Then SORA moves from near-zero to 3.7% over twenty months.

Expectedly, the REITs sell off. The banks were supposed to hold. Instead they move down too, repriced alongside everything else as the market simultaneously processes rate risk, recession risk, and a global portfolio rotation out of yield-sensitive equity. Your two largest positions, the ones supposed to be doing different things, start moving in the same direction. At the worst possible time.

You weren’t wrong about the fundamentals. You were wrong about what your portfolio would actually do under stress. And the tools you had weren’t built to tell you that. The number your correlation matrix shows you is an average, optimised for an environment that doesn’t exist during the periods that matter most.

What AI adds to this isn’t better mathematics. It’s the interpretation layer: the step between quantitative output and investment decision that, at most boutique AMs in Singapore, currently lives inside someone’s head or doesn’t exist at all

I built a tool to test how far this could go. Here’s what it found.

Click here for the Dashboard:

1) The risk story behind the stats

The math of portfolio risk management is universal. Any quant with a terminal can easily produce a correlation matrix or a factor exposure report. More sophisticated funds have access to Bloomberg PORT and FactSet, but the calculations have never been an issue.

Understanding what the output means and what to do about it is a different story altogether. You need experience, judgement, and ideally enough bandwidth to run your back-testing. In a firm with $500M AUM and four investment professionals, it’s easy for this to slip through the cracks.

AI Analysis

The three local banks — DBS, OCBC, and UOB — are moving together tightly enough that holding all three is closer to a single-factor bet on Singapore financial conditions than a diversified position, with the OCBC-UOB pair at 0.73 and DBS-OCBC at 0.72 representing the portfolio’s most acute hidden concentration.

Layered on top of this, the two industrial REITs (Mapletree Industrial and CapitaLand Ascendas) are co-moving at 0.72 as well, meaning the portfolio effectively carries two dense correlation clusters — banks and industrial REITs — that the headline average correlation of 0.30 flatters by blending in genuinely diversifying names.

That low average is almost entirely driven by Bumitama Agri (BS6.SI), which is near-orthogonal to virtually everything else in the book, including SGX and Genting Singapore, suggesting it is doing outsized diversification work that the position sizing may not reflect.

The PM should watch whether the bank cluster correlation compresses or widens during the next MAS policy or Fed rate decision, as that is the event most likely to activate both clusters simultaneously and expose the true undiversified risk.

What the AI layer does is close the distance between the output and the decision. Instead of a heatmap the PM files away because there isn’t bandwidth to interpret it, the tool generates a specific, portfolio-calibrated narrative: that your two largest sector positions are behaving as one risk factor under stress; that your rate sensitivity is not a fixed property but a regime-conditional one that was at its highest precisely when SORA was moving; that the hedges which held during 2020–2021 Bull conditions showed meaningfully different behaviour during the periods they were actually needed.

Connecting the quantitative pattern to the investment implication, in plain English, calibrated to the specific portfolio in front of you, is the layer that doesn’t exist at most boutique AMs in Singapore. Not because the firms lack intelligence, but because producing it previously required a quant to run the analysis and a senior analyst to translate it. The tool collapses that chain. The numbers are the same. The infrastructure required to act on them is not.

2) Where AI replaces the missing analyst

The next (and more useful) question that separates infrastructure from an expensive dashboard is: what do you actually do about these stats?

Most tools stop at the diagnosis. They show you the risk profile. They don’t help you navigate a 50-name universe to find what reduces it. And they don’t apply any judgment about whether the candidates they’d theoretically suggest are actually appropriate for an SGX portfolio in practice.

This is where the architecture diverges. Based on your preferences – maximizing Sharpe, reducing volatility or tail risk – the model can recommend changes to your existing portfolio. It’s not just optimizing weightages of existing holdings, but introducing and removing assets that historically boost the portfolio.

Here’s how it works:

Pass 1: AI doing analyst work

Before any heavy computation runs, Claude receives the portfolio’s full risk profile — factor exposures, regime-conditional correlations, CVaR, concentration flags — and a universe of 50 SGX names. Metadata only: sector, market cap tier, approximate average daily volume, liquidity flag.

