As I speak and publish on the subject of Artificial Intelligence in the legal industry, the question I am asked most often is “How do I start? What is the best way to begin my (or my firm’s) participation in this change?”
For attorneys working in law firms, the best way to begin is to is to develop an algorithm to predict which of your clients are most likely to need which of your services in the near future. (Some may quibble that this is not full blown AI, replete with neural networks, machine learning and natural language processing, but it’s a fine way to start.) The legal services potentially needed can include just about anything from a transaction (perhaps an opportunity created by regulatory changes) to a dispute (e.g., filing of a shareholder class action).
The exercise described below will make your marketing more efficient in that you will be pitching your services to those most likely to be in imminent need. It will benefit your clients in that you will not be wasting their time, and you’re likely to alert them to an upcoming risk or opportunity of which they would not otherwise be aware.
In the example of a possible shareholder class action, here are the steps I suggest you take:
- Identify your client’s risk or opportunity you’d like to predict.
- List all of the firm’s large corporate clients. This is statistics, so the more the better; perhaps your top 200 clients, or even 500. It’s fine to include prospects too.
- Brainstorm all of the characteristics and events that could possibly be predictors of the event. For instance, with filing of a shareholder class action against one of your large corporate clients you might include:
- Has there been recent turnover in the client’s Board?
- Has there been a recent restatement of earnings?
- How has their stock price behaved for the past year?
- Have there been an unusual number of large stock transactions in the past year?
- Have Board members bought or sold substantial blocks of stock in the past year?
- Has there been recent negative publicity about the company?
- Has there been recent negative publicity about any of the Board members?
- Has the company had substantial recent litigation of any kind?
- Has there been any previous securities class action?
- Has the company recently been involved in any substantial deals, especially failed deals/takeovers?
- Have they made other substantial investments (e.g., research projects)?
- Have other companies in the client’s industry recently faced securities class actions?
- Does the company plan to do anything that might cause an action (e.g., does their strategic plan or website indicate intent to expand into unstable countries or markets)?
- Have individuals at this client been opening and reading your client alerts regarding class actions.
- Have individuals at this client visited relevant pages on you firm’s website?
Be inclusive, adding as many possible predictors as you can come up with. Don’t try to screen them for “reasonableness.” It’s fine to include a hundred!
2. Collect data for all of the possible predictors. These data could come from internal sources (e.g., CRM, your billing system, readership of client alerts), subscription databases (e.g., Thomson Reuters, Lexis Nexus), news feeds (e.g., Manzama, Google alerts, Mondaq), company publications (10Ks, Strategic Plan, company website). Wherever possible, attach dates to the data elements (e.g., “earnings restatements occurred on 5/14/2006 and 2/1/2009,” “individuals from the company visited the Class Action pages of our website on 10/6/2005 and 2/10/1017”). Make sure to include dates of any prior occurrences of the event of interest (i.e., filing of a shareholder class action against the company).
3. Combine all of the data into one database.
4. Create your model by conducting regression analysis of the entire dataset with prior occurrences of the event of interest as your “dependent variable” and prior instances of the bulleted list above as your “independent variables.” (In AI terms, this is your “training database”). This will result in a model (equation) for predicting the event of interest. It will eliminate most of the variables as not useful in prediction and will show how to combine the remaining variables into a model. (More could be done with AI analytic tools such as Neural Networks, but in this case simple regression analysis is probably enough.)
5. Continue to collect the data included in your model and on a regular basis, perhaps monthly, run the analysis with updated data to predict which of your clients are likely to be faced with the event (in this example, the filing of a shareholder class action against them).
Make sense? Feel free to give me a call to talk about the details.