The Integration Tax

I was in Bucharest last weekend, talking with Paul, our first employee at CryptoCompare, about the portfolio tool. Neither of us had touched it in any real way since 2017, when it had over a million monthly active users [2]. Real people tracking holdings with categorisation and rules they'd built themselves. The data side of CryptoCompare kept growing through that period at exponential rates. We added exchanges, assets, chains, every kind of feed the institutional business needed. The portfolio got the same coverage as a side effect. My reasoning when I agreed to deprioritise the product work was that we'd focus on getting the data right and come back to the portfolio later, once the data was comprehensive enough. Later didn't come, the users didn't get a shutdown email, they noticed over years that the UI hadn't moved and that features they'd asked for weren't shipping. The portfolio still works and still has users, and it hasn't been seriously improved in nine years.

The portfolio is manual-input only. Users type their holdings in, which means the data is a virtual portfolio rather than a verified ledger of what anyone holds. Adding auto-integrations to exchange APIs and wallet addresses would duplicate the engineering work the data-licensing team is already doing, make the user data legally hot the moment it becomes verifiable, and put us on the receiving end of source-of-funds requests and tax-authority subpoenas every quarter. The retail revenue wouldn't fund the work, and even if it did, the moment we held a verified picture of a million users' real holdings we'd be one product manager's bad day away from finding ways to monetise it (this is not a hypothetical, every meeting about new revenue lines eventually surfaces it). Staying manual is good for the business and bad for the users.

The CryptoCompare portfolio is one version of a problem every personal-tracker company in history has faced. The structural reasons that kept us from rebuilding it are not specific to crypto, not specific to portfolios, not specific to one company's revenue mix. They apply to anyone whose business depends on holding a verified picture of users' lives.

Every personal tracker has an integration tax. Banks change their export formats, APIs deprecate, OAuth scopes tighten, fitness platforms get acquired and rename their endpoints, health data vendors change their units, medical providers rotate their login flows. The tax is paid in engineering time, forever, and it grows with every integration the product adds. The first five integrations are a team-month, the next twenty are a team, the hundred after that are a department. Any serious aggregator is running a small engineering function whose only job is keeping the existing integrations working. We had a six-person team at CCData on exchange data integrations alone (and exchanges are the easy ones, because they have engineering teams who care about API stability). Account integrations are harder.

The aggregator infrastructure between consumer trackers and source banks has the same problem. Yodlee, founded in 1999 as the consumer-PFM aggregation backbone that Mint and Money Dashboard ran on, sold to Envestnet in 2015 for $660 million [11] and pivoted to enterprise B2B because the consumer side couldn't sustain itself as a standalone product. Plaid, the modern equivalent, agreed to a $5.3 billion acquisition by Visa in 2020 that the DOJ blocked on antitrust grounds [10], and reached a $13.4 billion private valuation by selling enterprise APIs to banks and fintechs, not consumer experiences. The rails the consumer trackers run on are owned by businesses whose customers are the banks, not the people whose financial data flows through the pipes.

The tax has to be paid out of revenue the product generates. A personal tracker survives only if what its users pay exceeds the rolling cost of keeping the integrations from decaying. Across a decade of products and a dozen comparable categories, users don't pay enough to cover the integration tax plus the team around it plus the hosting plus the compliance.

So the category works in one of three ways. The product charges a subscription that most users cancel inside a year because the value is hard to feel month-to-month, and churn kills the team before the integrations can stabilise. Or the product gives up on direct revenue and monetises the data, selling anonymised transaction feeds to hedge funds, selling health metrics to insurers, selling behavioural signal to ad networks. That is the compromise an entire generation of free trackers ran on, and the reason most of them existed at all. Or the product hides a trading affiliate fee or a lending partnership or a rewards markup inside the experience, and the product's advice quietly bends around what earns the company revenue rather than what serves the user.

A fourth path exists and it doesn't work. The product ships honest, prices correctly for the work, gets undercut by a competitor running one of the three paths above, and either dies or pivots to one of the three paths itself.

