A Salesforce VP told Fortune last week that enterprises waste five to ten times more tokens cleaning up unstructured inputs than running the actual AI work. Cheaper models won’t fix that. The pipeline is leaky before the model even sees the data.

Financial advisory firms hit this wall hard. You’ve got client data in Xplan, emails in Outlook, meeting notes in OneNote, portfolio snapshots in PDFs, and KYC documents scattered across shared drives. When you ask an AI agent to draft an SOA or prep a client review, it chokes on the mess. The token meter spins while the model tries to reconcile three different spellings of the same client name, parse a scanned fact-find from 2019, and figure out which version of the risk profile is current.

The result is slow, expensive, and wrong often enough that your paraplanners stop trusting it. The problem isn’t the AI. It’s the data you’re feeding it.

The Real Cost of Unstructured Client Data

Most advisory firms don’t track how much time goes into wrangling client information before advice work can start. We see it in three places.

Meeting prep takes five to ten hours per adviser per week. You pull the portfolio snapshot, scroll through recent emails, check the last file note, and try to remember what the client said about their daughter’s university plans. Then you write a one-page brief so you don’t walk into the Zoom call cold. That’s billable time spent on admin.

Compliance documentation costs $3,000 to $8,000 in paraplanner time per advice document. SOAs and ROAs require pulling data from multiple systems, cross-checking it, formatting it into the firm’s template, and writing narrative that ties it all together. Cycle times stretch into weeks because the paraplanner is bouncing between systems and waiting for clarification.

Client onboarding drags on for 30 to 60 days. You send the fact-find, chase missing documents, re-key everything into Xplan, run the risk profile, and assemble the onboarding pack. New clients lose momentum. Some don’t make it to the first advice meeting.

The common thread is unstructured data. Client information exists, but it’s not in a form an AI agent can use. So you pay humans to structure it, or you pay the AI to fumble through it and burn tokens.

Why Cheaper Tokens Don’t Solve the Pipeline Problem

The Fortune article quotes a Salesforce VP explaining that firms focus on model cost while ignoring the preprocessing tax. If your input data is messy, the model spends most of its context window trying to make sense of it. You’re paying for cleanup, not insight.

Financial advisory firms see this when they try to use ChatGPT or Copilot to draft client communications. You paste in a meeting transcript, some portfolio data, and a few email threads. The model gives you something generic because it can’t tell what’s current, what’s relevant, or what the client actually cares about. You rewrite it yourself.

The token cost isn’t the issue. The issue is that you’re asking the AI to do data engineering on the fly. It’s like handing a chef a bag of unsorted ingredients and asking for a three-course meal in ten minutes. The chef can cook, but first they have to figure out what’s in the bag.

Advisory firms that want AI agents to work reliably need to audit and structure their client data before they feed it to the model. That means knowing where each data type lives, how clean it is, and whether it’s tagged in a way the agent can query. It means deciding which systems are the source of truth and which are just noise.

We call this the data foundation layer. It’s not exciting, but it’s the difference between an AI agent that saves your team ten hours a week and one that creates more work than it eliminates.

What Clean Data Looks Like for an Advisory Firm

Clean data doesn’t mean perfect data. It means structured enough that an AI agent can find what it needs without human intervention.

For client records, that means one canonical profile per client in your CRM or practice management system. Name, contact details, family structure, goals, risk profile, and current portfolio holdings all in consistent fields. If you’ve got duplicate records or clients with multiple spellings, merge them now.

For meeting notes, it means storing transcripts or structured summaries in a searchable format with timestamps and tags. If your advisers are still writing notes in Word docs named “Client Meeting 2026,” an AI agent can’t help you. It doesn’t know which meeting, which client, or what was discussed.

For compliance documents, it means templates with clearly defined sections and metadata. If your SOA template is a 40-page Word doc with manual formatting and no field tags, the AI can’t pull the right data into the right spots. You need structured templates that map to your data sources.

For portfolio data, it means automated feeds from your platform into your CRM. If your advisers are downloading PDFs and re-keying numbers, you’re adding friction every time the AI needs current holdings.

This is the work that happens before you turn on the agent. It’s the audit, the cleanup, and the integration layer that lets the agent run without constant human supervision. See Omni for financial advisory firms to see what that audit looks like in practice.

How AI Agents Use Structured Data

Once your data is clean, agents can do the work you hired them for. Let me walk through three examples we’ve built for advisory firms.

The Meeting Prep Agent pulls data from your CRM, email, and portfolio platform the morning of every client meeting. It generates a one-page brief with recent portfolio performance, upcoming goal milestones, recent communications, and any flags from the last meeting. Your adviser reads it in three minutes and walks into the call prepared. No more scrambling through systems at 8:50 AM.

The agent works because it knows where to look. Client profile in Xplan, emails in Outlook, portfolio data from the platform API, meeting notes in your structured repository. It queries each source, filters for relevance, and formats the output. If the data is messy, the agent can’t do that. It either guesses wrong or asks the adviser to clarify, which defeats the point.

The Advice Document Agent drafts SOAs and ROAs from meeting transcripts and your compliance template. It pulls the client’s current situation from the CRM, maps it to the template sections, writes narrative based on what was discussed in the meeting, and flags any gaps for the paraplanner to fill. Cycle time drops from two weeks to three days because the paraplanner is editing, not drafting from scratch.

This agent needs clean templates and structured meeting data. If your template is a free-form Word doc, the agent can’t tell where the client situation section ends and the strategy section begins. If your meeting notes are unstructured, the agent can’t pull out the key decisions. The data structure is what makes the agent useful.