From that alone, it shortlists 12 addition candidates and 6 removal candidates. One-sentence rationale per name, referencing the specific portfolio gap being addressed.

This isn’t a keyword filter or a sector-matching exercise. Claude applies contextual knowledge that no spreadsheet encodes: which names are family-controlled and carry governance risk that a correlation matrix won’t surface; which have ADV below SGD 1M and create real-world exit friction that makes theoretical optimisation irrelevant; which sectors appear different on a label but are structurally correlated to what the portfolio already holds.

This is the work a senior analyst does before running the numbers. At a boutique AM without that analyst, it previously didn’t happen. The construction analysis started from the full universe with no filter, which meant it either ran slowly, ran wrong, or didn’t run at all.

Pass 2: Python validates

The marginal contribution analysis runs only on the Pass 1 shortlist — not all 50 names. For each addition candidate, it simulates adding at 5%, funded by proportional trimming. For each removal candidate, it simulates removing and redistributing weight. All metrics are recomputed.

The optimisation targets stress-period behaviour specifically, not five-year averages. Candidates are scored on their correlation to the portfolio during Bear and Crisis regimes – exactly where diversification actually needs to work.

Three scenarios, one synthesis

Output: Conservative (1 add, 1 remove), Moderate (2+2), Aggressive (3+3). Before/after metrics for each. Then a 100-word Strategic Signal: the consistent finding across all three scenarios, regardless of transition magnitude.

The tool isn’t a stock picker. It’s a structured framework for having a better conversation about portfolio construction — one that starts from how the portfolio actually behaves under stress, not how it looks on a sector breakdown.

The infrastructure argument

Return to early 2022. Same portfolio. But before the rate cycle began, you ran this analysis.

Factor exposures showed net negative rate sensitivity. The rolling chart showed it had been rising for twenty months. Regime-conditional correlations flagged that your bank and REIT positions, which looked like a hedge on average, would converge under stress. Construction scenarios identified three SGX names - one industrial, two with genuinely low stress-period correlation to the existing book – that would have reduced that sensitivity without concentrating liquidity risk in thin-volume names.

Would it have prevented the drawdown? No.

But you would have gone in knowing exactly what you were carrying. With a quantitative basis for it, not just an intuition. And with specific, validated options for reducing the exposure that you could take to your investment committee with the reasoning already structured.

That’s the actual value proposition. Not prediction. Not automation of the investment decision. Infrastructure for thinking more clearly about risk before it becomes a problem.

The comparison that matters: producing this analysis manually — regime detection, rolling factor regressions, stress-period marginal contribution across 50 names, plain-English interpretation of every output — would take a quant analyst one to two days. This took twenty minutes. The analysis is identical. The infrastructure required to produce it is not.

What comes next

If you’re a PM or CIO at a boutique AM, family office, or PE firm in Singapore thinking through what AI-assisted risk infrastructure should look like at your scale — not Bloomberg PORT, not a Python script that requires a quant to run — I’d like to hear from you. The architecture is more accessible than most people assume. The harder problem is calibrating it to how your team actually makes decisions, and that’s best worked through with someone who understands both sides.

You read what you can find, call whoever you know and hope you haven’t missed anything important. You ask ChatGPT for help but triple-check every stat – the last time it made up a fake number that nearly cost you your job. Deep into your fifth coffee and third set of eyedrops, you’re sixty percent sure you have everything you need. Sixty percent. For a potential multi-million deal.

For Singapore AMs, this is part and parcel of the research cycle.

The latest AI models are incredibly intelligent. Ask covenant structures in private credit, or REIT capital recycling dynamics in Southeast Asia, and you’ll get a very impressive answer. But intelligence on its own is overrated, especially when you’re trying to get real work done.

Frontier models have 2 real problems. They makes things up, treating fake numbers like gospel truth. And they don’t know anything about your private data either. That’s why most PMs tend to be sceptical about what AI can do for them.

