface will be excellent. That will not help.

A covered auto is not a universal object anywhere in this country

The first mistake is making the vehicle record do legal work it cannot do. On the US Business Auto Coverage Form, coverage attaches through numbered symbols on the declarations, not through the vehicle schedule.

Symbol 1 is any auto for liability. Symbol 7 is described autos. Symbol 8 is hired autos. Symbol 9 is non-owned.

Physical damage can’t ride on symbol 1 or 9.

A hired unit needs symbol 8 on the physical-damage line, or the policy won’t answer the rental company’s collision bill. That alone defeats a simple data model.

Vehicle status, coverage grant, physical damage: three fields.

“This truck is covered” answers none of them.

Canada reaches a similar result through different machinery.

Provincial automobile wording is regulatory output.

Ontario uses the OAP 1 and its approved OPCF library.

Elsewhere the SPF No. 1 and SEF families apply.

Provincial Insurance Acts impose Statutory Conditions above it.

We cannot quietly rewrite standard wording because a shared service has a prettier product model.

For an Ontario fleet, the declarations describe the automobiles, and OPCF 21B can supply blanket fleet wording.

Units come on and off cover as we buy and sell them.

That is not the same mechanism as moving a unit between rows in a universal schedule.

Different control point. Different evidence. Different test.

I would make this a hard modelling boundary.

The platform may hold one fleet inventory.

It has to keep the attachment rule, form family, endorsement version, and effective period explicit beside it.

Otherwise we confuse an operational fact with a contractual one.

A truck belongs to a fleet.

That truck is covered today, for a reason.

Hired and non-owned are not a subcategory of vehicle

The next trap is ownership.

It is tempting to make it a drop-down.

Owned. Leased. Rented. Employee-owned.

Useful information.

Not the liability analysis.

Ontario’s Highway Traffic Act makes an owner liable for a driver’s negligence where the vehicle is operated with consent.

Comparable owner-liability provisions run across the provinces.

Employment law supplies another path.

An employer can answer for an employee driving a personal car on company business.

The exposure exists outside the insurance contract.

The policy is only our financial response.

So Canadian non-owned coverage sits on its own contract, the SPF No. 6 or OAP 6.

It insures third-party liability, not the borrowed vehicle.

Ontario uses OPCF 27 for that, elsewhere SEF 94.

One workflow treating an employee’s pickup, a rented truck, and an owner-operator’s unit as “non-owned” will promise the wrong thing.

The US symbol grid makes the contrast visible.

Symbol 8 covers autos leased, hired, rented, borrowed.

Symbol 9 handles autos we neither own nor hire.

Under symbol 1 both are subsumed for liability.

Our product need not imitate that grid.

But a platform serving cross-border fleet business is crossing a rulebook, not translating a vocabulary.

This is where our platform needs a refusal state.

Where the facts can’t establish whether we hired, borrowed, or merely benefited from the vehicle’s use, the system asks for evidence or refers the file.

It does not pick the nearest ownership value and issue an answer with a good user experience.

A fleet is a rating rule wearing the clothes of a master-data field

“Fleet” looks like a dimension in master data.

It is a rating rule.

The common bureau definition is five or more self-propelled autos under common ownership, though manuals and public insurers set their own counts.

Cross that line and we may move from class-and-territory treatment toward the risk’s own loss experience.

Manitoba and Saskatchewan run formal fleet safety-rating programmes that move rating with claims and violations.

The commercial mechanics are equally unromantic.

Composite rating charges a deposit premium against units, mileage, or receipts, then adjusts at audit.

Blanket coverage removes the need to report each acquisition.

Reporting conditions still govern the exposure base.

A breach is a coverage defence, not a data-quality exception.

So our “add vehicle” flow is deceptively consequential.

It updates a fleet inventory. It alters a premium exposure. It touches a blanket-coverage record. It creates a reporting obligation. It changes what we produce after a loss.

Those are not five implementations of one task.

They are five results of one event, under particular wording.

The test is correspondingly plain.

For a vehicle acquired, sold, rented, and driven on business, prove the expected attachment, exposure record, premium treatment, notice, and evidence for each provincial product.

A click-path test is not enough.

The new environment has to produce the same correct result.

A bound policy is not an authority to operate

Motor-carrier compliance breaks the fiction most cleanly.

Canadian extra-provincial carriers are licensed provincially, under legislation implementing the federal Motor Vehicle Transport Act.