The Client Onboarding Agent runs a guided fact-find with new clients, collects KYC documents, and prepares a clean onboarding pack for the adviser. It sends the fact-find link, reminds the client about missing documents, validates the inputs, and writes a summary. Your adviser gets a complete onboarding pack in five days instead of six weeks.

This agent works because it’s connected to your CRM and document storage. It knows which fields are required, which documents you need for compliance, and where to store everything so the adviser can find it. If your onboarding process is email-based and manual, the agent can’t help. It needs structured workflows and clean handoffs.

All three agents depend on the same thing: data that’s structured, current, and queryable. That’s the foundation. Without it, you’re back to paying humans to wrangle information or paying the AI to guess.

The Omni Audit Finds the Gaps

Most advisory firms don’t know how clean their data is until they try to use it. You think your client records are in good shape because they’re in Xplan, but then you discover 15 percent of clients have outdated risk profiles, 30 percent are missing email addresses, and half the meeting notes are stored outside the system.

The Omni Audit is a 60-minute session where we map your data sources, identify the gaps, and show you what needs to happen before you can run agents reliably. We don’t sell you software in that session. We give you three outputs: a data map, a prioritized cleanup list, and a blueprint for the first agent you should build.

We run this audit for financial advisory firms every week. The common pattern is that firms have more data than they realize, but it’s not connected. Client records in one system, emails in another, documents in a third. The AI agent can’t bridge those gaps on its own. You need integration work first.

The audit also shows you which workflows are ready for automation and which need more cleanup. Meeting prep is usually the easiest win because the data is already in your CRM and email. Advice documents take longer because the templates need work. Onboarding is somewhere in between.

Book a 60-min Omni Audit and we’ll walk through your systems. You’ll leave with a clear picture of what’s blocking your AI agents and what to fix first.

Why This Matters for Your P&L

The leaky pipeline problem isn’t abstract. It shows up in your cost structure.

If your advisers are spending eight hours a week on meeting prep and admin, that’s 400 hours a year per adviser. At $200 per hour of opportunity cost, that’s $80,000 per adviser in time that could go to client-facing work or new business development. For a firm with five advisers, that’s $400,000 a year.

If your paraplanners are taking two weeks to turn around an SOA, you’re capping how many advice documents the firm can produce. At $5,000 per document, the bottleneck costs you revenue. Speed up the cycle by cleaning the data and using an Advice Document Agent, and you can handle 30 percent more volume with the same team.

If your onboarding takes 45 days, you’re losing clients who get cold feet or find another adviser. Typical conversion from first meeting to signed client is 60 to 70 percent for firms with fast onboarding, and 40 to 50 percent for firms with slow onboarding. The difference is $50,000 to $150,000 in annual recurring revenue for every ten prospects.

These numbers are why we focus on the data foundation. Cheaper AI models don’t move the needle if the pipeline is leaky. Clean data does.

What Happens After the Audit

The audit gives you the roadmap. The next step is execution. For most advisory firms, that means three phases.

Phase one is data cleanup. Merge duplicate client records, fill in missing fields, tag meeting notes, and structure your compliance templates. This takes four to eight weeks depending on how much legacy data you’re carrying. It’s not glamorous, but it’s the work that makes everything else possible.

Phase two is integration. Connect your CRM, email, portfolio platform, and document storage so agents can query them. This is where Omni Ops comes in. We build the connectors and the agent logic so your Meeting Prep Agent can pull from five systems and deliver one clean brief.

Phase three is agent deployment. You start with one workflow, usually meeting prep or client onboarding, and run it for 30 days. You measure time saved, error rate, and user adoption. If it works, you expand to the next workflow. If it doesn’t, you adjust the data structure or the agent logic.

Most firms see measurable time savings within 60 days of starting phase one. The savings compound as you add more agents and more workflows. By month six, the typical firm is saving 15 to 25 hours per adviser per week and cutting compliance cycle times in half.

The firms that move fastest are the ones that treat this as a data project first and an AI project second. They audit, they clean, they integrate, then they deploy. The firms that struggle are the ones that buy an AI tool, plug it into messy data, and wonder why it doesn’t work.

The Salesforce Lesson for Advisory Firms

The Fortune article is a reminder that the AI hype cycle skips over the boring infrastructure work. Everyone wants to talk about GPT-5 and cheaper tokens. Nobody wants to talk about data hygiene and integration layers.

But the firms that win with AI are the ones that do the boring work first. They know where their data lives, they know how clean it is, and they know which workflows are ready for automation. They don’t wait for the perfect model. They build the foundation and ship agents that work.

If you’re running a financial advisory firm and you’re thinking about AI, start with the audit. Map your data, find the gaps, and prioritize the cleanup. Then build one agent that solves a real problem for your team. Measure it. If it works, build the next one.

The alternative is to keep paying humans to wrangle data or keep paying the AI to fumble through it. Both are expensive. Both are slow. Both are avoidable.

Book my Omni Audit and we’ll show you what’s leaking in your pipeline. Sixty minutes, three outputs, no deck. You’ll know what to fix and what to build. Then you can decide if you want our help or if you’d rather do it yourself.

For more on how advisory firms are using AI agents to automate compliance and client work, visit the AI audit for financial advisory firms or explore the broader Omni platform we’ve built for professional services. If you want to see how other firms are thinking about AI adoption, check out the EDNA insights library for case studies and tactical guides.

The data foundation isn’t optional. It’s the difference between AI that works and AI that wastes your money. Clean your data first, then feed it to the model. That’s the only way the pipeline stops leaking.