Clarista White Paper
AI cannot think the way your firm thinks, because your firm has never written down how it thinks.
A client of fourteen years asks a question in a review meeting. Why did we come out of that position in 2019?
Finding the trade takes about four minutes. There it is, with the date, the quantity and the price. Then somebody has to answer the actual question, and the room goes quiet in a particular way. One person thinks it was the concentration. Another is fairly sure it was tax. A third remembers a conversation but cannot recall what was said in it.
The client gets an answer. It is a reasonable answer, delivered with confidence, and everybody in the room knows it was assembled on the spot.
That meeting happens at every firm in this industry. The uncomfortable part is not that people forgot. Seven years is a long time. The uncomfortable part is that there was never anywhere to put it. The systems your firm owns have no field for why.
Ray Dalio started Bridgewater in 1975 out of a two-bedroom apartment. It became the largest hedge fund in the world. Ask him what the foundation was and he does not begin with markets.
From very early on, whenever I took a position in the markets, I wrote down the criteria I used to make my decision. Then, when I closed out a trade, I could reflect on how well these criteria had worked. It occurred to me that if I wrote those criteria into formulas, and then ran historical data through them, I could test how well my rules would have worked in the past.
Ray Dalio, Principles: Life and Work, 2017
Read that again with your own firm in mind, because the sequence is the whole argument.
He wrote down the reasoning. Writing it down let him check it against what actually happened. Checking it turned loose judgement into explicit criteria. Explicit criteria became formulas. Formulas became a system that could sit alongside him and decide in his style, at a scale one person cannot reach.
He was building a thinking system decades before the phrase existed. The raw material was not market data, which everyone had. It was his own written reasoning, which only he had.
The habit is older than the stories about him suggest. Bridgewater's Daily Observations, the note in which the firm works through its thinking in public, runs back to a piece Dalio wrote on 15 January 1978. Forty eight years of a firm writing down how it reads the world, dated and on the record.
Howard Marks began sending memos to Oaktree clients in 1990. In January 2025 he described how that went.
I never had a response … But yet I kept doing it.
Howard Marks, on the first decade of the Oaktree memos
Almost ten years of writing into silence. Readers neither praised the memos nor acknowledged receiving them. Then, on 2 January 2000, he published a memo on internet stocks. In his words, it was right, and it was right fast.
Warren Buffett later put it on the cover of Marks's book: when he sees a memo from Howard Marks in his mail, it is the first thing he opens and reads.
The lesson for a wealth manager has little to do with publishing. It is that writing down your thinking looks like pure overhead for years, and then turns into the most valuable thing the firm owns. The firms that start in a year when it feels unnecessary are the ones holding something in year five.
Draw your firm's technology on a whiteboard and it looks like a chain. Onboarding feeds planning. Planning feeds model assignment. Model assignment feeds selection, selection feeds trading, trading feeds accounting, accounting feeds reporting, and reporting feeds the next review.
Look closely and the chain has breaks in it. Between every pair of systems sits a person deciding something, and what the next system records is the result of that decision.
Name the decisions your firm actually makes, then look for where any of them are written down.
Each of those is judgement, and each is a point where your firm differs from the firm down the road. In every case the system records the outcome, dated and auditable, with the judgement left out.
Notice their shape, because it explains why this has stayed unsolved. A few of them happen at a handoff, where one system passes to the next. Most do not. They recur on a cadence, or something in the world triggers them, and they draw on four or five systems at once. A system built to own one part of the process was never going to own a decision that reaches across all of it.
Then look at the bottom of the ring. Aggregation, performance reporting, billing and the client portal sit there as one continuous block, because no judgement happens inside it. It is necessary and it is mechanical.
It is also what almost everybody already owns. Portfolio management software sits at ninety nine percent adoption among RIAs and client portals at ninety seven percent. Compare that with the tools closer to judgement: estate planning at fifty four percent, tax planning at fifty three percent, retirement distribution planning at eighteen percent. The infrastructure is universal, which is precisely why it cannot make you different. Thin adoption sits exactly where the thinking is.
There is a useful precedent outside this industry, and it is worth knowing because the mistake is so ordinary.
