Do Not Boil the Citrus: Why Law Firms Should Redesign the Workflow Before Adding AI

Charles Austin Klein
Jul 11
5 min read

Updated: 5 days ago

Disclosure: This article was developed by Charles Austin Klein, AI Strategist, with the assistance of Gemini, ChatGPT, and Claude for drafting and structural refinement. This work draws upon my professional experience in designing and implementing secure, closed-loop AI content, including the development of custom-configured GPTs (OpenAI). All strategic arguments, privacy frameworks, and recommendations regarding data security and privilege preservation were independently verified and validated by the author to ensure compliance with professional standards. The author maintains full responsibility for the final content.

Disclaimer: I am an AI Integration Strategist, not a lawyer. This article addresses operational controls, technical risk management, and data governance. It is not legal advice. Canadian Law Firms should adapt the proposed controls to their jurisdictions, clients, practice areas, systems, contractual commitments, insurance requirements, and tolerance for risk.

Tall sailing ship on rough sea at sunset, framed by lantern, oranges, compass, scales, files, and circuit icons in a moody collage

In 1747, Royal Navy surgeon James Lind tested several treatments on 12 sailors suffering from scurvy. The sailors who received oranges and lemons showed the clearest improvement. Yet the Royal Navy did not make lemon juice standard issue until 1795 — 48 years later [1][2].

The episode illustrates two ways institutions mishandle useful innovation:

They give greater credibility to complicated theories than to a practical intervention that appears too simple.
They force the intervention into existing systems in ways that weaken its value. Lind advocated preserving citrus juice as a concentrated syrup known as a rob. The heating used in its preparation reduced the vitamin C on which its effectiveness depended [1][3].

Small law firms face both risks with artificial intelligence. Some assume meaningful adoption must begin with enterprise platforms or autonomous agents, refusing the "citrus" because the solution appears too ordinary. Others place an AI product over disorganized files, inconsistent precedents, and fragmented workflows — they boil the citrus.

The question is not whether AI can produce a draft or a summary; it is whether the firm can use that capability to improve the complete delivery of legal service.

Governance Clears the Use; Workflow Design Determines the Value

AI governance determines whether a proposed tool is approved, appropriate for the task, protected, and subject to verification. This article focuses on what happens after the governance box is checked. For a small firm, the objective is not to become an "AI-first" shop; it is to become a better-run practice that uses AI deliberately.

The Complexity Trap: Refusing the Citrus

Small-firm leaders often lose months evaluating sophisticated systems before defining the problem they want to solve. Most do not need autonomous agents or private infrastructure. They need to solve recurring operational problems: time-sinks, stalls, rework, and inconsistent intake.

Before buying a product, ask:

What repeatedly consumes lawyer or staff time?
Where do matters stall?
Which steps do not require legal judgment?
Where is information repeatedly incomplete?
What result can be measured within 90 days?

The best first use case is the smallest recurring problem that matters enough to measure.

Boiling the Citrus: Adding AI Without Fixing the Workflow

Consider AI-assisted contract review. A system may produce a draft in minutes, but if the firm relies on outdated precedents, scattered instructions, and unclear file naming, the lawyer still spends hours reconciling the mess. One step is faster, but the matter is not moving faster.

The Workflow-First Approach:

Tool-first: Which platform should we buy, and how many tasks can we automate?
Workflow-first: What must happen from the moment work begins until the client receives a dependable result?

Redesigning the Workflow

Consider a recurring contract-review process:

Before: Information arrives via disparate emails and telephone calls. The lawyer hunts for the most recent precedent. Partner preferences emerge only at the final review stage. Substantial time is spent reconstructing instructions.
After: The matter begins with a structured intake form. The firm uses a controlled source bank (labeled by jurisdiction/version). AI compares the agreement against the baseline to prepare an issue table.
The Fail-Safe: The workflow must explicitly define a "Fail-Safe" point — a mandatory human check where the AI's logic is tested against the original instructions before the draft is finalized.

Standardize, Then Automate

Many apparent AI opportunities are actually standardization opportunities. Before introducing AI, the firm must:

Identify the current version of each core precedent.
Separate approved models from working drafts.
Distinguish reusable knowledge from client-specific work.

The Golden Test: Would the redesigned workflow still be better if the AI component were removed? If yes, you have improved the process. If no, you are simply boiling the citrus.

Measure the Complete Law Firm Workflow

AI demonstrations emphasize generation time. A practical scorecard must include:

Total turnaround time and total lawyer/staff effort.
Review and rework hours.
Errors or near misses.
Incentive Alignment: Does the firm's billing model actually support this? If you save five hours on a fixed-fee matter, you have increased your margin. If you do it on an hourly file, you have effectively cut your revenue unless you reallocate that capacity to other matters or higher-value work.

If there is no credible plan for how recovered capacity will be used, time saved is not financial value.

Preserve Lawyer Development

If AI performs the initial production, the workflow must identify where the "learning" will now occur. Assign tasks that require drafting from scratch and require associates to explain their reasoning for overriding AI-suggested clauses. The goal is to produce competent lawyers, not just efficient reviewers.

A Practical 90-Day Starting Plan

Days 1–30 (Select and Map): Map the process from instruction to output. Record the current baseline of time, rework, and errors.
Days 31–60 (Redesign and Test): Standardize intake and clean source materials. Define the Fail-Safe point.
Days 61–90 (Pilot and Decide): Run a controlled group. Record where the process fails. Scale only after the process is proven dependable.

The Voyage Ahead

The strongest AI strategy is not to automate more work. It is to preserve the conditions that make the work valuable and scale only what survives the test of rigorous workflow design. The challenge is not merely to bring the citrus aboard; it is to avoid boiling it.

For the companion decision framework governing whether a proposed AI use may proceed — including approval, task suitability, information protection, verification, and supervision — see AI Governance for Small Law Firms: The Approved, Appropriate, Protected and Proven Framework.

A Note on the History

The historical reasons for the Royal Navy's delayed adoption of citrus were more complex than this brief account suggests, and historians continue to debate how much credit Lind himself deserves [3]. The episode is used here as an implementation analogy rather than as a complete causal history.