How much customization should a financial platform offer?

Apr 10, 2026

The surface of this question is about customization — how much to offer, where to constrain, whether to run an open platform or a curated one. I gave a talk at Future Alpha recently on customization at scale, and the conversations afterward sharpened what I’d been thinking in ways I didn’t expect. I’ve come to believe the surface question is a distraction from the real one, which is structural: in financial advisory, the platform’s direct customer is not the ultimate beneficiary of what the platform does, and most of the hard decisions flow from that fact.

In most software businesses, buyer and beneficiary are the same. Stripe sells to the merchant; the merchant’s outcome is what Stripe optimizes. Figma sells to the designer; the designer’s work is what Figma makes better. Even where there’s a chain, it’s usually aligned — Shopify serves merchants who serve shoppers, and merchants want shoppers happy. Customization in those businesses is a straightforward product decision: let the customer do what they want, within sensible bounds, because what the customer wants is broadly what their customer wants too.

Financial advisory doesn’t work like that. The FA is the platform’s customer. The investor is the FA’s customer. The FA’s interests and the investor’s interests diverge in a particular way — not through malice in most cases, but through ordinary human drift. FAs develop preferences, habits, stylistic commitments, and a sense of craft. They come to believe their way of doing things is the right way, and they often believe this with more confidence than the evidence supports. This isn’t a character flaw; it’s how expertise feels from the inside, in any profession. But it means that when the platform asks “what does the FA want?”, the answer isn’t reliably “what’s good for the investor.” Those answers are often the same. They are also often, quietly, not the same.

One founder friend framed the three-party structure from a different angle. His wealth management startup, he told me, is a service company, not a tech company — they don’t even maintain a client-facing website. The tech is the factory that lets reps do the work; the reps are the interface to clients. Clients never touch the platform directly. They touch the rep, and the rep touches the tech. That framing isn’t quite the open-versus-opinionated question, but it sharpens it. Once you accept that the rep is the interface and not the tech, the platform’s influence on investor outcomes runs entirely through the shape of the tools it hands the rep. “Neutral tools” in that context is a fiction. Every tool shapes what its user can easily do, and a platform that pretends otherwise is just declining to take responsibility for the shape it’s already imposing.

This is the part of the problem that customization-as-usual doesn’t handle. An open platform treats the FA as sovereign: tell us how you want to run your book, and we’ll build it. That approach onboards firms quickly, it flatters the FA’s sense of craft, and it relieves the platform of the hardest question in the business, which is what good investing actually looks like. It also, inescapably, makes the platform the FA’s agent rather than the investor’s. The platform is saying: your preferences are our product. Whatever the investor gets, they get through you, and we don’t have a view about what that should be.

I don’t think that’s tenable. Not because it’s illegal — it isn’t, exactly — but because it’s a posture the platform can’t actually hold up under scrutiny. If an FA runs a strategy on the platform that harms their clients, and the platform built the strategy-running tool without a view about what a defensible strategy looks like, the platform has contributed to the harm even if it can’t be held formally responsible. Operating in a fiduciary industry without a fiduciary posture is a choice, and it’s a choice that eventually gets noticed — by regulators, by clients, by the FAs who do take their duty seriously and don’t want to be on a platform that treats them the same as the ones who don’t.

Another industry leader friend takes the harder line directly: he’s against open platforms in this industry as a matter of principle, and thinks the tech should encode best practices and financially sound strategies. I’m closer to his position than to the service-company founder’s, but I’d pull back slightly from “encode.” The strong version of his view is that there’s a set of right answers and the platform enforces them. The defensible version, I think, is softer: most of what FAs want to customize isn’t genuine edge, and the platform should make the defensible center easy while requiring articulation for deviation. Encoding is too strong when you’re not certain, and in investing the honest answer is that the evidence is very strong on some things and weak on others.

Which is where I want to be careful about the opposite extreme. A platform that tells FAs there is exactly one right way to invest, and forces everyone into it, is overclaiming. The evidence is strong that fees matter, that diversification helps, that most active management doesn’t beat its benchmark net of costs, and that behavior is more dangerous to returns than market volatility. It is weaker on many other things. Legitimate specialization exists. An FA who has spent twenty years on municipal bonds knows things the platform doesn’t. A firm that focuses on a specific client profile has real insight into that profile. Treating all FA variation as preference to be overridden would throw away genuine edge along with noise.

