The Deck That Built Itself — and Automated a $30,000 Workflow

Each quarter, the firm's Investment Policy Committee required a comprehensive Meeting Book. The deck had to cover market index performance across U.S. domestic, international, emerging markets, and fixed income. It needed model portfolio returns with risk metrics including alpha, beta, Sharpe ratios, and capture ratios. It required SMA manager performance versus benchmarks, asset allocation tables, and risk-reward scatter charts with capital market lines.

Building this manually meant four grinding steps: data collection, spreadsheet work, PowerPoint assembly, and error checking. The entire process consumed roughly 40 hours to complete properly. The firm employed a part-time analyst at approximately $30,000 per year whose primary job was producing these decks.

One detail captures how broken the process really was. Morningstar tables and Excel-built index grids couldn't be easily exported into PowerPoint with their formatting intact. So the analyst would take screenshots — large screenshots of dense data tables. These had to be manually compressed to fit the deck's slide layout, producing visibly distorted images: blurry numbers, pixelated text, and color banding. The firm's Investment Policy Committee, stewards of client assets, was reviewing fiduciary recommendations on slides that looked like photocopies of faxes.

It wasn't incompetence. It was a rational response to a toolchain that offered no clean path from data to presentation. The analyst wasn't bad at their job. The workflow was.

The Merger Complication

A merger between the firm and another advisory practice created an unexpected problem. A miscommunication about who would maintain the data subscription meant that when it came time to build the Q2 2026 deck, the advisor found himself without access to the platform every prior quarterly deck had depended on.

He had two choices: manually rebuild from whatever free sources he could find, or find a better way.

AdvisorClaw is an AI agent platform that combines large language models with file system access, code execution, and browser automation. The advisor described what he needed in plain English: a quarterly IPC deck built from free public data, matching the format and analytical depth of the original. AdvisorClaw understood the domain, asked clarifying questions about periodicity and benchmark methodology, and built the pipeline. No code was written by a human.

The result is a fully automated pipeline: live data pulls from Yahoo Finance for all index ETFs, benchmark proxies, and individual holdings; automated computation of weighted model returns, benchmark composites, and quarterly risk metrics; a professionally styled deck rendered directly to a 16:9 landscape PDF with institutional-grade typography; and a modular data architecture where quarterly updates require changing only two numbers.

The comparison is stark. Same analytical content, same committee, same fiduciary responsibility. But one looks like a compliance document from 2003, while the other looks like it came from an institutional design team.

Producing index data without a paid subscription

The original deck had always sourced index data from a paid analytics platform. Without access, the only viable free alternative was Yahoo Finance, which provides ETF total return data but not direct index-level returns for most benchmarks.

The challenge was to map each of the roughly 40 indices in the deck to the best available ETF proxy. This wasn't as simple as picking the most popular ETF for each index. AdvisorClaw conducted an analysis evaluating each candidate on exposure fit, tracking error from fees, and data availability.

Some indexes presented unique challenges. MSCI AC Far East has no direct ETF, so AdvisorClaw built a synthetic proxy using a 60/40 blend of EWJ and AAXJ. Russell 2500 Growth/Value have no dedicated ETFs, so VBK and VBR were selected as closest available proxies. The Credit Suisse Hedge Fund Index was discontinued, so AdvisorClaw stitched together EHFI 400 data with QAI ETF returns to create a continuous proxy series.

The resulting ETF proxy map was validated against known index data and locked in. It produces index-level returns that match paid-platform data within basis points. The firm no longer needs the subscription.

The alpha/beta discrepancy

During the build, AdvisorClaw detected something odd. Alpha and beta figures for the same models appeared differently on different slides of the Q1 manual deck. On the model performance summary page, the Growth Model showed an alpha of -1.36 and beta of 0.41, but the risk-reward scatter chart pages implied different numbers.