Enter RAG.

Retrieval-Augmented Generation (RAG) combines the best of both worlds – state-of-the-art reasoning and your own internal documents: board packs, deal memos, IC minutes, broker reports etc. It makes intelligence actually useful. The end output doesn’t just know finance in general, but your portfolio specifically.

To test it out, I built a RAG chatbot trained on Keppel Data Centre’s official financial and operational reports. I was testing for accuracy, and (more importantly) whether it would tell me if it didn’t have the data I was asking for.

Here’s what I found.

Click here to see the Dashboard:

Click here to see the Documentation:

What makes RAG useful

Before I get to the build, two properties of this architecture are worth understanding.

First, it significantly reduces hallucination risk. LLMs can make things up, particularly when they don’t know something. For a consumer chatbot that’s annoying. For an IC memo or a compliance decision it’s a serious liability. RAG constrains the model to answer from documents you’ve explicitly provided. Every claim can be traced back to a specific source, page, and paragraph. That traceability matters enormously when someone asks you to justify a call.

Second, nobody else sees your documents. When you use a public LLM interface and paste in a board pack or an internal research note, you’re sending that content somewhere. Depending on the provider and the settings, it may be used to improve the model. For a consumer query that’s fine. For non-public information about a private portfolio company, it isn’t. RAG keeps the knowledge base on your own infrastructure. The model reasons over your documents without storing them or training on them. You get the intelligence layer without the data exposure. Clean separation.

You can also layer access controls. A junior analyst queries public filings. A PM queries everything including restricted deal memos. A compliance officer has a view tuned to covenant documents and regulatory records. One knowledge base, segmented access.

What I actually built: Keppel Data Centres

To test how useful this architecture could be in practice, I built a RAG system over Keppel DC REIT. A well-covered SGX name with enough public documentation to stress-test the retrieval pipeline.

The knowledge base has three document types: financial reports from Keppel DC REIT, SGX filings and historical price data. The mix of structured financial tables and time-series price data is a reasonable proxy for what an AM would actually deal with across their listed holdings.

Some examples of what the system handles well:

“Based on the last three annual reports, is KDC’s debt maturity profile getting longer or shorter — and what does management say about refinancing risk?” — tests cross-document synthesis and whether the system can connect a quantitative trend to a qualitative commentary

“Which markets in KDC’s portfolio have seen the highest occupancy growth, and how does that compare to where they’ve been deploying capex?” — requires joining operational data across multiple periods

"What is the consensus view on Keppel DC's market performance?" — the system declined. No broker reports in the knowledge base, so rather than guess, it said so directly and pointed to what it did have. That's exactly the behaviour you want from a system handling investment decisions. A RAG setup that knows the boundary of its own knowledge is more useful than one that fills the gap with something plausible-sounding.

Retrieval works across heterogeneous document types. The system handles the transition between a financial table, a paragraph of management’s narrative, and a price series without losing coherence. That’s the foundational capability before extending to something more complex.

Where it gets really interesting: public and private, unified

Consider a Singapore AM holding two significant positions. One is Keppel Data Centres. The other is a Series B fintech in Indonesia.

Keppel has everything: Bloomberg page, SGX filings, MAS disclosures, four years of REIT financials, a dozen broker notes. The challenge is pulling the relevant signal from a large and fragmented document set quickly and reliably.

The Indonesian fintech has none of that. No Bloomberg page. No public filings. What the firm does have are a hodgepodge of data: quarterly board packs, emailed analyst notes, scanned pdfs of loan agreement, a site visit memo the PM wrote in 2023, the list goes on.

RAG is what makes this internal data intelligent, transforming a static document archive into something you can actually interrogate, cross-reference, and derive structured insight from.

The same RAG pipeline that indexes Keppel’s public filings can ingest the fintech’s board packs. Same query surface. Same interface. From the analyst’s perspective, they’re exactly the same. Ask a question and get a sourced answer, regardless of whether the underlying company is publicly listed or not.

