The model has been declared a disappointment, and the verdict is being drawn from the wrong evidence. Agency was designed carefully, launched confidently, and then defended past the point the evidence supported it. What failed wasn't the distribution model. It was the refusal to change it once it was live — and the brands now deciding whether to pause, proceed or reverse are making that call on a misdiagnosis.
A third of dealers in Startline's tracker say agency hasn't worked for most of them. Roughly the same proportion believe manufacturers are quietly walking away from their plans. The 2026 outlooks describe the volatility of the trade partly through what they call the stalling of the agency model. Launches have slipped. At least one brand reversed out altogether.
Read that as a verdict on the model and you reach an obvious conclusion: agency was a theory that didn't survive contact with a forecourt.
I don't think that's what happened. Every one of those data points is a measure of execution. Every one of them has been recorded against design. That is the same accounting error the network makes internally every week — the delivery delay caused by sales landing on the workshop, the survey miss caused at booking landing on the advisor — only this time it is being made at network level, about a decision worth hundreds of millions.
What failed wasn't the distribution model. It was the refusal to change it once it was live.The distinction the pause-or-proceed decision now rests on
Mercedes-Benz is the exception in almost every piece written on this. Agency landed there. It has held. It is generally accepted, including by the retailers operating under it, as having worked.
The interesting question is why — and the answers usually offered don't hold up. That Mercedes went first is not an advantage; going first means nobody has a template and every fault is found the expensive way. That the premium positioning carried it explains nothing about order-to-delivery, invoicing or handover. What actually distinguished it was less glamorous, and it came down to three decisions.
Not a pilot on selected models. Not a phased introduction by segment. Not a period of running the new mix half one way and half the other. The whole new car range moved, and it moved on terms that were enforced rather than encouraged.
Certification carried a pass requirement. People had to demonstrate they could do the job under the new model, not simply attend a session about it. That sounds like an HR detail. It is the difference between a network that can transact on day one and a network that has been informed about transacting.
New moved entirely to agency: new process, new systems, new customer journey. Used stayed exactly as it was, retailer-owned, operating as it always had. One line, drawn where the business was already divided — and easy to hold in your head on a Tuesday morning with a customer in front of you.
The retailer kept a business they already knew how to run while they learned the one they didn't. And where the two met — part-exchange, finance, handover, the customer buying one and selling the other in the same conversation — there were defined processes for those specific elements rather than an assumption that people would work it out. That is what let sites run an agency new car profile alongside a franchise-led used operation without the join showing to the customer.
Not because a stage gate said it was time to review them. The people closest to the transaction were listened to when they said a process wouldn't land, and they were listened to early enough for the change to be cheap.
Those three look contradictory, and the tension in them is the whole lesson. Be rigid about scope and standard. Be flexible about process. Most struggling implementations got that exactly the wrong way round: they negotiated the scope, softened the training requirement to get the network over the line, and then defended the process design long after it had stopped working.
None of this is a property of the agency model. It is a property of how that programme was run — which means it is available to any brand that wants it, and it is exactly what has been absent everywhere the model has struggled.
Most other brands drew their line somewhere far more damaging. They put BEV on agency and left ICE on the franchise model, systems and processes untouched. So the split didn't fall between new and used, where the business was already separated. It ran straight through the middle of the new car floor, by powertrain.
Consider what that does mid-deal. A customer comes in for a new ICE car and then, entirely normally, starts asking about the electric equivalent. Under a powertrain split that isn't a change of model — it's a change of commercial model. The deal has to be set up again on a different system, under different processes, with different pricing authority and different paperwork. It is genuinely complex for the person doing it, which means it looks considerably worse to the customer watching them do it.
Nobody experiences that as a distribution model in transition. They experience it as a retailer who can't seem to sell them a car.
Which produces the most self-defeating outcome in the whole exercise. The strategic case for putting BEV on agency was to support the transition to electric. The effect of putting only BEV on agency was to make the ICE-to-BEV switch the most awkward transaction in the building — the exact conversion the mandate is measured on, made harder by the design intended to promote it.
And neither model gets learned properly under those conditions. Every error is attributable to confusion rather than to design, which means the programme never finds out what is actually wrong with it. One foot in and one foot out is the most expensive position available.
