Claude for legal and professional services.

Legal and professional services firms sit in an unusual position with any AI tool. The work, drafting, review, research, analysis of documents, is exactly the kind of language heavy task Claude does well, so the upside is large. But the material these firms handle is privileged, confidential, and often subject to client specific obligations, which means the data terms are not a footnote to the decision. They are the decision. A firm cannot adopt a tool that puts client confidentiality at risk no matter how capable it is, and it cannot rely on a verbal assurance when professional duties and client contracts are at stake. So for these firms the evaluation runs on two tracks at once: does the model do the work well, and do the terms protect the client material it will touch. This piece sets out what legal and professional services buyers should expect from Claude, what to pin down in writing, and how the way these firms work shapes the commercial deal.

Why confidentiality is the gating question

For most enterprises data protection is important. For a law firm or an accountancy it is foundational, because the obligations are not only regulatory but professional and contractual. Privilege can be at stake. Client engagement letters frequently impose specific confidentiality and data handling requirements that the firm must be able to honor in every tool it uses. A breach is not just a compliance failure, it is a professional one that can damage client relationships and expose the firm. So before the capability question is even worth asking, the firm needs confidence that client material processed through Claude is handled in a way that satisfies these duties. This is why the data terms gate the decision rather than informing it at the margin, and why they belong at the front of the evaluation.

What to confirm in writing

The terms that matter to a professional services firm need to be confirmed in the agreement, not inferred from a marketing page or a sales conversation. The specifics to pin down include the following.

• Training use of your data. Confirm in writing whether content submitted through your enterprise or API usage is used to train models. For a firm handling privileged material, a clear contractual commitment that client content is not used for training is usually essential.
• Retention and deletion. How long inputs and outputs are retained, on what basis, and your ability to control or shorten that retention to match client obligations and your own records policy.
• Access and confidentiality. Who can access the data, under what controls, and the confidentiality commitments that apply, so you can map them against the duties you owe your clients.
• Data location. Where data is processed and stored, which matters when client engagements or regulation impose location requirements on where confidential material may reside.
• Audit and evidence. Your ability to obtain audit logs and records that let you demonstrate, to a client or a regulator, that the controls are in place and being followed.

The usage profile of a professional services firm

How these firms work also shapes the commercial side of the deal, because their usage profile has a distinctive shape. The work tends to be document heavy, with large inputs, contracts, filings, financial records, research material, processed for drafting, review, and analysis. A great deal of it is not real time in the sense that a client is watching a screen waiting for output. It feeds a professional's work product, which is reviewed and refined before it goes anywhere. And much of it carries large stable context, the same precedents, standards, and reference material applied across many matters. That profile, large documents, reviewable rather than instant, with substantial repeated context, happens to be an excellent fit for the optimization levers that reduce cost, which is something a firm can use to its advantage in the deal.

How the usage profile lowers the cost

Because professional services work fits the optimization levers so well, a firm that understands its own usage can negotiate from a much lower effective cost than the list rate implies. The large stable context, the standards, precedents, and reference material reused across matters, is exactly what prompt caching discounts, returning up to ninety percent on that repeated content. The reviewable nature of most of the work means a great deal of it can run on the Batch API at half the rate rather than paying the real time premium, since a professional reviews the output before it is used anyway. And not every task needs the most capable model, so routing routine review and extraction to a lighter model captures further saving. A firm that applies these levers brings its real run rate down before it ever discusses a commitment, which changes the number it can credibly take to the table.