In 1975 a member of Congress asked the Federal Reserve for transcripts of its policy meetings. The Chairman at the time, Arthur Burns, replied that they could not be supplied because they were routinely disposed of.
In 1993 it emerged that verbatim transcripts had in fact been kept all along. The revelation surprised even members of the committee, and the Fed was accused of having misled Congress for years. Since then the FOMC has released full transcripts on a five year lag, and they have become the primary material scholars use to understand how monetary policy is actually reasoned rather than merely announced.
Two things stand out. The reasoning was destroyed for years because it was treated as a by-product of the decision rather than as the valuable part. And the moment it was preserved and published, it became the most studied record the institution produces.
Here is where this stops being a philosophical point about record keeping.
Firms are connecting their data and putting an AI assistant on top of it, and the results are genuinely useful. It can tell you the account is on the balanced model, the position is fourteen percent of the household, the last review was in March. It summarises documents well. It drafts a decent follow-up email.
Then somebody asks it the question they actually care about. Given everything we know about this family, what should we be doing that we are not?
The answer comes back competent and generic, and the reason is structural. The system was grounded in records of what happened. Your records of what happened look very much like every other firm's records of what happened. The same custodians, the same instruments, the same account types, the same report formats. What makes your firm different is the reasoning that produced those outcomes, and that is the one thing the system was never given.
This is why a general assistant delivers productivity and very little differentiation. Summarising a document is useful, and it is equally available to every competitor at the same price.
The first gap is connection, and most firms have started on it.
A capital call falls due in three weeks. The quarterly tax payment lands in the same window. A tuition draw sits between them. Household cash covers about a third of the total. The position that would normally fund the gap is pledged as collateral on a line of credit, and it has fallen far enough that selling into it creates a second problem.
Every part of that is simple. Every fact is known to somebody. The call is in an inbox, the tax estimate is with the CPA, the tuition draw is in the plan, the cash is at the custodian, the pledge is at the bank. Joining five facts across five places is what no single system does, which is why the firm finds out in October rather than September.
The second gap is memory, and it is the one almost no firm is working on.
Suppose the system does surface that collision, and the advisor looks at it and decides to leave the concentrated position alone, because it is already earmarked for a charitable gift in the fourth quarter. That reason is correct, it is important, and it exists in no system anywhere in the firm.
Six months later the gift has not happened, and the reason it was left alone has gone with the conversation.
Closing the first gap is a data problem, and the industry understands data problems. Closing the second is different work, and it is where the durable advantage sits, because your competitors are leaving it alone too.
This would be an interesting argument in any decade. It is an urgent one now, for a reason that has nothing to do with technology.
Cerulli reported in August 2026 that thirty five percent of US financial advisors plan to retire within ten years, and that they currently manage forty percent of industry assets. More than one in four of the advisors expected to transition inside that decade are unsure of their succession plan.
Set that against what firms have written down. DeVoe found that forty two percent of RIAs have a written succession plan, the lowest level since they began tracking in 2019, down from fifty percent the year before. A separate survey of firm owners nearing retirement put the share with a comprehensive, funded plan at six percent.
Read those numbers next to the argument in this paper and the shape of the problem changes. Forty percent of the industry's assets are managed by people whose method exists mainly in their own heads, and most of those firms have written down neither the succession nor the method.
There is a quieter version of the same problem that shows up long before anyone retires. Work by Linnainmaa, Melzer and Previtero, published in the Journal of Finance in 2021, found that advisors' own personal portfolios closely mirror the advice they give their clients. Advice tracks what the individual advisor believes. At most firms, the unit of judgement is the advisor, and the firm is the container.
That is fine while the advisor is there and while the advisor is right. It is a problem when a client moves to a colleague, when a new advisor joins and has to learn the method by sitting next to somebody, and when an acquirer asks what exactly they are buying.
Schwab's benchmarking work keeps finding the same separation. Among the firms it classifies as top performing, sixty three percent have documented training and development processes, against forty eight percent of everyone else. The gap is in whether the method is written down.
The word institutionalise gets used loosely, so it is worth being concrete. It means taking something that currently lives in a person and putting it somewhere the firm holds: a written method, an application that enforces it, a record of the judgements made inside it, and a reason attached to each one.
