Welcome back. This week, the foundations. Not the deals and not the boundary lines: the layers underneath them. Asset managers have spent the summer quietly rebuilding the foundations of their operating models. The data layer first: Rathbones, Impax and Pictet all announced or progressed governed data platforms within weeks of each other. Then the dependency layer: the UK designated its first four critical third parties, and the EU's register-of-information cycle became an annual, enforced routine. Then the servicing layer: BNY brought fully native tokenised fund servicing to market. None of this makes front pages. All of it determines what your firm can build for the next decade.

Reconciliation automation starts paying measurable dividends. Duco's case study with DWS Asset Management is the clearest public evidence yet that data automation in the middle office is delivering hard numbers. DWS's multi-year move off legacy reconciliation platforms lifted match rates from 60–65% to 80%, with 85% now the target. Bhaskar Raj, who leads reconciliation operations for APAC and EMEA, is blunt about why: "If in the next two years our business grows twice the size of what we are today, can we do the same operations with the same set of people?" The migration also surfaced what the case study politely calls shadow IT: Excel macros running reconciliation processes that were "fragile, opaque, and difficult to audit." The stated test for the platform was whether Ops could build and change rules itself, rather than queueing behind IT. For DWS, the answer was yes, and the cost-income ratio is the metric they track to prove it.

The replatforming wave reaches the mid-tier. Wasatch Global Investors, a $23bn Salt Lake City manager, went live on Ridgeline on 4 August after more than three decades on legacy systems. Two details matter. First, the framing: the move is about "redirecting toward higher-value work" and handing off infrastructure management (servers, patching, uptime) entirely. Second, the vendor's number: Ridgeline customers replace an average of nine disconnected systems with one platform. Nine. That is a statement about how fragmented the average mid-size manager's stack still is, and about who owns the integration risk when nobody owns the data model.

Cost pressure becomes the operating model's forcing function. Formidium's Achim Denkel, writing in Hedgeweek, names three structural trends: relentless fee compression, regulatory complexity, and a talent market where experienced fund operations professionals are scarce and expect technology to remove repetitive work. His conclusion: "operational efficiency is becoming a strategic differentiator rather than simply a cost centre." His diagnosis of the status quo is worth pinning to the wall: many operations teams "still spend too much time moving data between systems, reconciling information and preparing reports rather than creating value." Cost pressure is not new. The claim that the operating model, not the product shelf, is where the margin gets defended, is becoming the consensus.

Something bigger than any single deal is happening beneath the platform market. Managers are rebuilding the data layer itself.

The benchmark number: Cutter Associates' 2026 outlook reports that 67% of firms responding to its Data Management Benchmarking Survey are changing their data management solution or planning to, with 20% planning a proprietary cloud solution. Cutter's advice is in the title: don't reinvent, integrate or partner. Its reasoning is the interesting part: AI "has brought the need for high-quality data to the forefront as never before," and the platform's job is to provide golden source data to downstream applications, risk engines, compliance tools and, now, AI models.

The live case study: Rathbones Asset Management's COO Stephen Wood tells IT Pro how a two-year programme on Snowflake built a "golden source of data for each of our foundational domains," now nine domain data sets covering performance, risk, sustainability and unstructured information. The sequence deserves study: a point-to-point share from Charles River first, then static fund reference data, then the harder domains. Agentic pilots on Snowflake's CoWork and CoCo are running in what Wood calls "a near-production environment" on millions of rows. The data foundation came first; the agents came second.

The governed-AI pattern: Impax Asset Management's selection of ICS's ATHENA platform in July makes the linkage explicit. The £22.3bn manager is consolidating positions, transactions, sustainability data, pricing and reference data into "a single, governed source of truth," with ATHENA MCP letting investment and operations teams query live data in plain language, every interaction logged, traceable and read-only, running inside Impax's own Azure tenant. That is the template the industry is converging on: data governance first, AI access second, never the reverse.

And the platform vendors have noticed. Pictet Asset Management extended SimCorp's remit in June to replace fragmented, disconnected data sets with a unified front-to-back data layer, moving from batch to real time. When the front-to-back platform story becomes a data layer story, the battleground has shifted.

Architecture Canvas lens: the data layer is becoming the first layer of the operating model, the layer every other layer reads from. Last issue was about where work sits. This issue is about what work sits on. Every process, every control, every AI agent eventually asks the same question: which number do I trust? The firms that can answer it without a meeting are the ones that rebuilt the foundation. The firms that cannot will discover, when their first production agent reconciles against a shadow spreadsheet, that the foundation was the project all along.

