Most in-house teams say AI is already shrinking the work they send to law firms. Here is how to update outside counsel guidelines so the savings show up.
In-house legal teams say AI is changing what they buy from law firms, but most outside counsel guidelines were written before anyone asked how a firm uses AI or who keeps the time it saves. A new survey puts numbers on the shift, and it makes the guidelines the obvious place to start.
Everlaw’s In-House Innovation Report, produced with Wakefield Research and announced October 20, surveyed 300 U.S. in-house legal professionals between June 26 and July 30, 2026. This playbook uses those findings to suggest what a legal operations team should add to its guidelines, and what it should check before claiming any savings.
What the Survey Found
Among respondents whose departments use outside counsel, 54% say AI has already reduced the volume of work they send to law firms, and another 44% expect it to within a year. Only 2% have done neither and do not expect to. At departments that use generative AI extensively, 73% are already sending less work out.
The savings are reaching the invoices too, by the respondents’ own account. Of those working with law firms that use generative AI, 92% report AI-related savings from those firms, though most of that is modest: 13% call the savings significant and 56% moderate, and Everlaw’s own write-up adds that 23% describe them as slight. Only 4% say their outside firms are not using generative AI at all.
The expectations are the useful part for guideline drafting. Among respondents who engage law firms, 62% expect clear disclosure of AI use, 61% expect measurable gains in efficiency or outcomes, and 60% expect lower fees or fewer billed hours when AI is used. On pricing, 81% expect outcome-based pricing to overtake the billable hour within five years, and 57% expect flat or fixed fees to be the most widely adopted alternative for AI-assisted work.
“AI rebalances the relationship between in-house legal teams and their law firms,” said Gloria Lee, Chief Legal Officer at Everlaw, in the release.
Keep the source in mind. Everlaw sells litigation software to the same legal departments it surveyed, the sample is 300 people, and the release gives a margin of error of plus or minus 5.7 percentage points. The direction is believable. The precise percentages deserve less weight.
What to Put in Your Guidelines
Start with disclosure, because it is the cheapest change and the one most respondents already expect. Ask firms to state, at engagement, which generative AI tools they will use on your matters and whether those tools keep or train on your data. That is a confidentiality question as much as a billing one, and it gives you a record to point to later.
Next, set the billing rule for AI-assisted time in writing. The ABA’s Formal Opinion 512 already gives a baseline: if a lawyer spends 15 minutes using a tool to draft a pleading, the lawyer may charge for that time plus the time needed to review the draft, and in most circumstances may not charge the client for learning how to use the tool. Writing that into your guidelines turns an ethics principle into a term the firm agreed to.
Then decide which work stays outside. Respondents said they mainly want to pay firms for business and commercial insight (56%) and for managing complex matters and specialized expertise (54% each), while fewer than 30% picked document review or routine work. In litigation, the tasks respondents most often expect to bring in-house are early case assessment and matter scoping (58%) and outside counsel performance and billing review (54%). If your department plans to do that work itself, say so in the guidelines so firms do not staff it by default.
Finally, ask for verification. One third of respondents said they had encountered AI-hallucinated evidence or case law in litigation, including 12% from their own teams. A guideline that requires firms to confirm every citation before filing, and to tell you if an AI error reaches a court, costs nothing to add.
How to Check the Savings
Guidelines only matter if someone reads the invoices against them. The data to do that is starting to exist. Legora and Laurel announced on October 13 that Laurel will turn tasks lawyers complete in Legora into draft time entries, each tied to a client and matter with a description for the lawyer to review. The companies say a later phase will develop measures of how lawyers use Legora and what that work is worth to a firm. If your firms use tools like these, ask whether AI-assisted entries can be flagged on your invoices.
Be realistic about the pace. In-house expectations run well ahead of how most firms bill today. The 8am billing report covered by LawSites on October 22 found that 47.7% of the small and midsize firms in its data billed entirely by the hour in the first half of 2026, and flat fees made up 15% of invoices, down from 16%. That data comes from a different market than corporate panels, but it is a reminder that firms do not change pricing just because clients predict they will.
Common Questions
Can a legal department require firms to disclose AI use?
Yes, as a contract term. Outside counsel guidelines are commercial terms the firm accepts, so a department can ask for disclosure of tools and how they are used, subject to whatever the firm agrees to.
Should we ban firms from using AI on our matters?
The survey suggests few departments want that, since most expect savings from firms’ AI use. A narrower approach is to require approved tools, data protection terms and citation checks.
How do we know if AI savings are real?
Compare hours and staffing on similar matters before and after a firm adopts AI tools, and ask firms to identify AI-assisted entries. Self-reported savings in a survey are not the same as a lower invoice.
What to Watch
Watch whether firms start offering fixed fees for work that AI makes predictable, and whether time capture tools give clients a view of AI-assisted hours. The departments that write these terms now will have a baseline to negotiate from when outcome-based pricing proposals arrive.
