Be useful

I spent last week looking at some large electronic bills of costs.

After a hesitant start, I am a great fan of the electronic bill of costs. I know a lot of lawyers who tolerate it, or even loathe it. They prefer the familiar three column bill of costs of yesteryear, which is easier to read, digest and understand. By way of contrast, the electronic bill of costs in spreadsheet format, can appear as an amorphous, indigestible mass of detail, running to thousands of entries, with page after page of closely written detail. But to look at an electronic bill of costs in this way is to miss a trick.

It is not meant to be read, to be understood. Even less useful, is any attempt to print it out, and comprehend the claim for costs that it makes, with the naked eye. Instead an electronic bill of costs is meant to be understood digitally. Through the magic of the filtering function in Excel, you can find virtually any single item of costs, or any conceivable aggregation of costs, consider them and evaluate them effortlessly. You can amend the bill, removing or reducing items, and simultaneosly, the bill will recalculate itself, enabling you to see the overal effect of the alterations on the claim for costs. It is like magic, but in order to cast your spells, you have to learn to speak Excel.

What strikes me however, about all these electronic bills of costs, is that although the claim for costs is presented in digital format, the underlying claim has to be calculated the hard way in analogue fashion, with costs draftspeople and costs lawyers, reckoning up the time and value of a solicitors file in a way that has not fundamentally changed in 50 years, recording the entries by typing them out and categorising them, in each column of the bill of costs. At least this is what I assume they do, as otherwise I cannot understand why claims for the costs of drawing the bill, regularly top 100 hours. But is it meant to be this way?

Returning to the Jackson Report of 2009, the potential for an improved and electronic bill of costs was noted and to be judged in these terms:

5.3 The three requirements. There are three requirements which have to be satisfied:
(i) The bill must provide more transparent explanation than is currently provided, about what work was done in the various time periods and why.
(ii) The bill must provide a user-friendly synopsis of the work done, how long it took and why. This is in contrast to bills in the present format, which are turgid to read and present no clear overall picture.
(iii) The bill must be inexpensive to prepare. This is in contrast to the present bills, which typically cost many thousands of pounds to assemble.

Ho ho ho. My view is that electronic bills are more expensive to prepare than their predecessors. But how was this cheap ‘n’ cheerful new bill of costs to be prepared? The Final Report continues:

5.4 How to meet those requirements. In my view, modern technology provides the solution. Time recording systems must capture relevant information as work proceeds. The bill format must be compatible with existing time recording systems, so that at any given point in a piece of litigation a bill of costs can be generated automatically. Such a bill of costs must contain the necessary explanatory material, which is currently lacking from the bills prepared for detailed assessment. Crucially, the costs software must be capable of presenting the bill at different levels of generality. This will enable the solicitor to provide either (a) a user-friendly synopsis or (b) a detailed bill with all the information and explanation needed for a detailed assessment or (c) an intermediate document somewhere between (a) and (b). The software must provide for work which is not chargeable or work which is written off
to be allocated to a separate file.

5.5 Armed with such a software system, solicitors should be able to produce upto-date costs information for the client or schedules of costs for summary assessment at whatever level of generality may be required. Also, at the end of the case, the solicitors will be able to produce a detailed bill of costs, which can be used either for negotiating costs with the other side or for a detailed assessment hearing.

5.6 I therefore recommend that work should be put in hand to develop existing software systems, so that they can (a) capture relevant information as work proceeds and (b) automatically generate bills of costs at whatever level of generality may be required. Two of my assessors, Senior Costs Judge Peter Hurst and Jeremy Morgan QC, have discussed the possibility of such software being developed with a firm of law costs consultants. The current proposals are that a bill should be presented in the order “phase, task, activity”. According to these proposals, the bill is divided into five “phases”: (1) case assessment, advice and administration; (2) pleadings and interim applications; (3) disclosure; (4) trial preparation and trial; and (5) detailed assessment. Each phase is then broken down to identify different tasks. A summary sheet lists the profit costs and disbursements in respect of each task in each phase. In the body of the bill itself, each task in each phase is set out in chronological order, with an indication of the time spent and the amount claimed. A bill in this form could easily be transmitted in electronic form, provided that all those involved had compatible IT software. If bills were to be prepared along the lines suggested, and dealt with electronically, there would potentially be large savings in time and costs. One advantage of the proposed system is that costs information can be extracted at different levels of generality. The electronic formatting of bills should, in principle, provide greater transparency.

