Archive for July, 2016

One source of frustration with a time-based charging structure (“billable hours”) is that it is difficult to accurately estimate how long a piece of work will take. This post looks at ways we can address this. (Or at least puts down my thoughts on virtual paper .)

Many professional services are priced based on an hourly (or day) rate. This is true of private practice patent attorneys. Although there are critics, the persistence of the billable hour suggests it may be one of the least worst systems available.

Most day-to-day patent work in private practice consists of relatively small items of work. Here “small” means around £1k to £10k, as compared to the £1m cases or transactions of large law firms. These small items of work typically stretch over a few weeks or months.

When performing patent work an unforeseen issue or a overly long publication can easily derail a cost estimate. For example, it is relatively easy to find that a few more hours are needed after looking into an examiner objection or piece of prior art in more detail. This often presents a lose-lose situation for both attorney and client – the work needs to be done, so either the attorney has to cap their charges in line with an estimate or the client needs to pay above an estimate to complete the job. This is not just an issue for patent attorneys – try comparing any quote from a builder or plumber with the actual cost of the work.

This got me thinking about taxis. They have been around for a while, and recent services like Uber offer you a price on your phone that you then accept. This is a nice system for both customer and driver – the customer gets a set price and likewise the driver gets a fare proportional to her time. Could something like that work for patent work?

For taxi services, the underlying variable is miles (or kilometres depending on your Brexit stance). A cost is calculated by adding a mile-based rate to a basic charge, with minimum and cancellation charges.

For patent work, one underlying variable is words. Take an examination report (or “office action”). The amount of time it takes to respond to novelty and inventive step objections is typically proportional to the length of the patent specification in question, the number of claims and the number of prior art citations.

Now, we can use EPO OPS to retrieve the full text of a patent application, including  description and claims. We can also retrieve details of citations and their relevance to the patent application (e.g. category ‘X’, ‘Y’ or ‘A’). I am working on parsing PDF documents such as examination reports to extract the text therein. In any case, this information can be quickly entered from a 5 minute parse of an examination report.

Wikipedia also tells me that an average reading rate for learning or comprehension is around 150-200 words per minute.

This suggests that we can automate a time estimate based on:

  • Words in the description of a published patent application – WPA (based on a need to read the patent application);
  • Number of claims – NCPA (applications with 100s of claims take a lot longer to work on);
  • Words in the claims – WCPA (claims with more words are likely to take more time to understand);
  • For each relevant citations (category ‘X’ or ‘Y’ – this could be a long sparse vector) :
    • Words in the description of the citation – WCITx (as you need to read these to deal with novelty or inventive step objections); and
  • A base time estimate multiplied by the number of objections raised – BTo (possibly weighted by type):
    • E.g. x amount of time per clarity objection, y amount of time per novelty objection.

Even better we need not work out the relationship ourselves. We can create a numeric feature vector with the above information and let a machine learning system figure it out. This would work based on a database of stored invoicing data (where an actual time spent or time billed amount may be extracted to associate with the feature vector).

The result would be an automated system for pricing an examination report response based on publically available data. We could host this on Heroku. By doing this we have just created a marketplace for patent responses – a single weight could be used by patent firms to set their pricing.

Similar pricing models could also be applied to patent drafting. The cost of a draft may be estimated based on a set length, number of drawings, number of claims, number of independent claims, length of invention disclosure and length of known prior art. The variables for a response are similar for a European opposition or an appeal, just with different weights and expanded feature vectors to cover multiple parties.

This would yield a compromise between billable hours and fixed fees. For example, a variable yet fair fixed fee estimate may be generated automatically before the work is performed. The client gets predictability and consistency in pricing. The attorney gets paid in proportion to her efforts.