The Legal Innovation Scoring Tool is Live!

Based on Professor William Henderson’s Diffusion of Innovation Worksheet [(c) Legal Evolution PBC, 2019-2025], users input attributes of a proposed innovation, the nature of the system into which it will be deployed, and the efforts possible via change agents to predict chances of successful uptake.
The tool calculates a total score ranging from -27 to 40, with scores below 0 unlikely to succeed, and generates a report with recommendations. 

Background

Bill Henderson taught my favorite class at Northwestern Law in the Master of Science in Law (MSL) program: Innovation Diffusion in the Legal Industry. He was a visiting professor at Northwestern while also serving as Professor and Stephen F. Burns Chair on the Legal Profession at the Indiana University Maurer School of Law. As a successful legaltech co-founder (Lawyer Metrics from 2010 to 2016), extensive publisher on his site legalevolution.org, and co-founder of the nonprofit Institute for the Future of Law Practice (IFLP, bought by the Law School Admission Council), Bill was uniquely suited to train the skills and methods needed to usher legal professionals into a more efficient tech-enabled future.

Bill’s diffusion worksheet is underpinned by Everett Rogers’ research, notably collected in his Diffusion of Innovations book, along with his own research and practical experience. It forces users to analyze a proposed innovation (new software platform, tool, work process, etc.) in five separate categories:

• Perceived Attributes of Innovation (Score -15 to +15)
  • Is it markedly better? Easy to try? Easy to see results?
• Type of Innovation Decision (Score -6 to +6)
  • How/where is the decision to use made? Individually like workers sneaking prompts into ChatGPT? Or does the CEO need to approve? Somewhere between?
• Communication Channels (Score 0 to +6)
  • Will others hear about successes/failures? How are the innovation and its advantages publicized?
• Nature of Social System (Score -6 to +6)
  • Would the organization’s culture be described as “very traditional” (law firms) or “very modern” (tech startups)? Are groups inside the organization connected or isolated?
• Efforts of Change Agents (Score 0 to +7)
  • Who is in the field advocating for change and what are their attributes? Seven binary choices:
    • Personal contact: Frequent or rare?
    • Orientation/intent: Problem solving or making a sale?
    • Empathy: Understanding or ambivalent?
    • Homophily: Similar to audience or an outsider?
    • Credibility: Trustworthy and knowledgeable or ill-informed?
    • Working thru opinion leaders: Access to trusted/powerful insiders or solo?
    • Technical education: Teaching independence or lingering presence?

Users total numbers in the subsections, then tally all scores together for a single number score. With that simple math, they get two immediate insights: The innovation’s strength/weakness in each of the five categories, and the overall likelihood of success. One fascinating result I often see when walking people through the scoring is how much their initial predictions differ from their final score.
Fun fact: Bill was a firefighter-paramedic before his law career, and the Apgar infant assessment scorecard (published 1953) works the same way.

Over the last 6+ years I’ve used this assessment method for guiding legaltech marketing efforts, scoping pilot projects, setting platform scaling plans, and even adjusting product feature roadmaps (to name a few). But despite how useful this method is, or how often I’ve shared the PDF or site link, it’s still relatively “flat”: you need an actual paper copy to write on, or somehow track your section scores while you go through the assessment. Considering how often individuals need to assess innovations, it’s not really a problem (or worth the developer time to “appify” it).

But what if I could turn it into an app quickly, cheaply, and by myself? I’d improve the adoption likelihood of the worksheet by increasing the Perceived Attributes of Innovation score in four out of five subsections:

• Compatibility
  • Anyone with a computer/tablet/phone can use it
• Complexity
  • Calculator no longer needed, easily adjust and update scores
• Trialability
  • One click to open the tool, with embedded links to the explanatory sections of Bill’s post
• Observability
  • Immediately see assessed scores, with ability to export scoresets and narrative reports.

Enter GPT-5

After OpenAI launched GPT-5 in August 2025, I ran a test to see how good it really was by asking several simultaneous questions. Could it:

• Code a functional app that didn’t require rework?
• Locate a specific website page and reference the material therein?
• Read over and understand a scoring tool (PDF and JPG available)?
• Do all this on the first shot?

So with no prep or additional context, I fed it this prompt:

Success

I specifically didn’t reference Bill’s website (legalevolution.org) to test GPT-5’s seek/find capabilities, and it passed with flying colors:

It even added the ability to download inputted settings as a JSON (a widely-used file type used to share information in programming backends), which I didn’t initially ask for.
In 997 lines of code generated in less than 15 seconds, I had exactly what I asked for. Prototyping used to be much harder than this. Here’s the initial functional result.

The great part about building with HTML (the basic language used to render websites) is that testing, deploying, and sharing is super easy. Anyone with a browser can open the file, as opposed to many other languages (e.g. Python) that require local machine setup, downloading of specific libraries, and other environment-specific steps. Yes, HTML can only run relatively simple tools compared to Python, but it’s hard to beat for rapid testing.

Polishing

To be fair, GPT-5’s output waned as I asked for specific changes and improvements, and when it broke the code completely I switched to the AI-assisted coding app Cursor.
NB: Replit, Windsurf, and other AI coding apps work great too, I just like the skill/capability ratio of Cursor (a bit more work to spin up but very powerful capabilities).

With Cursor I:

• Added colored pills for each section to resemble the worksheet
• Created a sliding drawer feature below the header to show/hide section totals
• Set grading scales for all scores e.g. “Poor – Some barriers present”, or “Excellent – Strong innovation design”
• Embedded links for each section to the corresponding explainer in the original post
• Added a feature that produces a sharable analysis report with
  • Name field for innovation
  • Text explainers for every metric
  • Dynamic strategic recommendation bullet list based on section scores
  • Copy-to-clipboard feature (drop into an email!)
  • Download report as HTML (I tried DOCX and PDF but way more work)

Generated on-screen report (above), slide-out header score drawer (below)

I thought about using LLMs to actually write a narrative report, but as I often say “you’d be surprised how far you can get with a decision tree”. So I worked up a fairly simple logic for generating text and ran with it. One example of simplicity ruling the day: Setting the strategic recommendation generator to limit the number of recommendations if the overall score was above a certain threshold. The better primed you are for success, the less advice you probably need…

Overall a fun, fast project to turn useful applied research into a prototype application.
Many thanks of course to Professor Henderson and his cutting-edge work!

And in case you missed it, open the tool here.