February 12, 2024

Where AI Helps a Firm Today, and Where It Does Not

Partners are hearing bold claims about what AI can do for a firm. A useful answer is more modest and depends on the task, so it helps to sort tasks into three groups.

Tasks that suit assistance

Summarizing long documents, drafting a first version of a routine section, reorganizing notes, and finding relevant passages in your own library all fit. They are time-consuming, low in risk when checked, and easy to verify.

The test for a good candidate is whether a colleague can check the result in a couple of minutes. If checking takes as long as doing the task from scratch, the tool is not saving anything. Tasks with clear source material usually pass this test.

Tasks that need a person in charge

Deciding whether to bid, setting a fee, describing the firm's actual experience, and making claims about a client all need a person. A tool can prepare material for these decisions. It should not make them.

Write down the boundary between the groups and share it. Staff often want to use new tools and are unsure where the limits are. A one-page note that names the permitted tasks removes guesswork and reduces the risk of well-meant mistakes.

Tasks that need caution

Anything involving client confidential information, regulated advice or statements a buyer will rely on needs a firm policy before a tool is used. If you cannot say where the information goes and who can see it, do not put it in.

Ask the vendor where submitted text goes, who can see it, and whether it is kept. If the answers are unclear, use only material that is already public until they are settled. That protects your clients and costs you very little in practice.

Test with your own work

Choose three real tasks from last month and try an assisted approach on each, with a colleague comparing the output with the original. Time saved is a rough planning assumption until you have measured it across a fair number of tasks.

Keep the comparison fair. Include the time it took to write the instructions, to check the output and to correct it. Some tasks look faster only because the checking was left out. A modest saving that is real is better than a large one that is imagined.

Expect uneven quality

Tools can produce fluent writing that contains errors, and the errors are not always obvious. Build a review habit before you build a dependence. The cost of checking belongs in the estimate of time saved.

Train people on what errors look like: invented references, wrong dates, confident statements about things the tool had no way of knowing. Ask reviewers to look specifically for these. A short exercise with real examples does more than a policy memo.

Start small and write down what you learn

Pick one team and one or two tasks. After a month, record what worked, what needed rework and what surprised people. Expand only where the record supports it.

Share results with the wider firm in plain terms, including the disappointments. A frank summary builds trust among partners who are sceptical, and it prevents the enthusiasm of early adopters from setting expectations the tool cannot meet.

Keep partners informed without the jargon

Partners do not need to understand how a tool works, but they do need to understand what it does with their clients' material and how its output is checked. A short briefing in plain terms, with examples from your own work, is usually enough.

Encourage questions, and record the ones you cannot answer yet. Unanswered questions are a sensible reason to wait, and a firm that waits for clear answers is not falling behind. It is being careful with something it does not own. There is no prize for being first, and there is a real cost to explaining a preventable error to a client who trusted you with sensitive material.

VIPMarketing Editorial