The practical result: a cross-portfolio query that no Bloomberg subscription answers. “Which of our holdings - public and private - have revenue concentration in the Indonesian consumer segment?” currently requires an analyst to manually trawl multiple disconnected environments. With a unified RAG setup, it’s one query.

Others are already doing versions of this

The architecture isn’t speculative. Larger institutions are already running versions of it, and the gap is widening.

Morgan Stanley deployed an internal GPT-powered assistant over more than 100,000 research documents which can surface relevant content in natural language rather than keyword search. The result isn’t just speed. It’s institutional memory that doesn’t walk out the door when an analyst leaves.

BlackRock has been more public about where Aladdin is heading: AI-assisted research synthesis layered on top of existing portfolio analytics. The direction of travel is the same: frontier reasoning grounded in proprietary data.

Closer to home, DBS has built internal AI tools over their own document sets. A bank operating across multiple Southeast Asian markets, with all the regulatory complexity that entails, found it worth the investment. That’s a useful data point for any Singapore-based firm thinking about the compliance dimension.

What these firms have in common is a decision to treat their internal document library as infrastructure rather than an archive. The AI layer on top is almost secondary.

What this looks like day-to-day

The Keppel DC and Indonesian fintech example is simple. Two companies, two document environments, one unified query. But the underlying architecture touches nearly every core process an AM runs, because every core process has the same problem at its heart: someone needs to find the right information, from the right document, at the right time.

Earnings deep-dives. IC memo drafting. Cross-portfolio exposure checks. Covenant monitoring on private credit positions. Regulatory watch across MAS and SGX circulars. The specific workflow doesn’t matter much.

What they all have in common is a retrieval problem that currently gets solved by an analyst spending time they don’t have, searching through documents that aren’t organised for searching.

RAG doesn’t change the judgement call at the end. It compresses everything that happens before it.

RAG will be table stakes in three years. The question is who builds it first.

Think about what Excel looked like twenty years ago. The firms that built rigorous, well-structured models early didn’t just work faster. They developed a fluency with their own data that compounded over time. The model got better as more history went in. The analyst who built it understood the business more deeply because structuring the data forced precision.

By the time Excel proficiency became an assumed baseline, those firms had a head start measured in years, not months.

RAG is at that same inflection point now.

Within three years, having a RAG system over your firm’s internal knowledge base will be as standard as having a Bloomberg Terminal. Not because it’s a nice-to-have, but because the competitive pressure from firms that have built it will make the gap impossible to ignore.

And that gap will continue to grow. A better LLM released in 2027 immediately makes your 2024 document library more powerful. The firms that treated their internal documents as a strategic asset – indexed, structured, access-controlled – will sit on infrastructure that compounds in value as the models on top of it improve. The investment in the data layer pays forward

For mid-sized Singapore AMs specifically, the window is narrow but real. The institutional players – BlackRock, Morgan Stanley – have already moved. The question is whether the gap closes further before local firms act.

What comes next

I built the Keppel DC system to understand the architecture before extending it to private company document sets - the board packs and deal documentation that represent the knowledge layer most AMs aren’t yet querying systematically. That’s the next phase.

If you’re working through similar questions – what document types to prioritise, how to handle messy private company reporting, how to build access controls that actually hold – I’d like to hear from you. The infrastructure is more accessible than most people assume. The harder problem is workflow design, and that’s best solved with people who understand how an investment team actually makes decisions.

Something has quietly changed in how the best-resourced investment teams do research - and if you haven’t heard of AlphaSense yet, you likely will soon. Founded in 2011, it has grown sharply in the last few years, crossing 5,000 corporate and financial clients globally as of 2024 .

The pitch to a lean buy-side team is simple: stop reading transcripts manually and let AI surface the signals for you. Pricing runs $20–40k per year - enough to make any COO pause.

I decided to find out how much of it was actually replicable. I built a version of one of its most interesting features, measured how close I could get, and hit a wall I didn’t expect. Here’s what I found.