The market already prices this, and the numbers are striking.
Advisor Growth Strategies reported a median RIA valuation of 11.6 times EBITDA in 2025, the highest on record, against a range for a five hundred million dollar firm running from roughly nine to fifteen times depending on how the firm is built. Six turns of EBITDA separate two firms of identical size.
In the same research, buyers were shown two hypothetical firms of the same size. One had twelve employees, two owners and ninety five percent recurring revenue, and attracted interest from eighty percent of the buyers surveyed. The other had ten employees, three owners and seventy five percent recurring revenue, and attracted interest from none of them. Same assets under management, opposite outcome.
When the same firm set out what earns a premium, the list read: leadership depth beyond the founder, revenue diversification across advisors and referral sources, operational systems that deliver consistency rather than reliance on exceptional individuals, documented succession, and strategic fit. Their summary line is worth keeping.
Growth attracts attention. Durability, institutional quality, and strategic alignment determine what that attention is worth.
Advisor Growth Strategies, August 2026
Deal volume rose twenty seven percent in 2025, the fastest acceleration since 2021. Whether or not you intend to sell, you are operating in a market where the buyers have already decided that a documented method is worth more than a talented individual.
One honest caveat. Valuation practitioners agree that concentration of knowledge in one person depresses value, and they describe the mechanism clearly, but none of them publishes a number for it. Treat the direction as well established and treat any specific discount figure with suspicion.
Every platform you buy encodes a method. It has a view about how suitability should be assessed, how a model should be assigned, what a review looks like and what a report should say. Most of the time that view is sensible, and adopting it saves you the trouble of having one.
The difficulty appears where your method is the reason clients hire you. There the platform pulls quietly in the other direction. The workflow is the vendor's, so the process becomes the vendor's, and over a few years the firm drifts toward the average of everyone else using the same software. The work that made you distinctive moves into a spreadsheet beside the system, which is where a great deal of this industry's real thinking currently sits.
The industry data suggests firms feel this without naming it. Advisor satisfaction with technology fell in twenty of twenty three categories in the most recent Kitces research. Integration scores about six out of ten, and fewer than one in three advisors reports genuinely integrated systems. The T3 survey observes that advisors rarely migrate between portfolio management platforms, and suggests the reason is that moving client portfolio data is such a burden. Low satisfaction combined with high switching cost is the signature of a market people are stuck in rather than happy with.
One number in that survey matters more than the rest. Only 3.27 percent of firms run anything resembling a data warehouse. Almost every firm in this industry holds its own history inside somebody else's data model, on somebody else's terms.
There are two obvious conclusions here and both of them are expensive.
The first is to set out and replace the stack. Estimates for a firm building genuinely proprietary technology run from one to five million dollars, and the practical threshold is usually placed somewhere around two billion in assets. A firm that starts there spends two years rebuilding things that already worked, and reaches the end with nothing recorded.
The second is to treat the systems you own as settled. They are not. Every one of them leaves you fitting some part of your firm to the way it works: a field used for something it was never designed to hold, a workaround everyone knows about, a report the team has learned to read around. Much of the time that is a fair trade, and buying the software was the right call. It stops being a fair trade at the point where the thing you are bending is the thing that makes you good.
So start with the reasoning systems, where the fit is worst and the difference is yours, and let the boundary move afterwards on evidence. It does move. A decision system usually turns out to need a piece of record keeping to work properly. A trading application needs positions at tax lot level through the day rather than at the close, so the bookkeeping comes with it, and once that exists the case for keeping the old module weakens on its own. Extend when that happens, because by then you know exactly what the new piece has to do and why. Replacing a system of record can be a good decision. It is a poor place to begin.
Build in the gaps instead. The applications worth owning are small, specific and unglamorous: the suitability review your firm actually runs, the model assignment logic your investment committee actually applies, the manager selection method your team argues about, the liquidity view that pulls the household together. Each one takes a process out of a spreadsheet, applies it consistently, and records the judgement and the reason inside it.
