Recommind panel session: predictive coding and life after Pyrrho

The judgment in Pyrrho Investments Ltd v MWB Property Ltd & Ors was delivered on 16 February, and Recommind was quick off the mark to get a discussion panel organised for 15 March. The crypt of St Paul’s was packed and people had been turned away, something which is good news for those of us who have an interest in encouraging the use of predictive coding and other technology solutions for dealing with eDisclosure / eDiscovery.

Simon Price, Managing Director of Recommind in the UK, was the affable and able moderator. The panel comprised consisted of Tim Brown from RPC, James Levy of Ashurst, Celina McGregor of Herbert Smith Freehills, Damian Murphy of UBS, James Peters of Ofgem and me.


Rather than gamely trying to report what each panel member said across 90 minutes, it is more helpful to focus on the main points which emerged (plus a few which did not but which reflect the fact that this discussions is about far more than the use of any particular class or make of technology).

It is not possible to read all the documents

The primary point is that it is simply not possible to use conventional means to search through very large volumes of documents within either the time limits imposed by the most tolerant court or within the bounds of proportionality. Furthermore, anyone who thinks that manual review achieves a better or more reliable result is kidding themselves. The lawyers must establish at an early stage how much potentially-disclosable material they have got and set that against the available time and resources. That is what proportionality means.

Hold TAR to the same standard as manual review

I quoted US Magistrate Judge Andrew Peck’s point from Da Silva Moore that we “should not hold TAR to a higher standard than keywords or manual review”. One of the lawyers on the panel described the process by which 300,000 incoming documents were reduced to 45,000 going through to final review. “Did we miss a smoking gun? I have no idea, but we were content with the results”.

This is critical to the understanding of the use of technology to reduce volumes. It is not a substitute for reading the disclosed documents (although some are happy to do this in appropriate circumstances – see below) but a means of filtering out the irrelevant material and prioritising the rest. It is not neglect or defeatism to accept the possibility that some relevant documents may be missed; that is consistent with both the Rules (proportionality, the “reasonable search” required by Rule 31.7, and so on) and brute practicality.

You are not required to look under every stone

We have case law going back to 2007 to offer to those who think they are required to be 100% certain that they have found 100% of the relevant documents. In Nichia Corp v Argos Ltd [2007] EWCA Civ 741, Jacob LJ attacked the idea that “No stone, however small, should remain unturned”.

That reference to turning every stone recurs in Digicel (St. Lucia) Ltd & Ors v Cable & Wireless Plc & Ors [2008] EWHC 2522 (Ch)  where Morgan J said (at Para 46):

it must be remembered that what is generally required by an order for standard disclosure is “a reasonable search” for relevant documents. Thus, the rules do not require that no stone should be left unturned. This may mean that a relevant document, even “a smoking gun” is not found. This attitude is justified by considerations of proportionality.

I give these examples (which I did not on the panel) to show that the possibility of overlooking documents is acceptable where it would be disproportionate to require everything to be found. You never aspired to that in a non-technology disclosure exercise (and if you did you were fooling yourself), so why do you expect it with the use of technology? For myself, I would back the consistency of the technology against the vagaries of human response, quite apart from the costs savings.


Technology which takes relevance decisions from suitably-skilled lawyers delivers documents in a presumed order of relevance. Different applications manage prioritisation in different ways, but the common element is that the documents most likely to be relevant can be reviewed quickly and by appropriately qualified people while the rest are relegated both in time terms and in terms of the skills and resources applied to them.

Ask yourself this: would you rather have the documents put into a draft order of relevance or are you content to start at the beginning (perhaps chronologically) and work your way through to the end?

Prioritisation enables early predictions to be made about timing and cost, informing the strategy of the party giving disclosure, the reaction of opponents, and the decision-making of the court.

Quality Assurance

There were repeated references to using advanced technology solutions like predictive coding as a QA tool, that is, to double-check results achieved by other means. Once you have used this mixture of technology and human input to pool and prioritise documents, it is possible (by sampling or by applying concept search tools) to test results, looking for irrelevant documents in the “relevant” pool or relevant ones in the discard pool.