This is a structural feature of a product category whose costs are unbounded and whose willingness-to-pay is modest. The CryptoCompare portfolio is the version I have first-hand experience with. The category has buried more products than it has shipped, across forty years and two continents, in patterns that repeat closely enough to be diagnostic.

What dies when a personal tracker dies is the interpretation layer the user built on top. The raw numbers are usually recoverable from the source. The categorisation, the tagging, the goals, the rules, the queries they saved, the habits they wired up around the product's particular shape, all of that is the user's, in the sense that nobody else made it, and it's the company's, in the sense that the company's database is the only place it lives. When the company moves on, the layer goes with it.

The shutdown cases come first. Wesabe launched in November 2006, ten months before Mint, as the first personal finance tracker to attempt automatic transaction aggregation by building its own scraper rather than relying on Yodlee, which Wesabe's founders considered insufficient on privacy grounds. It led the category until Mint launched in September 2007 using Yodlee for its aggregation layer, and shut down in July 2010 after burning through its raised capital and surviving nine more months on revenue alone [3]. Mint reached approximately twenty million active users at its 2016 peak, was acquired by Intuit in 2009 for $170 million, and shut down in March 2024 with users migrated to Credit Karma, which lacked equivalent budgeting, transaction rules, or custom categorisation [4]. The UK's largest personal finance manager, Money Dashboard, was acquired by ClearScore in 2022 and shut down by ClearScore in October 2023 with a closure statement saying the company "could not find a sustainable business model for the apps." ClearScore kept the open-banking infrastructure that powered Money Dashboard and rebuilt it as a B2B product called D•One, sold to lenders [5].

The longest-running personal tracker on record is Quicken, founded in 1983. Intuit sold it to private equity in 2016 after concluding it was "essentially a dead end," contributing $51 million to Intuit's $4.2 billion total revenue. H.I.G. Capital resold the business to Aquiline Capital Partners in 2021. Quicken moved to a mandatory subscription model in 2018, with users trapped in a proprietary file format and integration quality declining year over year. Forty-three years in, it survives by holding its install base captive [6].

Personal Capital reached three million users on a free dashboard funded by AUM fees on the wealth-management business that paid the bills, and got absorbed into a $1T retirement provider in 2020 [7]. CoinTracker bundled the dashboard with crypto-tax filing, which is regulatory-mandatory in most jurisdictions, raised a $100 million Series A in January 2022 at a $1.3 billion valuation, and has not raised a follow-on round in the four years since [8]. The crypto-tax bundle is the closest thing the category has to a structural commercial advantage. The runway clock is the relevant variable, not the product quality. Personal Capital made the lead-magnet model work by being absorbed into a parent business that needed customer acquisition. CoinTracker hasn't found that parent yet.

Six trackers across forty years and two continents. Three shut down outright; one survives in proprietary captivity; one was absorbed into a wealth-management parent; one is burning down its Series A on the way to one of the first three. None survived without compromise.

Lefteris has been building Rotki for years [1]. Local-first portfolio tracker, runs on your hardware, your data stays on your machine, open source, no cloud dependency, the kind of architecture I'd recommend to anyone who reads this site. The categorisation is yours and stays yours. The integrations are open source and the export is whatever format you want. Rotki is the right architecture for the category.

It has also struggled to get traction the entire time Lefteris has been building it. The product is genuinely good, and that hasn't been enough to overcome the category. The market for self-hosted portfolio tools is small, the price ceiling is modest, every integration is forever maintenance work against a revenue base that doesn't fund a team, and the users who would benefit most don't know the product exists or wouldn't run a self-hosted thing if they did. The right architecture exists, has been built by someone with the skill and the values to do it correctly, and is having to fight uphill against the same economics that pushed the CryptoCompare portfolio into nine years of standstill.