Try the Dashboard here:

See the Technical documentation:

What AlphaSense Does

AlphaSense is an enterprise AI search and market intelligence platform built for financial professionals. At its core, it lets users search across a vast corpus of documents: earnings transcripts, broker research, SEC filings, news etc. using natural language rather than keyword strings.

For a $2–5B AUM manager with a lean team, this directly empowers its analysts to spend less time reading and more time, well, analysing.

Pricing sits in the $20–40k per year range depending on seat count and data modules. To understand how reasonable this is, here are some comparable platforms:

Bloomberg Terminal (~$27,000/seat/year) aka. the one no one can live without. Real-time market data, trading tools, quantitative screens, news, and fixed income analytics. Its depth on structured financial data and execution workflows remains unmatched.

FactSet (~$12,000/seat/year): strong on financial modelling, Excel integration, and structured data. It particularly ‘Excels’ in investment banking with superior Excel integration and financial modelling capabilities.

Factiva (~$5,000/year) from Dow Jones covers news and media aggregation at the lower end of the price range. It offers extensive media coverage but with less sophisticated search technology and minimal AI integration.

In a very competitive landscape, AlphaSense’s value proposition is simple: it fills the gap none of the above close cleanly. Bloomberg gives you data. FactSet gives you models. Factiva gives you news. AlphaSense gives you meaning in the form of AI-native qualitative research and earnings intelligence, wrapped in a modern interface a lean buy-side team can actually use without a dedicated data team behind it.

In particular, a feature that caught my attention was earnings tone and transcript sentiment analysis. The idea is that management language is a leading indicator that most quantitative models don’t capture. Before guidance gets cut, before consensus moves and (hopefully) before price action reflects it, there is often a change in how management speaks.

More hedging. Less conviction. Shorter answers under analyst pressure.

AlphaSense tracks this systematically. Smart synonym matching across quarters. Sentiment scoring on prepared remarks vs. Q&A. Theme tracking across peer companies. I wanted to know: how much of this is actually replicable with off-the-shelf tools and a few evenings of work?

Three Things an Analyst Actually Cares About

Before building anything, I mapped out use cases a PM or senior analyst would actually be interested enough to pay for.

Guidance softening before consensus moves. Management rarely says “we’re cutting guidance.” They say “we remain cautious given the evolving macro environment” or “we’ll have more visibility by next quarter.” Detecting that language shift systematically - across 40–80 names - is nearly impossible to do manually with any consistency.

Analyst pushback under the surface. A single quarter of hedged Q&A language might be noise. But when analyst questions grow sharper and management answers grow shorter across multiple quarters, that’s a pattern - and one that rarely shows up in sell-side notes. Tracking it manually requires someone to re-read six transcripts per name and score them subjectively.

Q&A tone diverging from prepared remarks. When management’s prepared statement is confident but their Q&A answers are evasive or heavily qualified, it often means analysts are pressing on something management would prefer not to discuss directly.

What I Built, and What Happened with Apple in Q2 2025

The stack is straightforward. I used Claude to handle the language analysis and took transcript data directly from EarningsCall, which offers Apple and Microsoft data for free.

Output is organised across four tabs. The Overview tab gives you an at-a-glance read: a management confidence score, a forward guidance score, and an analyst pushback level.

Trends tab plots these scores quarter-over-quarter so you can see movement, not just a snapshot.

Themes and Hedging tab breaks down the specific language driving the scores - hedging density, recurring themes, and the terms the model flagged as significant.

Ask the Data tab is a simple chatbot interface where you can ask natural language questions directly against the transcript: “What did management say about capital expenditure?” or “How did the CFO respond to the margin question?”

A good example of interesting output was Apple in Q2 2025. The confidence score dropped sharply, notable precisely because Apple’s management is typically very consistent in how they communicate. What drove it was a cluster of signals arriving together. Hedging density in the guidance section was elevated. Responses to analyst questions on the tariff impact were noticeably shorter and more qualified than Apple’s historical baseline. These were phrases oriented around uncertainty rather than execution.

