Why do most firms struggle to get results from AI tools?

iManage's 2026 Knowledge Work Benchmark found that 85% of legal organizations are piloting or implementing AI, but only 17% have fully integrated it. The gap isn't the AI itself — it's the underlying infrastructure. Firms that got AI working first organized their documents, standardized their file naming, set access controls, and built a knowledge layer their AI tools could actually use. Firms still struggling usually have AI tools sitting on top of disorganized, inconsistent, or inaccessible document environments. The AI can only work with what it can find and access.

What does 'knowledge infrastructure' mean for a small professional services firm?

Knowledge infrastructure is the organized foundation your AI tools run on: your document storage structure, naming conventions, access permissions, and how consistently your team saves work to shared systems vs. personal drives or email. For a 10-person accounting firm, this might mean ensuring all client workpapers are in a standardized folder structure in your practice management system. For a law firm, it means matters are organized consistently in your document management system (NetDocuments, iManage, or even a well-structured SharePoint) so that AI tools can find and use the right documents. Without this, AI tools waste time on irrelevant documents, miss key files, and produce inconsistent results.

What is iManage MCP and when will it be available?

iManage announced in early 2026 that it will launch Model Context Protocol (MCP) integration in H1 2026. MCP allows AI tools — including Harvey, Legora, and Microsoft Copilot — to connect directly to iManage's document management system and access firm documents in context. This means an AI assistant can retrieve and reason over actual firm documents (contracts, precedents, case files) rather than requiring manual uploads. For firms already using iManage, MCP will significantly improve AI output quality. For firms not on iManage, the announcement is still a signal: AI tools work better when documents are organized in a system they can connect to.

What should a small firm do before buying another AI tool?

Run a 5-question infrastructure check before the next AI purchase: (1) Can your team find a specific client document in under 60 seconds without searching their own email? (2) Do you have a consistent folder structure that everyone follows? (3) Are your documents stored in a shared system (not personal drives or email attachments)? (4) Do you have access controls so the AI can't accidentally surface confidential documents to the wrong person? (5) Can you describe what types of documents you have and what formats they're in? If the answer to any of these is no, fix the infrastructure first. Buying more AI tools won't fix an organization problem.

Does this apply to accounting, consulting, and staffing firms — not just law?

Yes. The iManage research is primarily from the legal sector, but the infrastructure gap is identical across professional services. An accounting firm that has partner workpapers scattered across personal drives, email attachments, and a shared server with inconsistent naming will get the same poor results from AI tools as a law firm with disorganized document management. Consulting firms that store deliverables in project folders no one can consistently locate will find AI research assistants mostly useless. The principle is universal: AI amplifies whatever organization you already have. Good organization + AI = good results. Poor organization + AI = expensive frustration.

AI Infrastructure Gap: Why Tools Fail Firms

Published: March 2026 | By: The Crossing Report | 6 min read

Here's a conversation I've had with three different firm owners in the past few months. Different industries, different sizes, different AI tools — but the same story.

They bought an AI tool. They tried it for a few weeks. The results were inconsistent. Sometimes it worked well, sometimes it seemed to hallucinate or miss obvious things. After two months, usage dropped to one or two people still experimenting. The monthly subscription is still running.

"I think we're not using it right," one said.

They were right. But not in the way they meant. The problem wasn't how they were prompting the AI. The problem was what the AI had to work with.

The 85/17 Gap

In March 2026, iManage published its Knowledge Work Benchmark Report with a data point that should stop every firm owner cold: 85% of legal organizations are piloting or implementing AI. Only 17% have fully integrated it.

That's not a gap in AI tools. That's a gap in the underlying infrastructure those tools run on.

The firms in the 17% — the ones getting real results — had something in common before they bought the AI. They had organized, accessible, consistent document environments. Their files were findable. Their naming conventions were followed. Their knowledge — precedents, templates, past work product, client history — was stored in systems their AI tools could actually use.

The firms in the 85% had the same AI tools. They just put them on top of disorganized document environments and wondered why the results were inconsistent.

iManage is primarily a document management platform for law firms, but the finding mirrors what I've heard from accounting and consulting firm owners who've tested AI tools: the AI is only as good as what it can find.