The portfolio I built at CryptoCompare had a million monthly active users and the wrong architecture. Rotki has the right architecture and a small fraction of the users. The structural problem is that "user-friendly enough to win mass adoption" and "durable enough to survive its own creator" have never been the same product. I need to build tallyd to somehow be both (which is either the most ambitious thing I've ever signed up for, or the most naive).

This is the part that's uncomfortable to say out loud. The category cannot be saved by another company with better intentions. Every product named above was launched by people who genuinely thought they would treat users better than the previous generation, and every one of them was either failed by the economics or compromised by the temptation to monetise the data they were holding. The architecture has to remove the company, because the company is what fails.

Tallyd is where the numbers about your life live. Transactions, holdings, heart rate, sleep stages, screen time, commits, weight, workouts, medications. It is the budget app, the portfolio dashboard, the fitness tracker, the habit log, all at once and connected. The rest of the fleet reads from it. Fifty to a hundred entries a day across forty years comes out to about a million rows, comfortable in any reasonable database. Postgres handles a life today, and what it looks like in thirty years isn't a problem I need to solve now. The engineering this needs has always existed. The commercial reason to point it at one person never has.

Once a daemon knows your holdings, your cash flow, your goals, and your past decisions, it can push back when you're about to make one. You think you want to buy more of a stock you've watched for months. Shadowd points out that you already have ninety percent of your portfolio in that industry. The fleet tells you what you're missing while you decide. No existing company can build this.

It's me and AI on tallyd right now (I'm a very optimistic person), and I am not pretending that's a stable position. LocalGhost is one person and a lot of tooling. The integration tax doesn't disappear; AI tooling drops the cost of writing a parser from weeks to hours, the plugins that ship will be the ones I care about, and everything else waits for someone who cares enough to write it. If the project stalls, or I get hit by a bus, or I get excited about a new project (statistically the bus is the least likely), the data survives because it's on your hardware, the parsers stay open because they're on GitHub, and the categorisation is yours. There's no company here to fail. LocalGhost is a project I can dedicate the rest of my life to, and even if I don't, the architecture means the data doesn't get lost.

A standard for portfolio data exists. The OFX investment message set, shipped by Intuit, Microsoft, and CheckFree in 1997 and maintained by the Financial Data Exchange consortium [9], covers holdings, positions, and transactions. Almost no consumer tracker implements it as an export option. Not because the standard is hard, not because the implementation is expensive, not because users haven't asked. Because no company in the category has ever had a commercial reason to make leaving them easier than staying.

Tallyd holds the stats across a life, owned by you. The crypto-index work I did at CryptoCompare reduces a basket of correlated financial assets to a single tradable number, which is hard but tractable. Reducing a person to a number, where the inputs are a year of sleep, mood, decisions, holdings, commits, and weight, is a different kind of problem.

The technical part is real. The moral part is whether anyone should be reduced to a number, including by themselves, and whether having the data to do it changes the answer. That's the next post.

[1] Rotki, an open-source local-first portfolio tracker built and maintained by Lefteris Karapetsas, a former Ethereum Foundation engineer. Rotki runs on the user's own hardware, holds the user's data locally, supports a wide range of exchanges and chains through open-source plugins, and offers a free tier alongside a premium tier funded by user subscriptions and donations. Rotki demonstrates that the right architecture for a personal tracker is buildable; it also demonstrates the structural difficulty of getting traction for a self-hosted product against company-backed alternatives that monetise differently. Project site at rotki.com. Source code at github.com/rotki/rotki.

[2] The CryptoCompare portfolio tool reached over a million monthly active users by 2017. Source for the figure is the author's direct experience as CTO and co-founder of CryptoCompare during the period in question. The product remains live but has not been substantively redeveloped since 2017, with engineering effort redirected to the company's institutional data-licensing business (now operating as CCData following the CoinDesk acquisition announced 16 October 2024).