What I’ve landed on — tentatively — is something between those poles. Most of what FAs want to customize is preference, not edge. The platform should build defaults around the defensible center:That is, via regularization. the strategies it makes easy, the analytics it shows first, the workflows it optimizes for, the reports it generates by default. Deviation from the defaults shouldn’t be impossible, but it should require articulation. An FA who wants to run something the platform’s defaults don’t support should be able to, but should have to say why — and the platform should be willing to decide whether that “why” constitutes legitimate specialization or idiosyncratic preference. Some justifications the platform should accept. Others it shouldn’t.

There’s a way to sharpen this that I keep coming back to. Stated plainly: the platform exists to help FAs scale and to let investors transact with minimal friction and surprise. Everything downstream — the opinionation, the defaults, the articulation requirement — is in service of those two. The mechanism for both is commoditizing investor problems at the substrate layer. Every recurring problem the platform solves once — benchmark rebalancing, tax-loss harvesting, transition management, basic portfolio construction — is a problem the FA doesn’t solve separately for each client, and a source of potential surprise the investor doesn’t have to absorb. Opinionated defaults aren’t a constraint on the FA’s practice; they’re leverage for it and protection for the investor at once. The platform isn’t choosing between the FA’s interest and the investor’s. It’s choosing between the FA’s stated preference — do it my way, for every client — and the FA’s actual interest, which is serving more clients more consistently with less error. Those are genuinely different things. They don’t always feel different from the inside.

This is harder than open customization in every dimension that matters for growth. It slows onboarding. It creates conversations with prospective customers that end in them choosing someone else, because the relationship the platform offers isn’t vendor-to-customer in the usual sense — if you’re asking an FA to accept defaults they didn’t choose, you have to show up as a credible thinking partner rather than a service provider, and the conversation has to be about investment reasoning, not feature parity. That posture is openly at odds with the “customer is always right” mantra, and the tension is worth owning rather than dodging.The opposite of “the customer is always right” isn’t “the customer is wrong.” It’s “the customer deserves a vendor who actually thinks.” Some FAs will find it frustrating and leave, and that churn isn’t a failure; it’s a filter — be selective yourself about which FAs to convert, because the one who’s agitated by defaults in sales will stay agitated by them forever after. It requires investment professionals on staff who can defend a philosophy, not outsourced customer support — people carrying a view with real zeal while staying open to where a specific FA has legitimate edge.A large fraction of what arrives as a customization request is a feature-discovery problem in disguise — the FA wants X, the platform already supports X, but the naming is off or the workflow is buried. A professional closes the gap in conversation; a support rep files a ticket. And it asks FAs to change behavior, which is the single thing adults are least willing to do. Some FAs will experience this as paternalism, and they will not be entirely wrong — it is paternalism, in the specific sense that the platform is asserting a view about good practice that the FA didn’t ask for. The question is whether the paternalism is earned. If the platform’s defaults reflect defensible evidence, I think it is. If they reflect the platform team’s own preferences dressed up as evidence, it isn’t.

The operational argument for opinionation is worth restating, because I used to frame it shallowly and the sharper version does more work than I gave it credit for. The shallow version is that open customization is expensive because every edge case has to be supported. True, but not the point. The sharper version is that most customization requests, examined closely, are toil disguised as flexibility. Every bespoke workflow at scale is a compliance surface that isn’t instrumented the way the defaults are, a knowledge dependency on whoever built it, an ops risk that accumulates invisibly until an incident surfaces it. The cost isn’t in the support queue. It’s in the failure modes you can’t see until they’ve already happened, which in a fiduciary industry is the most expensive kind of cost there is. But the fiduciary argument is still the deeper one. Open customization, in this industry, forces the platform to abdicate the fiduciary question, and the platform can’t abdicate it, not really. The defaults the platform builds, the workflows it makes easy, the reports it generates out of the box — all of these are already answers to “what does good look like.” Pretending they aren’t is leaving those answers unexamined. The service-company founder is right that tech isn’t the interface. But that’s an argument for more care about the shape of the tech, not less — because the shape of the tech is what reps end up doing. The choice isn’t whether to have a view. It’s whether to have a considered one.

The intellectual frame that’s helped me most in this work isn’t from engineering literature. It’s from Deleuze and Guattari’s A Thousand Plateaus. Their central contrast is between the arborescent, which is tree-like, hierarchical, rooted in a central trunk from which everything descends, and the rhizomatic, which is horizontal, multi-entry, connection-forming, able to regenerate from any point. Their argument is that Western thought has overvalued the tree and that we should think rhizomatically instead. Taken at face value, this reads as a license for pure openness — no hierarchy, no center, no defaults. Everything connected to everything else.