BNY takes tokenised fund servicing operational. On 29 July, BNY launched its Digital Transfer Agency capability, extending TA services to digitally native funds across jurisdictions and blockchains. The structural claim is significant: where tokenised funds are issued on-chain from the start, "the legal title and economic value of the funds exist on-chain rather than remain in mirror-token or 'twin' models." On-chain books and records, fiat and stablecoin subscriptions and redemptions via new mint/burn capabilities, all connected to BNY's traditional TA recordkeeping. The first client is already live: the Baillie Gifford Enhanced Yield Fund (BAGEY) is, per BNY, the first publicly available, fully native UK-regulated tokenised fund. BlackRock is expected to use the same capabilities to launch BSTBL, a tokenised money market share class designed for stablecoin reserve requirements. The operations implication: the books and records question, which has always been the anchor of fund servicing, is now a design choice between on-chain and twin models, and the major servicers are building both rails.

National Bank of Canada chooses Multifonds to consolidate fund accounting. In an 11 August announcement, Multifonds confirmed it has been selected by National Bank of Canada, one of Canada's six systemically important banks, to modernise its fund accounting operations and support ETF growth. The stated objective: "replace siloed systems with a centralized platform" for operational resilience, faster product onboarding and consistent service delivery, with real-time processing, exception-driven workflows and more than 350 user-definable controls. ETF NAV operations are where volume, T+1 compression and complex fund structures collide, which makes them the most honest test environment for fund accounting platforms right now.

4Pines turns fund administration into a shared operating environment. The next generation of 4Pines' C2P platform, announced 29 July, is built on a simple idea: every task, from initial email to final deliverable, moves through one system visible to administrator and client simultaneously. Emails become structured tasks, validations execute inside the workflow rather than on spreadsheets beside it, and a templated AI assistant called Acorn answers questions against the firm's own live fund data. The administrator stops being a black box that reports after the fact and becomes a shared, auditable operating layer. Expect every administrator to be forced to answer the same design question within two years.

The dependency layer now has supervisors on both sides of the Channel.

UK: the Critical Third Parties regime is live. On 13 July, the Bank of England, PRA and FCA began joint oversight of the first four designated critical third parties: AWS, Google Cloud, Microsoft and Oracle, designated by HM Treasury under the Critical Third Parties (Designation) Regulations 2026. The regime is explicitly complementary: the regulators' announcement is careful to state that firms "remain responsible for managing risks arising from their third-party suppliers." For operations teams, the practical consequence is subtle but real: the cloud concentration question has moved from a procurement concern to a supervised one, and the evidence you keep about your own dependencies, exit plans and concentration risk now sits next to a regime that assumes the same dependencies are systemically important.

EU: DORA's register becomes an annual routine, and the subcontracting rules bite. The first full annual cycle of register-of-information submissions closed this spring; the CSSF, for example, required submissions between 11 February and 31 March 2026, with the ESAs' annual reporting process now set on a 31 March rhythm. Meanwhile the subcontracting RTS (Commission Delegated Regulation (EU) 2025/532), in force since 22 July 2025, requires firms to map subcontracting chains for ICT services supporting critical or important functions, with contractual audit, notification and termination rights through the chain. Deloitte's 2025 DORA European Survey found only 8% of surveyed institutions consider themselves fully compliant on ICT third-party risk, and notes the EU has now designated 19 critical ICT third-party providers for direct oversight. EY's framing is the right one for this audience: the question has shifted from "how do we comply?" to whether the dependency map is a living document. For asset managers, the register is not an IT artefact. It is an inventory of every ICT dependency behind NAV production, dealing, reporting and client service, and it now has an annual deadline, validation rules and a supervisor on the other end.

Moves from the past two months.

This week's operating model insight: The Golden Source Clause.

Every operating model carries a hidden clause: the model is only as good as the data it reads from. Most asset managers never signed it consciously. The data layer grew organically: custodian files, administrator extracts, vendor feeds, a few spreadsheets with macros nobody remembers writing. Then three things happened at once. AI arrived and made data quality a board topic. The regulators began supervising dependencies rather than just asking about them. And T+1 compression made the tolerance for multiple versions of the truth disappear.

The firms in this week's Change Watch understood the clause. Rathbones built the golden source first, then piloted the agents. Impax wrapped its governed data in a read-only AI interface. The sequence is the whole lesson.

Three things to do:

The firms treating the data layer as a plumbing project will find their AI strategy blocked by it. The firms treating it as the first layer of the operating model will be able to point at any number and say: that's where it comes from, that's who owns it, and here's how we know it's still right.

This week's companion piece on opmodal's Insights page goes deeper: The golden source is an operating model decision, not a technology project.

Target State is published weekly by opmodal. Subscribe free at target-state.beehiiv.com. Look out for the paid tier opening this autumn.

Architecture Canvas is a central component of the methodology developed by opmodal for capturing, governing, and improving operating models in complex organisations.