They file standardised certificates provincially: a Form E liability certificate, and a cargo counterpart where cargo must be filed.

Minimum limits vary by commodity and vehicle class.

A million is typical, and dangerous goods run higher.

So policy, licence, and filing hold different states.

A platform that bound a policy has not necessarily completed the carrier’s authority to operate.

Cancellation runs its own path too.

The registry learns of a lapse when we file a cancellation notice, so its record is authoritative about cancellations and stale about everything else.

Show it, and don’t pretend to collapse it.

Policy bound. Filing generated. Filing accepted. Certificate issued. Cancellation sent.

Distinct events, distinct evidence, distinct owners.

We may show a broker one status.

We must not store one status underneath.

The same holds for certificates going to shippers and lessors.

A certificate of insurance evidences a policy.

It cannot amend one.

Additional-insured status and a subrogation waiver need an endorsement.

A screen letting somebody type a certificate holder into a template is not allowed to create a contractual right by accident.

The province does not have one regulator, and the field does not exist

The architecture also has to stop asking for “the regulator” as though it were a field.

OSFI incorporates and prudentially supervises federally regulated insurers.

It does not license the transaction of insurance in a province and has no market-conduct mandate.

A federally incorporated carrier still needs a provincial licence.

Provincial bodies handle conduct, complaints, and rates.

So one insurer answers to several authorities, all of them correct and none of them complete.

The capital record makes the distinction tangible.

Canadian property and casualty insurers are measured under the MCT regime, with a supervisory target well above the minimum.

That is a prudential fact.

It tells our platform nothing about whether an Ontario form is approved, whether a carrier filing is current, or how a claims workflow must be run.

Rate authority is separate again.

Ontario’s FSRA integrates licensing, conduct, and rate filings.

Alberta has a stand-alone Automobile Insurance Rate Board.

Nova Scotia’s Utility and Review Board and New Brunswick’s Insurance Board carry rate functions, in arrangements nothing like a national insurance department.

In British Columbia, ICBC Basic rates go to the BC Utilities Commission while optional coverages are not rate-regulated.

Manitoba’s basic Autopac rates go to its Public Utilities Board.

Saskatchewan’s Rate Review Panel recommends rather than decides.

None of this is provincial trivia.

It tells the platform where a rate filing belongs, who has authority over it, and which record proves the change was administered properly.

A rate-rule deployment isn’t done because a service returned a number.

Conduct needs its own record as well.

The Canadian conduct return is the CCIR and CISRO Annual Statement on Market Conduct, collected through a portal the AMF administers for participating regulators.

That evidence concerns how we handle business.

It does not substitute for an MCT result, a form-approval record, or a Form E filing.

Publish the map, then say how each row is implemented

The answer is a less elegant architecture.

Don’t hide provincial rules behind a canonical workflow.

Ask them to publish a province-by-province evidence map.

For every rule that changes a quote, an attachment, a document, a filing, a claim route, or an outcome: province, function, source, version, product scope, owner, downstream service.

Then name the implementation honestly.

Configuration? Code? A document library? A workflow condition? A referral?

The distinction matters because each kind fails differently.

A configuration change deploys without touching an API.

A code change keeps a field name and reverses its meaning.

A human referral gets skipped because the interface mistook an incomplete fact for a negative answer.

Every row needs a test and an evidence expectation.

A form test verifies the OAP 1, OPCF, SPF, or SEF relationship for a term.

A fleet test verifies the treatment of the fifth self-propelled auto where that threshold matters.

A non-owned test separates third-party liability under SPF No. 6 or OAP 6 from physical damage added through OPCF 27 or SEF 94.

A carrier test separates issuance from Form E and cargo status.

A rate test names FSRA, the Automobile Insurance Rate Board, or whichever body applies, rather than writing “regulator approved.”

None of this pleads for manual operations.

It is a plea for explicit ones.

AI helps here, in one narrow place.

It compares wording, classifies filings, finds inconsistencies in a fleet record.

It should not infer that an endorsement attached, that a filing is effective, or that a claims treatment applies because the answer sounds nationally plausible.

The board hears about one customer interface.

We should arrive with the better question.

Which obligations are configuration, which are code, who owns each update, and how do we prove the change didn’t alter the outcome?

A national auto platform earns its national surface only by preserving its provincial machinery.