The effect of putting only BEV on agency was to make the ICE-to-BEV switch the most awkward transaction in the building.The exact conversion the mandate is measured on
Agency is usually presented as a commercial redesign: the OEM takes title, sets price, owns the customer, pays the retailer a handling fee. All of that is contractual, and all of it can be agreed in a room.
The implementation is a different animal. It is systems, process, order-to-delivery, invoicing, handover, part-exchange, finance, compliance, authentication, access levels, demonstrator rules, and the several hundred small operational habits a site uses to transact a car. Every one has to be rebuilt. Every one has a failure mode that only appears under live volume. Three consistently get underestimated.
If the DMS doesn't talk properly to the OEM's order and invoicing environment, nothing else you have designed matters, because the site cannot transact. This is knowable early and testable early. On the programmes I've delivered, the single highest-value decision was trialling integration well ahead of go-live rather than trusting it would be ready — and the point was never confidence. It was finding what was broken while there was still time to fix it cheaply.
If the handling fee and the incentive structure don't reflect what the job now actually takes on the floor, you have designed a model your own network is motivated to work around. That surfaces within weeks. It is rarely reported upward within months.
Telling a network what's changing is not the same as equipping it to do the new thing on a Tuesday morning with a customer in front of them. Go-live comms, video training, process maps, crib sheets, guides — that material is what converts a signed agreement into a functioning transaction, and it is routinely the last thing resourced and the first thing compressed when timelines slip.
The brands that held a pass requirement rather than an attendance record have since collected something they weren't aiming at. Product knowledge on electrification is now the live capability gap in retail, and it is a gap with a target attached: mandate pressure is measured in registrations, and registrations are made by people who can hold a credible conversation about range, charging, running costs and residuals. A network certified to a standard has that. A network that sat through the deck has a certificate. That difference was invisible in year one and is expensive now.
None of the above is secret. Most implementation managers could list it.
What goes unsaid is the next thing: when a programme starts producing evidence that some element of it isn't working, the incentive to report that honestly is close to zero.
The retailer who raises it looks unready. The implementation manager who raises it looks behind. The steering committee that raises it is questioning a decision the same committee signed off. So the signal gets softened at every level it passes through, and by the time it reaches the people who could change the design, it has been rendered as a local issue at a handful of sites rather than a structural fault.
The programme then continues to be measured against the plan, because the plan is the only agreed measure. And a plan is very good at telling you whether you are on schedule and completely silent on whether the thing you are building works.
This is how a good model acquires a bad reputation. Not through one visible failure, but through eighteen months of small unreported ones — at the end of which the only conclusion available to everybody is that agency doesn't work.
A plan tells you whether you are on schedule. It is completely silent on whether the thing you are building works.Why the reporting stayed green
The two things that actually separate the implementations that landed from the ones that didn't are unglamorous.
Adaptable means the design is treated as a hypothesis until live volume has tested it. It means having a route by which “this isn't working” travels upward fast and intact, and someone with authority at the end of it who can act without a change-control cycle that takes a quarter.
Realistic means being willing to say, out loud and in front of the brand, that a date is wrong or a process won't hold — before the network finds out the expensive way. That is an uncomfortable conversation and it is almost always cheaper than the alternative. Every implementation I have worked on had a moment where somebody had to say a thing nobody wanted to hear. The ones that went well are the ones where it was said early and received without defensiveness.
Neither is a capability you can buy in at go-live. Both are set by how the programme is governed from the beginning, and specifically by whether the people running it want to be told the truth about it.
There is a real strategic question about whether agency suits a given brand, network and market. The influx of new entrants choosing franchise precisely because dealers hold local market expertise the OEM does not is a genuine argument, and worth taking seriously on its merits.
But it should be argued on its merits — not on the back of a verdict reached by looking at execution and blaming design.
If you are pausing agency because the model doesn't work, check first whether what you actually have is a programme that couldn't change course. Those two things look identical from the top of the organisation. They require completely different responses, and only one of them is a reason to stop.
Written from the retail side of five agency models across five marques and more than 40 sites — onboarding, delivery, training, process mapping and post-live regression, delivered solo. Present at network briefings, on the steering groups, and on the phone at go-live when the integration didn't hold. Ground Truth Advisory works with OEMs, suppliers and investors on structural change in automotive retail. The work is confidential and unattributable.