This reflected something visible in the broader market at the time: tariff paralysis. Apple’s supply chain exposure to China meant that management didn’t know how to make sense of the geopolitical volatility, and that uncertainty bled into how they spoke publicly. The tool caught it.

Of course, Apple having a low confidence score means nothing, at least in isolation. But the real value is twofold: being able to read signals without the entire transcript, and being able to run inter-company or inter-industry comparisons over a multi-year period, quantified into a signal that can be fed into an analytical model.

Total cost to run this analysis: a few cents in API credits per transcript. Against a $30k annual license, the math is hard to ignore.

If the data is available, that is.

The Wall I Didn’t Expect: Asia Data

Let me be honest about what my tool is and isn’t. The intelligence layer - signal extraction from a clean transcript - holds up well.

The operational reality around it does not: reliability, maintenance, auditability, and compliance are real problems that require engineering and full-time attention. A production-grade version of this is buildable, but it is months of focused work away, not evenings. None of that is the most interesting gap, though.

The biggest problem has nothing to do with AI: many companies in Asia simply don’t publish transcripts. Some offer video recordings, and others might do it in in a local language. It’s completely unreliable and unstructured.

What AlphaSense has spent years building is not primarily an AI system. It is a system of relationships with IR teams, transcript providers and document sources, which allows them to get the right data as quickly and accurately as humanly possible. That infrastructure is the real moat.

For a Singapore-based manager with meaningful exposure to Asian names, the enterprise platform has a real, defensible moat. And it’s worth paying for - not because of the AI layer, but because of what sits underneath it.

A Decision Framework for Boutique Managers

Rather than a blanket buy-vs-build recommendation, the more useful question is: what does your portfolio actually look like?

If 80% or more of your names are US-listed large-caps: the build case is strong. Transcript coverage is reliable, English-language, and increasingly available through low-cost APIs. A technically capable hire - or even a motivated analyst willing to learn - could replicate the core earnings intelligence layer for a fraction of the enterprise cost. The AI layer is not the barrier.

If you run significant Asia exposure: the enterprise platform has a genuine case, but do your homework. The coverage gap for Asian names is real, and many intelligence providers tend to be more US-centric.

A dual approach might make sense too. Build the US-name capability internally and use an alternate provider selectively for Asian coverage, reducing seat cost while preserving the relevant coverage.

The question to ask yourself is not “is AlphaSense good?” It clearly is. The question is: “Am I paying for the AI, or the data?”

Where AI Moats in Finance Actually Sit

This experiment sharpened something I’d been thinking about loosely.

The intelligence layer of these platforms: the LLM, the sentiment model, the natural language interface etc. is rapidly becoming a commodity. The underlying AI is no longer the hard part.

Running a capable language model costs cents per transcript and is accessible to anyone with basic technical skills. Any reasonably capable developer can build a system that extracts structured signals from a transcript with impressive accuracy. The gap between a well-prompted Claude API call and an enterprise sentiment model has narrowed considerably in the last 18 months.

What is not commoditising is the data infrastructure underneath. Proprietary transcript archives. Normalised historical coverage across markets and document types. IR relationships that mean your corpus updates in real time. Compliance-grade audit trails that a regulated fund manager can point to. These are hard to replicate, and they’re what the best platforms in this space have quietly been building for a decade while the AI layer was an afterthought.

For any asset manager thinking about the next three years of vendor evaluation: the question to ask is not “how good is your AI?” Every vendor will tell you their AI is excellent, and increasingly, they’ll all be roughly right. The question is: “How clean and complete is your underlying data, and do you have coverage in my universe?”

The platforms that survive the next wave of commoditisation are not AI businesses that happen to have data. They are data businesses that happen to use AI.

By 9am, he has read - really read - maybe four of them.

The other ten will be skimmed. Some will be missed entirely.

This is not a talent problem. It is a bandwidth problem. And for the first time in the history of asset management, there is a structural solution available that does not involve hiring more analysts.