The initial assumption was a methodology difference, but AdvisorClaw's investigation revealed something simpler and more concerning. Different slides were using different periodicity. Some calculations used monthly return observations, while others used quarterly. Neither was mathematically wrong, but they produced meaningfully different results.

AdvisorClaw standardized everything to quarterly frequency across all pages. Every alpha, beta, Sharpe ratio, and capture ratio is now computed identically — from the performance summary table to the scatter charts to the appendix methodology notes. The discrepancy was invisible to the human eye; only by cross-referencing all calculations did AdvisorClaw spot it.

Proactive improvements: narrative summaries

The original manual deck's market index pages were dense tables of returns: nineteen rows of U.S. domestic indices, sixteen rows of international and emerging market indices, and seven rows of fixed income indices. Useful data, but it asked the reader to interpret raw numbers without context.

Without being asked, AdvisorClaw added narrative summaries above each index table. This wasn't a generic AI making lucky guesses. AdvisorClaw was purpose-built for financial advisors and wealth managers. It understands what an IPC committee needs: not just numbers, but the story behind them.

These narratives transform index pages from reference tables into briefing documents. Committee members get context before they dive into the numbers. This feature did not exist in any prior quarterly deck.

Cutting the fluff

Rebuilding the deck from its data sources rather than copying the previous quarter's PowerPoint revealed something the advisor had suspected but never had time to address. The 57-page manual deck contained significant redundancy: separate SMA and mutual fund model books, individual watch list pages for dozens of funds, and full-page manager profiles that added limited value to the committee conversation.

The automated deck consolidates everything into 21 pages. Same analytical content, zero filler. The advisor described it as “what the deck should have been all along.”

The transformation is quantifiable across every dimension that matters: time, cost, quality, and analytical depth. The automated deck doesn't just replicate the original — it improves on it in ways that weren't possible with manual assembly.

But the qualitative shift matters more. Every risk metric uses the same methodology, every table uses the same formatting, and every page follows the same grid. The result looks like it came from an institutional design team — no cross-page inconsistencies, no copy-paste mistakes, and no outdated period labels. And 21 pages deliver more actionable insight than the original 57.

The risk-reward scatter charts are another example of what automation enables. In the manual deck, these charts were only available when sourced from the paid analytics platform. Without that subscription, the Q2 deck would have lost a key analytical tool. Instead, AdvisorClaw built them from the ground up — with capital market lines, consistent quarterly methodology, and professional presentation.

In the process of automating the deck, AdvisorClaw constructed monthly return series for every model holding. It geometrically linked individual security and SMA returns into model-level returns, and these series extend back to each model's inception, reaching 2008 and 2009 for the legacy models.

This created something that didn't exist before: a structured, queryable database of the firm's model performance history. Before, historical returns were scattered across old quarterly decks, Excel files, and SMA provider statements. Now the advisor can ask natural-language questions and get instant answers.

This was never the goal of the project, but it's become one of the most valuable outcomes. The quarterly refresh process is now trivially simple: receive two numbers from the SMA provider, feed them to AdvisorClaw with a single prompt, and wait roughly five minutes for the PDF to render.

The work that consumed a part-time role costing approximately $30,000 per year is now automated. The firm also avoids the cost of the data platform subscription. And it gained a queryable database of historical performance that turns ad-hoc analytical questions into instant answers.

The advisor's first test of AdvisorClaw's analytical depth wasn't the deck. It was a manager research exercise. He asked it to find the best managers in large cap, mid cap, and small cap — the kind of due diligence that typically takes his team six hours and requires paid data subscriptions. AdvisorClaw did it without accessing any of them. In each category, the managers it recommended matched what his team had already identified internally, sometimes exactly, sometimes closely. And it surfaced one new name worth investigating.

The story isn't about replacing people with software. It's about eliminating work that shouldn't exist in the first place: manual data pulls, screenshot gymnastics, cross-page error checking, and formatting that degrades with every copy-paste cycle. The analyst wasn't bad at their job. The workflow was. And now the workflow is gone.