That is a different proposition from building a platform. You are not replacing the systems of record. You are occupying the space between them, which is the space no vendor can sell you anything for, because what belongs there is specific to you.
Three questions sort the candidates quickly.
Does our approach differ from what the standard tool assumes?
Every wealth platform encodes a method. Where yours matches, use the platform. Where it differs, the platform quietly pushes you back toward the average.
Is the real work happening in a spreadsheet, or in somebody's head?
A spreadsheet is a process that outgrew its system. Somebody's head is a process that never had one.
Would we struggle to explain this decision two years from now?
If that is true for a client, a regulator or a successor advisor, the reasoning is not being kept.
Two yeses make a process a candidate.
The same areas come up across firms: client suitability, model and strategy assignment at the account level, private markets manager selection, trading, research, and household liquidity. What they share is that the judgement is the work.
The useful instruction is to be unambitious. Look for where your process fails to fit neatly into a standard application, or where people are still working by hand. Start there, and do it properly, which means capturing why you acted and why you did not.
Asking people to keep a journal will fail at firm scale. Dalio's method worked because he was one person with one process. Inside a firm, the capture has to sit inside the work, and it has to cost seconds.
Four things make that possible.
When the system raises something, it shows the facts that produced it, each with the system it came from and the date it was last true. An advisor should be able to confirm or reject it in fifteen seconds without leaving the screen. Anything slower gets ignored, and an ignored channel is worse than no channel at all.
Acted, considered and declined, and this is not a real signal. The middle one is the valuable one. A simple accept or dismiss throws away the case where the system was right and the answer was still no, which in practice is the most common case of all.
A short list of specific reasons, free text alongside, and a proposal rather than a blank box. The moment an advisor writes that the position is pledged and earmarked for a gift in the fourth quarter, the firm knows something it did not know before, and it knows it somewhere that outlasts the advisor.
Every reason for not acting is true until something changes. Record what that something is. When the gift has not settled by the end of December, the item comes back, quoting the reason and its date. That is the difference between a system that stores a dismissal and one that holds the firm to its own thinking.
Bridgewater's record has been challenged, most thoroughly in Rob Copeland's 2023 book, and one of the criticisms bears directly on this argument. A firm that writes everything down can still forecast a dozen downturns and take credit only for the one that arrived.
That is a fair point, and it strengthens the case rather than weakening it. Go back to what Dalio actually said he wrote for: so that when he closed out a trade, he could reflect on how well the criteria had worked. The record earns its keep at the moment you read the ones that went wrong.
A record you never audit is storage. A record you review is a method. The difference is a standing meeting and the willingness to look.
Precision matters here, because this area attracts overstatement.
There is no SEC rule requiring an adviser to document the rationale for a recommendation, or for declining to make one. Rule 204-2 requires preservation of written communications relating to advice that already exist. It does not require a rationale record to be created.
What the SEC staff has said is more useful than a mandate. The 2023 staff bulletin on care obligations states plainly that there is no documentation requirement, and then says it may be difficult for a firm to demonstrate compliance without documenting the basis for certain recommendations, and that firms should consider documenting the process and reasoning behind recommendations involving complex, risky or expensive products.
Treat that as a benefit rather than the reason to act. The reason to act is that the reasoning is worth keeping.
In the first ninety days one process comes out of spreadsheets, and the first reasons start landing.
By the end of year one the firm has a defensible file for the decisions that matter most, and it can answer questions about its own practice that it used to answer by asking around.
By year three the record is large enough to be taught. Patterns in the reasoning become firm policy. The system raises things the way your senior people raise them, because it has watched your senior people do it several thousand times. That is Dalio's sequence, run at the scale of a firm rather than one desk.
Read the next few paragraphs as a thought experiment rather than a forecast. It is written as though from five years out, because the mechanism is easier to see from the far side of it.
A thought experiment, dated 2031
The surprise was how little the models had to do with it.
By 2029 every firm had good AI. The assistants were fluent, they read documents accurately, they drafted well, and they answered planning questions to a standard that would have impressed anybody in 2026. A client could get a competent second opinion from a consumer application in about ninety seconds, free.
That was the problem. Competent stopped being worth paying for. The floor came up for everyone at once, and the floor is not a business.