Other tools – de-duplication and keywords

Predictive coding is but one tool, whose use is to be considered alongside others. De-duplication is the most powerful single way of reducing volumes without risk; near-duping has to be treated with caution because a minor change in content may conceal a major change in meaning or effect, but it does have the merit of grouping like documents together.

There was a discussion about the adequacy of keywords. Most saw them as valuable, not least for taking what you might call “ranging shots” when making rough assessments of the volumes likely to be returned by any particular keyword – you would want a rough idea of the implications of any keyword or combination of keywords before agreeing a list of them, wouldn’t you? You wouldn’t? Oh. I see. No wonder we have a problem with volumes).

Keywords have the advantage of clarity – everyone knows what you mean when you refer to a particular keyword – but that is offset by the potential for inaccuracy. Common sense is required in their use – I mentioned the lawyers who (allegedly) recommended the use of the word “Barclays” as a keyword in disclosure in a case involving that bank, whose name appeared in the footer of every email they sent. One of the panel members would like to see the end of keywords as selection tool; most see a value in them, but the very fact that this causes debate shows how important it is to be aware of the implications of such choices.

How much do you need to know?

There was discussion as to how much the lead lawyers needed to know about how the technology works. Those responsible for disclosure need to know enough to decide on the use of technology, to choose between alternative technologies, and to be responsible for the overall process and result. Again, different views were expressed by the panel members, which is what made this session interesting. My own view  is that a detailed knowledge of statistics and other scientific / technological matters is not required so long as someone, whether internal or external, is engaged on your side who does understand it.

Knowledge Transfer

There was discussion (and unanimity) about the critical importance of knowledge transfer, that is, the exchange of information and understanding between lawyers and their clients, within teams, between opponents, and between a party and the court. I told the story of the lawyers in the Alabama Admiralty Court who agreed in correspondence that they would use “technology assisted review” without, as it transpired, either of them having any idea what that term meant.

Get in a room

The most important discussions are those between the parties. Ed Spencer of Taylor Wessing (one of those actually involved in Pyrrho) urged lawyers to “Get in a room” with the other side, accompanied by whatever expert help was needed to have an informed discussion. Meetings like this can cut through the great piles of correspondence, often contentious, which consume vast amounts of time and cost. I drew attention to the fact that the Rules now require a meeting or a telephone call anyway. If you need a reference for that, it is in Rule 31.5(5) (a Jackson insert) which provides that:

Not less than seven days before the first case management conference, and on any other occasion as the court may direct, the parties must, at a meeting or by telephone, discuss and seek to agree a proposal in relation to disclosure that meets the overriding objective.

The tools and techniques

Practice Direction 31B requires the parties to discuss “the tools and techniques (if any) which should be considered to reduce the burden and cost of disclosure of Electronic Documents” and “the use of agreed software tools”.

The Master gave the impression that the term “predictive coding” meant the same as “technology assisted review”; I was keen to emphasise that predictive coding is a specialised subset of the much wider term “technology assisted review” and its variants. It was necessary (and this was perhaps my most important single message that this event) to go and see the various tools and to understand, at least at a high level, what they did, what they cost, what they could achieve, and when they should be used.

Even if you have no intention of using them yourself, for whatever reason, you are at a distinct disadvantage when you come up against an opponent who wants to enter into a discussion about their own use of this kind of technology.

Technology as enabler of smaller players

Is technology like this only for big firms? On the contrary, I said. The use of technology like predictive coding, in alliance with a technology provider and a good managed review provider, could enable a small firm to take on a very big one, with the lawyer providing the legal, strategic and tactical input and contractors (who may include barristers) doing the rest. Damian Murphy told us of the very large case which had introduced him to electronic disclosure in which his side were structured in that way.

