AI Is Changing What Patent Services Cost, Not Just How Fast They Are Delivered

Most of the public conversation about AI in professional services has settled into a frame that is comfortable but limited: AI helps experts do individual tasks faster. Draft a clause quicker. Summarise a document quicker. Find a reference quicker. Task-level AI is real, and it is useful, and it is nowhere near the interesting part of the story. 

What is also changing in patent law (and, shortly, in several adjacent professions) is that AI has moved out of the toolbox. It is starting to define the workflow itself. Every stage of the patent lifecycle, from the first conversation with an inventor through to opposition and litigation, looks different now from how it looked a short time ago. And when a technology reshapes every stage of a long process, the economics of the work change in ways a faster task never does. 

From tool to process 

Not long ago, AI in patent practice meant a handful of point tools. A prior art search assistant in a browser tab. A claim checker in Word. Wikipedia for information gathering and fact checking. Each tool lived on its own, and it was the attorney who still did the underlying work by hand, saving a few minutes here and there. Useful enough. Each did not massively change how the day was organised. 

What has emerged since is that the AI now sits inside the workflow, structuring the process itself rather than assisting with pieces of it. An invention disclosure is turned into a structured draft disclosure by means of AI before the attorney has opened the file. A patent application is written against the disclosure, with the drafter reviewing and directing the AI output rather than producing from a blank page. An office action comes in and is parsed, categorised and mapped against the specification and file history automatically, directing attention to the rejections that actually require thought. Opposition strategy is sketched against the prior art before anyone sits down to seriously consider it.  

None of this replaces the attorney; it changes what the attorney does. A working day used to be full of production: writing, mapping, cross-referencing, formatting. Now much more of it is judgement. What matters in this draft. What is weak in that argument. Whether the examiner’s position is worth fighting or better narrowed by amendment. That was always the part of the work that mattered most, and it is now most of what the attorney will be doing. 

Where AI is saving time 

The efficiency gains are real, and they are spread unevenly across the patent lifecycle. It is worth being specific about where the time actually goes. 

Invention capture. Ask any patent attorney about the gap between what inventors explain in a meeting and what ends up in the disclosure form that lands on a desk a week later, and you will get a rueful answer. Something is always lost in translation. The inventor is a brilliant engineer; the disclosure form is a blunt instrument; the attorney is catching up on a Tuesday afternoon from a form that does not quite match what was actually invented. AI tools that sit in, or replace, the disclosure meeting itself fix a problem the profession has lived with for decades. The saving is not just the hours that used to go on piecing the disclosure together afterwards. It is the rework avoided downstream, when the attorney discovers two weeks into drafting that nobody asked the inventor how the system behaves under certain important conditions. 

Drafting. This is where the public debate has concentrated. A generalist chatbot will produce a first draft that is not fit to file unless very carefully directed, and anyone who tells you otherwise has either not tried it or has not read the result carefully. But specialist systems, built to follow good patent practice and ideal patent specification structure, are a different matter. The first draft is genuinely usable. The attorney’s job shifts from writing to architecting: deciding claim scope, shaping the disclosure, identifying what the drafting assistant has missed or misjudged. 

Filing and formalities. Less glamorous, and probably the largest aggregate saving. Legal documents accompanying the application, such as assignments and inventor declarations, Power of Attorney, sequence listings, application forms, formality letters, fee calculations. The sheer volume of this work scales badly with portfolio size, and it is exactly the kind of work AI handles well. 

Prosecution. Replying to a patent office objection is where the efficiency gain is most visible, because the task is bounded and the output is discrete. A recent response in a moderately complex case ran to around two hours from start to finish. The same case, handled the old way, would have taken most of a working day. The difference was not that the AI wrote the arguments. It was that they came together earlier, because the mechanical work of mapping claims against references and parsing the examiner’s reasoning was already done. 

Opposition and litigation. Once granted, patents are challenged and asserted. The gains here are earlier-stage and more variable. Prior art marshalling in opposition, claim construction analysis, document review prioritisation in litigation, infringement mapping against asserted claims. All of this is work that AI now accelerates materially. The strategic calls still rest with the attorney. Preparation time before those calls gets compressed instead. 

Looking at the stages together, the striking thing is that the saving is not concentrated anywhere in particular. It is distributed across the whole process. That is why there is a change in how the work is delivered, rather than just a change in how individual tasks are performed. 

Why this changes the cost, not just the speed 

If AI only made individual tasks faster, the economics would be straightforward. Firms would do the same work a little more cheaply, keep the margin, and clients would see a slight benefit. That is broadly what happened with earlier waves of legal technology: docketing systems, electronic filing, document management. Useful. Not disruptive. 

This is different. When AI restructures the whole workflow, the cost of delivering a patent application, a response to patent office objections, or an opposition case drops by a multiple rather than a percentage. And the drop does not come from working faster. It comes from the fact that one attorney with good AI-driven tooling can now deliver what previously needed two or three people. The work itself is being made differently. 

When how the work gets made changes, how it is priced has to change too. The billable hour made sense when the hour was a reasonable proxy for the production work that went into it, but it is not that any more. A client paying for hours on an AI-assisted matter is paying for a metric that has come loose from what they are actually receiving. Sophisticated in-house teams (the ones running serious patent portfolios) have noticed, and the procurement conversations are getting pointed. 

The profession’s response has been mixed. Some firms are moving to value and outcome-linked fee deals. Most are quietly holding the hourly line, and a few have gone further. A recent Financial Times piece quoted partners at several firms, one of them a patent attorney, suggesting that reviewing AI-generated correspondence from clients takes so long that fees should go up rather than down. That is not a position that survives serious examination. Clients can see the maths. 

Early evidence 

It is still early for firm-wide data, and anyone quoting precise percentages on overall savings should be treated carefully. The directional evidence, though, is consistent across firms that have genuinely adopted workflow AI rather than bolting point tools onto old processes. 

Per-matter cost drops enough that fixed pricing becomes attractive to the firm as well as the client. Both sides share the gain rather than one side capturing it. Inside the firm, the mix of work shifts visibly: less production, more strategy, more of the work that senior attorneys actually wanted to do when they became senior attorneys. And client behaviour changes: work consolidates with firms that price predictably, in-house budgets that used to be spent policing hourly bills get redirected, and the scope of matters a mid-sized in-house team can run externally expands, because the unit cost has dropped enough to make previously uneconomic work economic. 

Any one of those effects is easy to dismiss on its own. Together they describe a commercial relationship that is measurably different from the one that prevailed at the start of the decade. 

What this means beyond patent law 

Patent prosecution is a useful early indicator because the work is technical, the output is concrete, and the pricing is visible. But the pattern generalises. Any profession that bills expert time for artisanal output is on the same curve, at different points. Consulting. Compliance. Audit. Much of corporate law. Parts of medicine and finance. 

The sequence is recognisable. First, point tools that help with tasks. Then, workflow AI that reorganises the process. Then an awkward period when the cost of delivery has dropped but pricing has not yet caught up, and the firms that move first get to decide what the new model looks like. Then the model settles, and the firms that waited find out it was set without them. 

Patent practice is somewhere in the middle of that. Other professions are earlier on. The backdrop matters too: The Economist reported in March that profits per lawyer at the hundred largest firms have risen 54% since 2019, with hourly rates climbing at more than twice the pace of inflation. That is the cost base against which AI is now reshaping the work. Privately, many senior people in professional services know the billable hour is living on borrowed time in any discipline where AI can do the heavy lifting. The question is whether firms choose how the pricing changes, or wait to find out.