I think this is exactly the wrong lesson to take from it, and the inversion is where the frame starts earning its keep in engineering. (D&G themselves might see what I’m about to do as smuggling the tree back in through the side door. They’d be partly right. But the inversion is useful anyway.) The most productive rhizomatic systems I know of are built on profoundly arborescent substrates. The internet is the obvious example. Its topology is rhizomatic, but its protocols are opinionated and non-negotiable. TCP/IP is a tree. DNS is a tree. You can only get to the rhizome because everyone agreed to stand on the same trunk.

The pattern shows up everywhere once you see it. Apple looks arborescent at the surface — curated store, design language descending from Cupertino — but the App Store ecosystem itself is rhizomatic, with apps connecting through URL schemes, share sheets, universal links, continuity. The platform imposes rigid rules at one layer so fluid connection can happen at another. Android went the other direction, open at the platform layer, and the result isn’t a single rhizome but a proliferation of fragmented arborescences — each OEM and carrier growing its own little tree. Nominal openness produced more hierarchy, not less, because without a shared substrate there’s nothing for rhizomatic behavior to grow on.

Palantir’s Foundry works the same way. The ontology — the semantic layer that models a customer’s business as objects and relationships — is deeply opinionated. Every entity has a canonical definition; every relationship is explicitly modeled. That rigidity is the point. Once the ontology is in place, analysts, engineers, and executives can compose workflows rhizomatically above it, because they share a substrate. The arborescence of the ontology is what makes the rhizome above it possible. Plaid does this at the data-normalization layer — one opinionated schema across messy bank APIs, which is precisely what enables any fintech app to connect to any account without rebuilding twenty integrations.

The engineering principle I’ve taken from this is: decide consciously where the tree goes and where the rhizome lives. Choosing both is the job. Engineers who say they want “flexibility everywhere” usually produce fragmented trees rather than rhizomes, because flexibility without a substrate is just isolation. Engineers who say they want “opinionation everywhere” produce cages, because rigidity without fluidity above it has nowhere for real use to live. The art is in the layering — build a trunk rigid enough that rhizomatic behavior can safely emerge on top of it, and don’t confuse the trunk for the whole thing.

This reframes the platform question in a way I find clarifying. The open-platform position in financial advisory isn’t really rhizomatic; it’s Android-style — each FA grows their own little tree on the platform, and the platform itself has no substrate connecting them. The opinionated position isn’t a denial of FA fluidity; it’s the Apple move — opinionation at the substrate layer so that fluid, differentiated practice can happen above it. The defaults are the trunk. What FAs do with them is the rhizome. You need both. The question isn’t whether to have a tree, but whether to build one consciously at the layer where it does the most good.

I don’t know that I’d press this all the way to “there are only a few smart ways to do things.” That’s overclaim. But “there are not infinitely many smart ways, and most of the variation FAs want to express isn’t among them” is a claim I can defend, and it’s enough to justify the moderate position: build defaults around the defensible center, let FAs deviate with justification, and be willing to lose the FAs who can’t or won’t offer one. The FAs that costs you are mostly the ones who shouldn’t be on the platform in the first place.

Following the logic through to its end changes what the FA actually is. If the platform holds the investment philosophy, owns the defaults, and enforces the defensible center, then the FA is no longer delivering a fundamentally differentiated product. They are selling different flavors of the same thing — the same risk frameworks, the same core strategies, the same evidence-based defaults, applied with their own relational style to their own book of clients. The IP moves into the platform. What the FA brings is distribution, relationship, trust, judgment in context, the ability to sit with a nervous client in a bad quarter and say the right thing. That’s real work and worth paying for, but it isn’t the work most FAs believe they do. Most of them believe they’re delivering edge. In this world they’re delivering service on top of edge the platform produced. That’s a smaller self-image, and it’s the self-image the industry has been quietly resisting for a decade, which is why robo-advisors have been only partially disruptive — not because they were wrong about the economics, but because they forced FAs to confront what they were actually being paid for, and most FAs preferred not to.

The part I’m still uncertain about is the edge — what’s left after the platform internalizes the IP. My instinct is that real differentiation survives at the relational layer and in a narrow band of genuine specialization, and collapses almost everywhere else. More than 20% of current FA variation is legitimate, less than half. I’m guessing at the number, but I’m not guessing at the direction. The platform’s job is to discover the number by trying, getting it wrong in specific cases, and correcting — and to be honest with FAs along the way about what role they are actually being invited to play. Opinionation isn’t the same as certainty. It’s the willingness to say out loud what a platform that internalizes the IP is actually asking of its FAs: not to invent the investment product, but to deliver it well to clients who would otherwise have no one to sit across the table from. That’s a worthy job. It’s just not the one most of the industry thought it signed up for.