That led into a discussion about the possibility that barristers might take ground previously occupied by solicitors. Some of the solicitors beside me thought this unlikely (the terms “snobby”, “over-intellectual” and “insufficiently practical” have come up in this context), but it is happening already. Once barristers appreciate that disclosure is not just a mechanical exercise to be performed by mere solicitors, but a tool for getting at the evidence, we will see more examples of barristers taking the lead (and the work) involved in disclosure.

Handing over unread documents

One of the dafter objections to come my way about the use of predictive coding is that “our lawyers will not hand over documents they have not read”. One wonders what bit of marketing material ever gave them the idea that this is what predictive coding is for.

Certainly you can choose to do this in appropriate cases, but so you could with any form of review. The use of predictive coding might perhaps encourage you to do so where you might otherwise not, but it is not the reason why would would choose this route, still less the reason not to.

James Peters from Ofgem said that he rather liked the idea of parties handing over documents which they had not read, although he balanced that by saying that a regulator faced with large volumes of apparently unfiltered material would usually demand a more focused set of documents.

There is the possibility, of course, that the interests of both parties might lie in doing just that – the giver is relieved of the time and cost of filtering (save, usually, for privileged documents) while the receiver can apply technology which is capable of crunching large volumes quickly and make its own selection. This is the electronic version of what Lord Justice Jackson described as the “keys to the warehouse” approach to disclosure. I am not sure that I would want to do this as the giver, but can envisage circumstances in which a giver which was sure of its data (perhaps because it had a rigourous set of information governance and defensible deletion processes) might be willing to do it. The point is that the receiver armed with appropriate technology gets to the data quicker and can make its own selection.

How informed are the judges?

The courts, as well as the lawyers, must make themselves aware of the availability of technology as an aid to proportionality As one participant said, “courts setting timetables in cases like this are going to have to get their heads round the use of technology like predictive coding”.

I have heard it suggested that you are increasingly likely to find judges who ask why predictive coding has not been considered; whether or not that has been generally true hitherto (one hears a lot to the contrary), Pyrrho is likely to increase the chance of finding a case-managing judge who will question reliance on other and older ways of conducting eDisclosure.

Know the Rules and be informed of the options

My contribution to the discussion was as you would expect if you have read my article Reacting to the reactions to the Pyrrho predictive coding judgment. Much of it was stuff which the able practitioners on the panel were able to take for granted. Others, not so much.

Read some of the recent cases, especially those where good firms were taken to pieces by the judge for evident failures of process management – have a look at RBS Rights Issue Litigation (I wrote about it here) and the reference to:

….the commitment of increasing resource to the identification of documents, leaving a diminished resource for their assimilation, without properly taking stock as to whether the process had overtaken the purpose and/or whether a more confined process should be adopted, perhaps with the agreement of the Claimants or the blessing of the Court;

Start informing yourself not only of the technology but of the people you will want on your side when the next big case arises;

RTFR (where the first R is “read”, the last R is “rules”, and the F was not to be uttered in the crypt of St Paul’s;

Bear in mind not only Pyrrho, with its helpful ten summary points at the end, but the basic first principles in the rules, those which require proportionality, require the parties to discuss scope, method and costs, and require the court to make an informed decision (which, as I have said above), it can only do if the parties give it the material on which to base a decision).



This was an excellent event, allowing us the opportunity to range over many subjects going beyond the pure technology. It perhaps helps to be Taylor Wessing, RPC, Ashurst or Herbert Smith Freehills, but there is no reason why the use of predictive coding and similar technology is limited to these very large and very good firms. Quite apart from anything else, you may find yourself against them next week and will want to understand what they mean when they pick up the phone to discuss the use of analytics in disclosure.


[Photographs supplied by Recommind]



About Chris Dale

I have been an English solicitor since 1980. I run the e-Disclosure Information Project which collects and comments on information about electronic disclosure / eDiscovery and related subjects in the UK, the US, AsiaPac and elsewhere
This entry was posted in Discovery, eDisclosure, eDiscovery, Electronic disclosure, Predictive Coding, Recommind, Technology Assisted Review and tagged , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s