What separated firms was something duller. A minority had spent the late twenties recording why they did things. Their systems could answer a client who asked why, could explain a position to a regulator with the evidence attached, and could hand a departing advisor's book to a colleague without the method leaving too. Their AI gave answers that sounded like the firm, because it had been trained on the firm.
The rest had connected their data beautifully and had nothing to say with it. Their assistants gave the same answers as everybody else's, which clients noticed, slowly, and then all at once.
A good number of clients had stopped logging in at all. Their own assistants asked the questions, nightly, and formed opinions. Firms with something worth saying to those assistants kept the relationships. Firms with a portal and a quarterly PDF discovered how little either had been worth.
The consolidation that followed was not really about scale. Buyers paid for firms whose method survived the founder, and passed on firms where the method was the founder. That had always been true. AI simply made the difference legible.
Step back out and the mechanism is straightforward enough to state in one line. Improvements in general AI raise the baseline for every firm at the same time, which is exactly why they cannot differentiate any firm. Whatever any competitor can buy this year, they can buy next year too, and slightly cheaper.
So the interesting question is what a competitor cannot buy. The models are available to everyone. The data feeds come from the same custodians. The reporting is something every firm already has. What remains is the record of how your firm decides, and the reason it is not for sale is that it has to be accumulated rather than purchased.
That leads to an uncomfortable implication worth stating directly. A firm that starts recording its reasoning this year cannot be caught next year by a firm with more money, because the thing that matters takes time to make rather than capital to buy. Advantages with that shape are rare in this industry, and they do not stay available indefinitely.
Generation Alpha and Beta are turning into a prompt generation, automating their tasks with agents. They are growing up instructing software rather than signing into it, and they will inherit the wealth your firm manages.
That habit is not confined to children. Younger principals, founders and next generation family members are already working this way, and in plenty of families they are the ones deciding now.
Think about what they will want from an advisor. Signing into a portal to read a PDF will strike them as an odd way to learn about their own money. What they will want is to point their own agent at their wealth and ask it questions, at whatever hour suits them, in whatever form they like.
That has two consequences, and the second is the larger one.
The first is access. A firm will need to hand a client's agent a governed, permissioned and audited view of the household: holdings, cash, obligations, commitments, documents and the current picture across every entity. Few firms can do that today, because the picture exists in pieces and there is no layer to serve it from. Doing it safely is a governance problem as much as a technical one, since the firm is deciding what an outside agent may see, on whose authority, and with what record of having seen it.
The second consequence is the one worth sitting with. Once every client's agent can read the data, the data stops being the service. Any competent agent can compute an allocation, spot a drift and summarise a statement. What a client's agent cannot produce on its own is judgement: which of these observations actually matters for this family, what the firm thinks about it, and why.
So the differentiation moves again. Access to information stopped being scarce years ago. The ability to assemble the picture is going the same way. What stays scarce is the quality of the thinking, and the only version of that which can reach a client's agent is the version somebody wrote down.
Advisors can start differentiating on signal quality immediately, for their own use, long before any client agent arrives. That is the useful part. The work that makes you the firm worth talking to in 2031 is the same work that makes your own Monday morning better in 2026, and the record it depends on takes years to accumulate.
A firm that recorded why it acted can teach a system to think the way it does. A firm holding only the outcomes starts from zero, however good the models get. That is the argument for starting now rather than waiting. The technology will keep improving on its own. The record will not build itself.
The sequence runs backwards from the question, which is the opposite of how data projects usually start.
Start with the questions that matter. Only the five or six a principal actually asks and currently cannot answer well.
Work back to the decisions behind them. A good question is usually a stack of decisions. Answering it well means getting those decisions right and consistent.
Build applications around those decisions. Each one takes knowledge out of a spreadsheet, a mental model or a manual process, with room built in to record why. These do not have to be ambitious. The value is consistency, saved time, fewer errors, and a reason attached to every decision.
Land the data those applications and answers need. Data last, because now it has a job before it arrives. In practice this runs in parallel, with sources aggregating into a data lake the client owns while the first application is being built.