[3] Wesabe launched in November 2006, ten months before Mint, as the first online personal finance tracker to attempt automatic transaction aggregation. Founders Marc Hedlund and Jason Knight chose to build their own scraper rather than license aggregator Yodlee, citing privacy concerns about Yodlee's handling of user credentials. The company raised $4.7M across two funding rounds, reached profitability on revenue alone for nine months between November 2009 and July 2010, and shut down in July 2010. Hedlund's post-mortem "Why Wesabe Lost to Mint" attributed the loss in significant part to user-experience differences arising from the data-aggregation choice. Source for the launch order, the Yodlee decision, the funding history, and the founder's post-mortem framing. Hedlund's post-mortem is preserved at Medium (originally published at Precipice). Additional company history and timeline verification at Failory.

[4] Mint launched September 2007 at the inaugural TechCrunch 40 conference and was acquired by Intuit in November 2009 for $170 million. The product reached approximately 20 million active users at its 2016 peak, declining to 3.6 million by 2021 according to Bloomberg. Intuit announced the shutdown in November 2023 and completed the sunset on 23 March 2024. Users were directed to migrate to Credit Karma, also owned by Intuit, which preserved net worth and a subset of transaction history but did not replicate Mint's budgeting, transaction rules, or custom categorisation. Source for the acquisition price, peak user count, and the Credit Karma migration gap. Bloomberg shutdown coverage at bloomberg.com/intuit-winds-down-mint. CNBC user-reaction coverage at cnbc.com/budgeting-app-mint-is-shutting-down.

[5] Money Dashboard was a UK personal finance manager founded in 2010, regarded as the UK's largest PFM by 2015 with over 100,000 users, and powered by Yodlee for account aggregation. ClearScore Group acquired Money Dashboard in May 2022 in a deal whose terms were not disclosed. ClearScore subsequently shut down the Money Dashboard Neon and Classic consumer apps on 31 October 2023, with the company's official closure statement reading: "we could not find a sustainable business model for the apps, and are therefore unable to continue supporting the services." ClearScore retained the open-banking technology that powered Money Dashboard and rebuilt it as a B2B product called D•One, launched in January 2023, sold to lenders for transaction categorisation and open-banking connectivity. Acquisition coverage at dailybusinessgroup.co.uk/clearscore-buys-money-dashboard. Closure announcement coverage at moneytothemasses.com/money-dashboard-closing. D•One launch coverage at openbankingexpo.com/clearscore-d-one-launch.

[6] Quicken was founded in 1983 and is widely considered the first consumer personal finance tracker. Intuit sold the product to H.I.G. Capital in 2016 after Intuit CEO Brad Smith called it a "desktop-centric business" that "doesn't strengthen the small business or tax ecosystems," contributing $51 million to Intuit's $4.2 billion total revenue at the time. H.I.G. resold the business to Aquiline Capital Partners in 2021. Quicken moved to a mandatory subscription model in 2018, with users locked in a proprietary file format and recurring complaints about deteriorating integration quality across the user community. Source for the Intuit divestiture and the "dead end" framing at computerworld.com/intuit-sells-quicken. Aquiline transaction at techcrunch.com/quicken-resold-private-equity.

[7] Personal Capital, founded in 2009, offered a free auto-aggregating personal finance dashboard alongside paid wealth-management advisory services. By 2020 the platform had over 2.5 million users tracking $771 billion in household assets, with 31,800+ advisory clients holding $28.5 billion in managed assets. Empower Retirement, the second-largest US retirement plan provider, acquired Personal Capital for up to $1 billion enterprise value in 2020. The free dashboard was rebranded as Empower Personal Dashboard in 2023 and continues to operate as a customer acquisition channel for Empower's wealth-management business. Acquisition announcement at empower.com/empower-retirement-completes-personal-capital-acquisition.