The asset management industry in Southeast Asia is at an inflection point. The signals are visible to anyone paying attention: Singapore’s MAS is actively funding fintech and AI adoption through initiatives like the Financial Sector Technology and Innovation scheme. Regional sovereign wealth funds are not just allocating to AI; they are deploying it internally. Global firms opening regional offices are arriving with AI toolchains already built.

The question for mid-sized regional asset managers is not whether AI will reshape the investment process. It already is. The question is whether your firm will lead that shift or spend the next three years catching up to firms that did.

This newsletter exists to make that question easier to answer.

What “AI in Asset Management” Actually Means

There is a version of this conversation that happens at a lot of firms. Usually held in a strategy meeting, or triggered by someone reading a consultant’s report.

AI gets discussed as a future initiative. Something to pilot or perhaps form a committee around.

That version of the conversation is already obsolete.

The firms building an edge right now are not piloting AI as an add-on to their existing process. They are rebuilding the investment workflow itself around a set of complementary tools - what I call the AI ensemble.

Not one model. Not one vendor. A coordinated stack, each component doing what it does best, integrated into the moments that matter most in the investment cycle.

The chart below shows what that looks like across the three core processes that every asset management firm runs, every day.

The framework breaks the investment workflow into three distinct processes.

Synthesize: From Raw Data to Investment Thesis

This is where most of an analyst’s day goes. Reading. Parsing. Cross-referencing. Forming a view.

The traditional toolkit is well-known: Bloomberg for data pulls, S&P Global and FactSet for structured financial information, internal databases for historical positioning. The output is an analyst who has read enough to have an opinion, and a process that scales exactly as well as the number of analysts on the desk. Which, for many mid-sized asset managers, isn’t much.

The AI ensemble changes the input-to-insight ratio fundamentally. Tools like Alphasense and Daloopa are not search engines bolted onto research libraries. Alphasense uses language models trained on financial text to surface thematic connections across earnings calls, broker notes, and regulatory filings - connections that would otherwise require hours of manual reading to identify.

Daloopa automates the most tedious part of fundamental analysis: the extraction and normalisation of financial data from unstructured documents. What used to take a junior analyst an afternoon now takes minutes, with lower error rates.

Exabel and Yipit extend this into alternative data - the signals that do not appear in company filings at all. Consumer transaction data, foot traffic, web scraping, supply chain proxies. These were once the exclusive domain of large quant shops with proprietary data budgets. The commoditisation of access changes that calculus for mid-sized active managers.

For Southeast Asia specifically, there is a capability that does not often appear in Western discussions of AI in finance: cross-language pipelines.

A Singapore-listed consumer staples company with significant Indonesia operations will generate material information in Bahasa Indonesia. Management interviews, local press and regulatory commentary, all of which are rarely seen by most analysts at English-language firms. LLMs with multilingual capability change that. The competitive edge available to a regional firm that can actually read Southeast Asian markets - in the languages those markets operate in - is significant and underappreciated.

The outcome is not that AI replaces the analyst’s judgment. It is that the analyst’s judgment gets applied to a dramatically richer information set. Three hours of reading compressed to fifteen minutes. Quant-style factor screening available without a dedicated quant hire.

Execute: From Thesis to Live Portfolio

This is where the gap between intention and outcome has historically lived.

The Investment Committee meets. A view is formed. The portfolio manager knows what the target allocation should look like. What happens between that decision and the actual portfolio - the sequencing of trades, the timing within each session, the management of market impact - is where slippage accumulates.

The traditional execution model relies on portfolio management systems like Bloomberg PORT, discretionary trader judgment on timing, and standard algorithms (VWAP, TWAP). They do not adapt to intraday liquidity conditions. They do not dynamically rebalance around transaction cost estimates. And they do not optimise across the interaction effects of multiple concurrent positions.

The AI execution layer changes all three of these.

SimCorp Dimension and Clearwater Analytics provide the portfolio-level infrastructure: real-time position reconciliation, exposure modelling, constraint monitoring. But it is the algo execution layer that represents the sharpest change. JPMorgan’s LOXM is the best-known institutional example of ML-optimised execution: a model trained on millions of historical trades that selects and parameterises execution strategies dynamically, based on real-time market microstructure.