Start agents on the simple work. Meeting preparation and real-time questions about a household are useful immediately. Agents that reason the way your firm reasons come later, as the record accumulates. Promising those on day one would be dishonest, because the material they need does not exist yet.
This is the approach we take at Clarista, and the part worth saying out loud is the sequencing. Firms have been sold the warehouse first for twenty years, which is why so many of them own one that sits unused. Start with the question, and the data arrives with a purpose.
Two tracks run in parallel from the first week, because each one is slow on its own and each needs the other.
Aggregating sources is table stakes, and a long list of vendors do it. The work that decides whether any of it is usable comes after: linking entities and accounts so a household is one thing rather than nine, defining what your firm means by each term so two reports stop disagreeing, verifying values against their sources, and monitoring for the moment something breaks. It lands in a data lake the client owns, on the client's own infrastructure.
We take one process where your method differs from what the standard tool assumes, and build the application that runs it, with a place to record why built into the work rather than bolted on. Suitability. Model and strategy assignment at the account level. Manager selection. Household liquidity. Each one takes the process out of a spreadsheet, makes it run the same way every time, and starts the record.
The distinction here matters more than anything else in this section.
Productivity agents work from the data. Meeting preparation, portfolio analysis, tax analysis, pulling terms out of trust deeds and fund documents. They save real time, they are worth having immediately, and within a couple of years every firm in this industry will have them. That is exactly why they cannot be the strategy.
Reasoning agents work from the data and the recorded reasoning together, and they sound different. Why is this household on this allocation, and what would we change it for. The reason you gave in March for leaving that position alone has expired, because the gift has not settled and the pledge is still in place. Three demands land on this family's cash inside nineteen days and the position that usually funds it is collateral on a line. The firm has faced this eleven times and decided the same way nine of them, and this case differs in one respect.
Those only exist at a firm that has been recording. They cannot be bought, they improve every month the record grows, and they are the reason the first two tracks are worth running.
Apps and agents run under your controls, with access management, observability, and a record of what each agent did and on whose authority. This is the unglamorous part that decides whether any of it survives contact with a compliance review.
One honest note on sequencing. Productivity agents arrive early, because the data supports them. Reasoning agents arrive later, in proportion to how much reasoning has been recorded. Any vendor promising the second on day one is describing something that cannot exist yet.
Pick one process where your method differs from what your platform assumes. Write down the decisions inside it and who makes them today. Then ask what a reviewer would need in two years to understand one of those decisions, and check honestly whether anything your firm owns holds it.
That exercise takes an afternoon. It usually settles the question of where to begin, and it is the same afternoon Ray Dalio spent, one trade at a time, for fifty years.
Institutional knowledge is everything an organization knows that lives in its people rather than its systems: why decisions were made, which approaches failed and were abandoned, what a specific client needs before they ask. Systems record outcomes; institutional knowledge is the reasoning between them, and it walks out the door when people leave.
Tacit knowledge is expertise a person applies without being able to fully articulate it: pattern recognition, judgement, feel. It matters for AI because a model can only learn what has been recorded. A firm that writes down its reasoning converts tacit knowledge into training material; a firm that records only outcomes has nothing for the model to learn its judgement from.
Dalio wrote down his decision criteria before acting, each time, and then converted those written criteria into rules a machine could execute and test. The input was written reasoning, not market data. That order of operations, record the thinking first, automate it second, is the part most firms skip.
Make recording the why part of the workflow, not an archaeology project later: capture a short written rationale at the moment of each material decision, including decisions not to act, link it to the client and position it concerns, and store it where a governed system can retrieve it. With 35% of US advisors planning to retire within ten years, the window to record their judgement is while they are still making decisions.
Choose which categories of cookies and tracking technologies you allow. Strictly necessary cookies cannot be disabled. Read our full Cookie Policy.
Required for security, authentication, fraud prevention, load balancing, and storing your consent preferences. Always on.
Remember choices you make to provide a personalized experience (language, region, theme, saved filters).
Help us understand how the site is used (Google Analytics, Vercel Analytics, Microsoft Clarity, etc.).
Used for conversion tracking, retargeting, and personalized ads (LinkedIn, Meta, Google Ads, Bing, etc.).