What "Infrastructure" Actually Means for a 10-Person Firm

You don't need enterprise document management to fix this. You need four things:

1. Documents in one shared place — not email, personal drives, or three different systems

If your team stores work in personal Dropbox folders, email attachments, a shared server, and a practice management system — the AI can't connect all of those. Pick one primary home for client work. Enforce it.

2. A consistent folder structure that your team actually follows

AI tools work best when they can navigate predictable structure. If your folders are organized differently for every client or every partner, the AI spends its effort figuring out where things are instead of analyzing what they contain.

3. Access controls that match your professional obligations

For law firms: attorney-client privilege means your AI tool should not be able to surface Client A's documents when a partner is asking about Client B. For accounting firms: client tax information should not bleed across engagements. Most AI tools respect the permissions you've set in your underlying document system — but only if you've set them.

4. A complete client file — not a partial record

AI tools summarize and analyze what's there. If you've been using the AI file for current documents but storing notes and correspondence in your personal email, the AI will give you an incomplete picture and you'll wonder why it missed something obvious.

This isn't glamorous work. It's not the kind of thing that makes a good demo. But it's the difference between the 17% and the 85%.

The iManage MCP Signal

In early 2026, iManage announced it will launch Model Context Protocol (MCP) integration in H1 2026. This allows AI tools — including Harvey, Legora, and Microsoft Copilot — to connect directly to iManage's document system and access firm documents in context, rather than requiring manual file uploads.

For iManage users, this is significant. An AI assistant that can pull the right precedent or contract version from your actual document system, rather than requiring you to find and upload it, will produce dramatically more useful results.

For firms not on iManage, the announcement is still a signal: the entire direction of professional services AI is toward deeper document integration. Tools are being built to work with organized, structured document environments. Firms that build that environment now will extract far more value from every AI tool they buy — now and over the next three to five years.

The 5-Question Infrastructure Check

Before you buy another AI tool, run this check. If you can't answer yes to all five, fix the infrastructure first.

1. Can anyone on your team find a specific client document in under 60 seconds without searching their own email?

If the answer is "sometimes" or "depends on the person," that's a no.

2. Do you have a consistent folder structure that everyone follows for client matters?

If partners organize files differently from associates, or if client folders look different depending on who set them up, that's a no.

3. Are your documents stored primarily in a shared, searchable system — not personal drives or email?

Email is where knowledge goes to die for AI purposes. If your team primarily stores work in email or personal cloud storage, the AI can't get to it.

4. Have you set document access permissions that match your professional obligations?

For law firms: conflicts, privilege, and matter confidentiality. For accounting: client data segregation. For consulting: engagement confidentiality clauses. These controls belong in your document system, not just in people's intentions.

5. Can you describe the types of documents you have and what formats they're in?

If you don't know what's in your own knowledge base, neither does the AI.

Where to Start This Week

Law firms: Run a document audit on your three most recent matters. Where are all the files? List the locations — shared drive, email, personal desktop, client portal. Consolidate to one primary location. Set that as your standard for the next matter you open.

Accounting firms: If you use a practice management system (Karbon, Canopy, Financial Cents), set a rule: all client documents go into the matter file, not email. Enable the document storage features you've been ignoring.

Consulting firms: If client deliverables are scattered across project folders with inconsistent names, spend two hours standardizing your folder template. Apply it to your next engagement from day one.

Staffing firms: Your knowledge infrastructure problem is usually candidate data — split across ATS, email, spreadsheets, and phone notes. Pick your ATS as the system of record. Everything else feeds into it.

Marketing agencies: Client assets and briefs living in a mix of Google Drive folders, Slack threads, and email attachments is the agency default. It doesn't have to be. One client folder in your project management tool, always.

The AI tools will still be there next quarter. The firms that fix their infrastructure first will get better results from the same tools than firms that don't.

You Bought the AI Tool. Why Isn't It Working? The Infrastructure Gap Firms Keep Hitting.