The result is measurable reduction in implementation shortfall. Not marginal, but structural over time.

For most mid-sized regional asset managers, the direct access to a system like LOXM is constrained by prime brokerage relationships. But the principle applies across accessible alternatives, and the direction of travel for algorithmic execution is unambiguous: static VWAP as the default execution strategy will look increasingly primitive within five years.

The more immediate opportunity is at the Snowflake layer - using cloud data infrastructure to close the loop between IC decision-making and portfolio reality in real time. The problem many mid-sized asset managers have is not that they make bad decisions; it is that the portfolio does not fully reflect their decisions because the execution and reconciliation cycle is too slow. Real-time data infrastructure, paired with optimisation-driven rebalancing tools, closes that gap.

What the IC decides, the portfolio actually reflects in real time. That is the outcome that matters.

Anchor: From Live Portfolio to Institutional Edge

This is the process that most AI discussions in asset management underweight, and where I believe some of the most durable competitive advantages will be built.

Anchoring covers everything that happens after the trade. Compliance monitoring, performance attribution, portfolio review, and the institutional memory that turns individual decisions into organisational knowledge. In the traditional model, this is largely backward-looking and manual: daily and weekly reports, periodic performance reviews, manual compliance checks against investment guidelines.

The costs of this model are less visible than they should be. Compliance breaches caught in the monthly report rather than the hour they occur. Post-mortem analysis that never actually happens because the team is too busy preparing for the next IC meeting. Institutional knowledge that lives in the heads of two or three senior people - and walks out the door when they leave.

BlackRock’s Aladdin has anchored large institutional portfolios for years, providing the risk modelling and compliance infrastructure that most mid-sized regional firms cannot build themselves. But Aladdin is primarily a risk system, not a knowledge system.

The newer tools in this layer are Hebbia and Kensho, paired with custom RAG implementations over internal research archives. They represent something different: the beginning of genuine institutional memory at the AI layer. Hebbia allows firms to build retrieval systems over their own document corpus: past investment memos, trade rationale notes, IC minutes, research archives going back years.

The ability to ask a question of your firm’s own history - “what was the investment committee’s view on Indonesian consumer exposure in 2019, and what happened to the positions that reflected that view?” - is a capability that no junior analyst on the team has, and that the senior PM who was there may not accurately recall.

Real-time breach detection, autonomous compliance copilots, systematic post-mortem workflows: these are not features. They are the difference between a firm whose institutional knowledge compounds over time and one that resets every time a key person leaves.

Why This Matters for SEA Asset Managers Now

The ensemble is not a future roadmap. Every tool named in this framework is available today, to firms willing to engage with the integration work required to make them interoperate.

That integration work is real. It is the part that consultants’ slide decks tend to understate. Connecting Alphasense outputs to an internal research workflow, building a RAG layer over a proprietary document archive, configuring real-time data pipelines between execution and portfolio management systems - none of this is plug-and-play.

But the firms that do it are not building a marginal improvement. They are building a different kind of investment operation: one where analysts think at higher abstraction levels, where the portfolio actually reflects IC intent, and where institutional knowledge does not decay when people leave.

In a regional market where the talent pool for investment professionals is finite and the competition for assets under management is intensifying - from global firms with larger balance sheets and from robo-advisory platforms attacking the retail segment - operational alpha is not a nice-to-have. It is a strategic imperative.

The next issues of Buyside AI will go deep on each of the three processes: how the synthesise stack actually gets built, what the execution layer looks like in practice for a regional active equity fund, and how to think about the anchor layer as a knowledge infrastructure problem rather than a compliance problem.

If you are building this at your firm, or thinking about how to start, I want to hear from you.

Buyside AI covers AI adoption across the investment workflow, with a focus on mid-sized active managers operating in Southeast Asia. If this issue was useful, forward it to one colleague who should be reading it.