AI Won't Fix a Broken Firm: It Just Makes the Chaos Faster

Integrating AI into a firm that has unreliable intake processes, manual handoffs, and disconnected systems doesn't resolve these issues. Instead, it magnifies problems, operating them faster but with less visibility. Firms that achieve measurable returns from AI initiatives first redesign their workflows, then automate them, and finally apply AI to areas requiring rapid judgment, following this order. Skipping the initial step is the primary reason most AI projects falter.

---

The Importance of Process First

This is not just an opinion; it's supported by research:

• Workflow redesign has the strongest correlation with financial returns from AI of 25 factors evaluated in McKinsey's 2025 State of AI survey, surpassing budget, talent, and model choice. Unfortunately, around 70% of organizations skip this step.
• High-performing companies are nearly three times more likely to significantly modify their workflows rather than merely add AI to existing ones (McKinsey, 2025).
• About 95% of enterprise generative AI pilots yielded no measurable bottom-line impact in a 2025 MIT study. Most of these failures resulted from brittle workflows and poor alignment with daily operations, rather than the AI model itself.
• Only 18% of professionals report their firms track AI ROI, and 40% are unsure if it's even measured (Thomson Reuters, 2026). Many companies are adopting AI without clear success criteria.

Signs of a Broken Process in Professional Services

You likely have a process issue, not an AI issue, if:

• Intake relies on the availability of individuals, causing response times to vary from same-day to a week later. Eligibility or conflict checks occur late in the process.
• Your tools are disconnected, requiring manual data entry between practice management, CRM, billing, and spreadsheets, leading to mistrust in monthly numbers.
• Handoffs are based on memory, with work transferring between people through notes that may or may not be written.
• The founder serves as the sole integration layer, with every non-standard case routed through one person's knowledge.
• There is no single source of truth, resulting in three different answers regarding how many active matters the firm has.

Applying AI to this type of environment only accelerates the distribution of incorrect answers, automates follow-ups on poor data, and introduces a layer that becomes impossible to audit.

How AI Can Worsen a Broken Process

• It obscures problems. A manual process reveals failures through complaints, whereas an automated process can fail silently until a client reacts.
• It increases error rates. If 1 in 10 intakes is mishandled today, automating intake without addressing the underlying issues means the same error rate just occurs faster and at a higher volume.
• It cements existing problems. Once an AI workflow is integrated into a flawed process, untangling it becomes more complex than the original manual version.
• It erodes trust internally. Teams may observe the “AI project” producing little value, leading to resistance against future automation efforts.

The Effective Sequence for Implementation

1. Map and quantify.

   Identify every workflow from the initial inquiry to the final invoice. Note where work waits for individuals and quantify the costs of delays in hours, write-offs, and lost inquiries.

2. Redesign the workflow.

   Eliminate steps that exist due to tool limitations or outdated habits. Decide who is responsible for each step and what constitutes “done.” This is where most of the value lies and is often free.

3. Connect the systems.

   Enable automatic data flow between core tools to avoid duplicate entries. Establishing a single source of truth is necessary for downstream processes.

4. Automate the deterministic parts.

   Focus on tasks with fixed rules, such as triggers, routing, reminders, document requests, and status updates. This is genuine automation and yields the majority of time savings.

5. Apply AI where judgment intersects with volume.

   Use AI for drafting initial documents, summarizing lengthy records, extracting data from unstructured files, and triaging inbound requests. At this stage, the AI operates on clean inputs within a defined process, allowing you to measure its effectiveness.

Quick Definitions

• Automation: Rule-based processes that operate without human intervention, such as triggers and reminders. This forms the backbone of efficiency, implement this step first.
• AI: Systems that handle language, judgment, and unstructured data. These capabilities are applied once the process has been streamlined.
• RPA (Robotic Process Automation): Bots that mimic user actions across unintegrated applications. This is a temporary solution when systems cannot connect, but fragile and should be avoided if proper integration is possible.
• A System: The intentional combination of tools and processes tailored to your workflows. This is the ultimate goal; the tools should serve as components rather than the primary focus.

Conclusion

AI is a valuable and practical tool; however, it should be the final step, not the first. Improve the process, integrate tools, and automate rules before introducing AI. This way, you create a solid foundation for AI to build upon, allowing you to demonstrate its positive impact.