[8] CoinTracker is a cryptocurrency portfolio and tax-reporting tool founded in 2017 by Jon Lerner and Chandan Lodha. The company reached unicorn status with a $1.3 billion valuation after a $100 million Series A in January 2022, and as of late 2025 reports approximately three million users across 100+ countries with integrations to 500+ exchanges, wallets, and chains. CoinTracker positions itself as a portfolio management platform first and tax software second, with the business funded primarily by crypto-tax compliance subscriptions, which are regulatory-mandatory in most jurisdictions. The IRS published Revenue Procedure 2024-28 in June 2024, requiring per-wallet cost-basis tracking starting in tax year 2025 (replacing the previous universal-account method), which substantially increased the regulatory burden on individual crypto holders and made automated cost-basis tracking effectively mandatory rather than optional. Total funding raised stands at approximately $101.6 million, with no Series B reported as of April 2026, four years after the Series A. CoinTracker is a private company that does not publish audited financials. Series A coverage at coindesk.com/cointracker-raises-100m. Funding history at crunchbase.com/organization/cointracker. Revenue Procedure 2024-28 at irs.gov/pub/irs-drop/rp-24-28.pdf.

[9] Open Financial Exchange (OFX) is an open data format for the exchange of financial information including bank transactions, credit card statements, and investment account data. Microsoft, Intuit, and CheckFree announced the OFX standard on 16 January 1997, and version 1.0 of the specification was released on 14 February 1997. Versions from 2.0 onwards are XML-based. The OFX investment message set covers security identifiers (CUSIP, ISIN, ticker), positions and quantities, transaction history (buys, sells, dividends), prices, and holdings. In 2019 the OFX consortium joined the Financial Data Exchange (FDX) consortium, which now maintains the OFX specification; the latest reference document was published in October 2020. While many US banks support OFX for bank-account transaction export, support for the OFX investment message set across consumer portfolio products is sparse: the open-source ofxtools Python library documentation explicitly notes that "full support of the OFX investment message set has been somewhat neglected by the Python community." OFX specification at financialdataexchange.org. OFX history at en.wikipedia.org/wiki/Open_Financial_Exchange. ofxtools investment message set note at ofxtools.readthedocs.io.

[10] Plaid is the dominant US bank-data aggregation infrastructure layer that most modern fintech and personal finance products depend on. Visa announced an agreement to acquire Plaid for $5.3 billion in January 2020, describing the company as "an essential infrastructure for the future of financial services." The US Department of Justice filed an antitrust lawsuit on 5 November 2020 challenging the acquisition under Section 2 of the Sherman Act, citing internal Visa communications that described Plaid as an "insurance policy" against threats to Visa's online debit business. Visa abandoned the acquisition in January 2021. Plaid raised a Series D at a $13.4 billion valuation in April 2021, with revenue concentrated in enterprise APIs sold to banks, lenders, and fintech infrastructure companies. As the aggregation layer that most consumer fintech products depend on, Plaid's commercial trajectory determines what data flows are possible across the consumer fintech category, regardless of any individual consumer product's intentions toward its users. DOJ complaint coverage at techcrunch.com/doj-antitrust-visa-plaid. Acquisition abandonment at cnbc.com/visa-plaid-call-off.

[11] Yodlee was the original consumer-facing personal financial management aggregation infrastructure, founded in 1999 and providing the data aggregation for Mint, Money Dashboard, and a generation of other personal finance products. Yodlee IPO'd on Nasdaq in 2014 and was acquired by Envestnet in November 2015 for $660 million in cash and stock, after which Yodlee was repositioned away from consumer-facing products toward enterprise wealth-management and bank-facing infrastructure. Envestnet itself was taken private by Bain Capital in a $4.5 billion transaction, and announced in June 2025 that Yodlee would be sold to private equity firm STG, with the deal expected to close in Q3 2025. Yodlee's pivot from consumer aggregation to enterprise infrastructure is one example of the broader pattern in which the aggregation layer that supports personal finance products migrates upmarket toward institutional buyers because the consumer-facing economics will not sustain it independently. Envestnet acquisition completion 8-K at sec.gov/Envestnet-Yodlee-completion. STG acquisition coverage at americanbanker.com/envestnet-yodlee-stg. Industry context at kitces.com/envestnet-yodlee-acquisition.