top of page

Casey Flaherty on the Limits of Incremental Improvements and “Razzle Dazzle AI”

I noted with interest Casey Flaherty’s recent post entitled The Limits of Incremental Improvements. This sentence captures the gist of it: “[O]nce you cut something in half, there is nothing you can do, save eliminating it entirely, that will ever again deliver the same raw level of improvement.”

I’ve enjoyed my interactions with Casey. He has had occasion to turn his gaze in my direction; see his 2018 post How Much of Lawyering Is Being a Copy-and-Paste Monkey? and his 2016 post Bad Lawyering, Not Bad Forms. And last year I did a City of Contracts podcast with him; go here.

But I’ve seen enough of Casey to be a little intimidated. Me, I have my little corner of the marketplace of ideas, and I patrol it with austere single-mindedness. (Recently I idly pondered what historical figure I could relate to, and Symeon Stylites popped into my head. He’s the guy who lived on top of a pillar for 37 years.) By contrast, Casey has a more voracious, wide-ranging mind, and he has the patter to go with it.

So with mild trepidation on my part, let’s consider this from Casey’s recent post:

In the legal context, for example, we have good reason to accelerate contract review. Start with a standard review that averages 20 minutes and reduce it by 60% through basic interventions (harmonized templates, checklists, playbooks, deviation matrices, etc.). You save 12 minutes per contract. Next, throw on some razzle dazzle AI that reduces average review time by another 60%. You save less than 5 additional minutes. That’s not nothing, especially with large volumes. But it comes at a cost, including the opportunity cost of addressing other chokepoints, constraints, and rate-limiting factors.

Contract review is obviously something I’m involved with. It’s a big part of LegalSifter’s business, and in my capacity as an advisor to LegalSifter I’ve devoted a chunk of the past two years working on LegalSifter’s technology—deciding what concepts to look for, figuring out how those concepts are expressed in contracts, and writing help text.

With that in mind, let’s consider Casey’s phrase “razzle dazzle AI.” I don’t know about anyone else’s artificial-intelligence technology, but “razzle dazzle” sure isn’t a phrase that comes to mind when I think of LegalSifter’s AI. Instead, I think of toil. Toil by me, and toil by the data-science team responsible for training the AI technology to look for the issues we target.

But a more fundamental point came to mind on reading that part of Casey’s post.

Like the two other examples that anchor his post, his contract-review scenario involves something that’s easy to measure—time. (The two other examples involve fuel consumption and, again, time.) LegalSifter is certainly keen to tell the world that contract review using its service saves time, but that’s just one benefit of what LegalSifter offers—expertise, delivered through its technology. The other benefit is more important. It’s improved quality.

That’s because—to return to themes discussed in Casey’s 2018 and 2016 posts—traditional contracts are dysfunctional, in terms of what they say and how they say it. Pretty much all of us are riding the copy-and-paste train, holding on for dear life as it barrels down the track, smoke billowing, sparks flying, machinery rattling. Few have the time, the expertise, or the authority to do the challenging and disruptive work of improving quality. Instead, we rely on dog-eared conventional wisdom and indulge in legalistic fantasy. I see that not only in templates but also in checklists and playbooks of the sort Casey mentions.

The dysfunction is so entrenched that the only way to make real progress in sweeping it aside is through technology. One front in that battle is automated contract creation; the other is automated contract review. It’s early days yet, but LegalSifter’s marriage of expertise and technology holds the promise of benefits from automated contract review that far exceed the meager incremental improvements Casey has in mind, although they're harder to measure. We’ll do our darnedest to make sure we deliver.

(Oh, the photo at the top of this post? It's from a performance of the musical Chicago, which features the song Razzle Dazzle. Go here for a clip of the song being performed in the movie version.)

Recent Posts

See All

Recently I noticed this article on Artificial Lawyer. The title is Generative Legal AI + “The Last Human Mile”, and it’s about limits to applying AI to legal work. It says this: The last mile problem

bottom of page