Now Practice Direction 47 is very clear about the contents of electronic bills, but it imposes no requirements on the parties’ solicitors to record their time in a particular way, or on particular software, using the dreaded “J codes” or any other system of time recording. The result is predictable: the electronic bill is not capable of being created in the space of a few clicks of a mouse, within software: instead it is the metaphorical equivalent of a twenty first century supercar: laboriously built by hand.

One of the puzzles of the last decade is why paying parties do not fundamentally challeng the time for compiling bills of costs by asserting that failing to record time in a way contemplated by the Jackson report is prima facie unreasonable, leading to unreasonably incurred costs and the only sums which should be allowed for drafting and checking the bill, should be based on the assumption that time will be recorded digitally for the purposes of bill compilation in a compatible format, which will produce fairly nominal figures compared with the sort of sums which are routinely claimed and awarded in detailed assessments.

But whether this nettle is grasped, is not perhaps the most presssing question. Would it be possible, even if time is not recorded in “J codes” or similar, for artificial intelligence to scan an electronic file and populate a bill of costs? Let us ask a biased bystander: ChatGPT.

Assume I am an English costs lawyer. I want to create an electronic bill of costs using an Excel spreadsheet in Precedent S. The solicitors file I have to work with, is an electronic collection of PDF documents. Is it possible for artificial intelligence to scan and read the various PDFs, note the dates, time and content of those documents, and populate the columns of the Precedent S, with accurate data? If so, how long would it take to create the electronic bill of costs, assuming that the solicitors file was 10,000 PDF pages long?

The answer it provides is as follows:

Yes, it is possible for artificial intelligence (AI) to scan and read various PDFs, extract relevant data such as dates, time entries, and content, and populate the columns of the Precedent S spreadsheet accurately. This process involves several steps and technologies, including Optical Character Recognition (OCR), Natural Language Processing (NLP), and data extraction and transformation techniques.

Here’s a high-level overview of how it could work:

  1. OCR (Optical Character Recognition): The AI would use OCR to convert the text in the PDF documents into machine-readable text. Tools like Adobe Acrobat, Tesseract, or commercial services like Google Cloud Vision or AWS Textract can be used for this purpose.

  2. NLP (Natural Language Processing): Once the text is extracted, NLP techniques can be used to identify and categorize the relevant data, such as dates, time entries, descriptions of work done, and other necessary details.

  3. Data Extraction and Transformation: The extracted data would then be mapped to the appropriate columns in the Precedent S Excel template. This would involve understanding the structure of the Precedent S and ensuring that the data extracted from the PDFs fits this structure.

  4. Validation and Verification: The AI would need to perform some level of validation to ensure the accuracy of the extracted data. This might involve checking for consistency in dates and times, ensuring that all necessary fields are populated, and possibly flagging any anomalies for human review.

Estimated Timeframe

Assuming you have a file with 10,000 PDF pages, the time required to complete this process can vary based on several factors:

  • Processing Speed: The speed at which the OCR and NLP algorithms can process the documents.
  • Complexity of Documents: How structured and consistent the documents are. Highly structured documents (e.g., standard forms) will be processed faster than unstructured documents (e.g., free-text letters).
  • Validation and Verification: The level of accuracy required and the extent of manual review needed.

Rough Time Estimate

  • OCR Processing: Depending on the OCR tool and the quality of the PDFs, it might take around 1-5 seconds per page. For 10,000 pages, this translates to approximately 2.8 to 14 hours.
  • NLP and Data Extraction: This step could take anywhere from a few milliseconds to a few seconds per document, translating to approximately 1-3 hours for 10,000 pages.
  • Data Mapping and Population: Assuming the AI can accurately map and populate the Precedent S, this step might take another 1-2 hours.
  • Validation and Verification: Depending on the level of human involvement, this could range from a few hours to several days.

Total Estimated Time

In an ideal scenario with high-quality, structured documents and efficient AI tools, the entire process could be completed in as little as 5-8 hours. However, in more complex scenarios, it could take 1-3 days or longer, especially if significant manual validation is required.

Conclusion

AI can significantly streamline the process of creating an electronic bill of costs from a large collection of PDF documents. However, the exact time frame will depend on the specifics of the documents and the tools used. It’s advisable to run a pilot project with a smaller set of documents to better estimate the time required for your particular case.

You read it here first!

One thought on “Be useful

  1. Fascinating! The trick, as so often is the case, is the part about digestion, comprehension and understanding. Regarding AI and ChatGPT, I’m sure our UK brothers and sisters heard of the American lawyer who got into a proverbial pickle for using it and ending up citing non-existent cases, or so the story went, as if there were not trouble enough citing cases which have been overruled or are not on point or relevant. Is the UK legal system addressing the ethical issues presented by new, ever newer, technologies?

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.