David Malkinson

[00:00:00] David Malcolmson, how are you this afternoon? Very well, Ted. Thank you for having me. Absolutely, man. This is a, a great follow-up. You and I did a LinkedIn Live at ILTAcon with, in partnership with ILTA, and we had such a good conversation that I knew we had to have a follow-up here on the podcast. Um, long time listener, first time caller, Ted, so, uh Awesome.

I love to hear it. Yeah. Um, well, cool, man. Before we get kicked off, why don't you, uh, just do a quick intro, what you're doing today, a little bit about your background. Cool. So, you know, I've been in legal technology since 1999. Started in the coalface in law firm IT in various different firms, and then got particularly jealous of all the consultants that came in doing all of the interesting meaty work and, uh, taking the glory.

So I jumped ship and joined a small startup called Phoenix Business Solutions back in 2007. 10 of us in a, in a, in a room doing document management implementation. So [00:01:00] over the next 12 years, we then built that up over four continents, about 120 people globally, and built one of the, like a really successful partnership with, with, with iManage.

It was, uh... We then exited to Morae in 2019. Uh, and, and inside Morae over the last, over six years, I did various roles. I ran the, the market team, which is the preset global pre-sales team. I ran all of delivery globally for all of the legal technology solutions. Uh, and latterly, I then started the strategy team, which looked after solution stacks, owned the partnerships, all of the messaging, go to market, and the firm, wide AI strategy as well.

Um, we sold both... I've had sort of an interesting view of, of law firm. We sold to law firms and legal departments, and also Morae was an ALSP at the same time. So you get this sort of this, this, this sort of fairly interesting view of the market. And as, as AI sort of moved on from [00:02:00] '23 to '24 to '25, it was pretty clear to me that, like, this landscape is confusing, right?

There's no, not least like the technology moves every 30 seconds. But there's also this conflicting messaging from like vendors who want you to buy their tool, big consultancies who wanna sell a job lot of services to you, social media algorithms making it even, even worse. So it-- How-- D- d- I set up Signal to help sort of navigate through that noise and try and get firms to get clarity decisions quickly and create decisions that last, you know?

Does it like real proper, um, so, so the tagline currently is AI strategy that lasts. We'll see what it is next week, but, uh, but, but that's me. Mm. Okay. Good stuff, man. That's a tall order, creating a strategy that lasts in this market. Hey, you go- we'll, uh, we'll, we'll discuss that today, I'm sure. Yes, we will.

Well, [00:03:00] why don't we jump right in? So while we were in Nashville, we talked a little bit about the, the market evolving. You know, it was MCP everywhere, right? Um, it really was last year too. It was kind of the beginning that I-- It was the first time I had seen MCP take such a central role in the messaging, um, amongst the vendors.

Yeah. But you know, our friend Ben Whitwick had a run MCP, you know- Oh, good ... uh, tagline. It was so good. Uh- It was so good. I loved it. Yeah. Um- I loved that. Yeah The, uh, well, why don't, why don't we talk a little bit about like traditional data access, which has historically been API, and like where MCP fits in?

Because I think people think MCP is the answer for everything, and there are limitations to MCP, right? Yeah. Like doing broad wholesale data operations on a, through MCP has some real challenges. You know, it operates [00:04:00] in a user context. Yeah. Um, so, you know, API still has a place, but like what is your take on kind of where MCP fits and, and where API access to systems- Yeah

and data still fits? So I, I'm particularly bullish on, on MCP, not necessarily because of the technology itself, but it's just a thing that everybody agrees on, you know? The analogy here, and I'll try not to do this too, take too long on this, 1983, okay? 1983, synthesizers and drum machines didn't really talk to each other.

So you know the song Donna Summer, I Feel Love? Mm-hmm. Absolute nightmare. You probably see some music stuff behind me. Absolute nightmare to try and get those drums and the synthesizers in time, okay? So then somebody then created a protocol, Juno did, that w- that made the drums and the synthesizers be in sync, and that was great.

But there was a separate protocol for Juno, there was a separate protocol for Oberheim, and all the-- They all had separate protocol. So nothing [00:05:00] could talk to each other until somebody came up with the idea of MIDI, where everything can talk to everything else, and it opened up the entire ecosystem of synth.

And it, it created the dance music that we see today, not because the protocol was the best, it was probably the worst of the lot, but everybody agreed on it. And so therefore everything can talk to everything else, and that's the exciting bit about MCP and its architecture. It's a stripped-down version of what the API can be.

So yes, if you want speed, if you want, if you want certain nuanced use cases that MCP can't do, g-go build an A-- go and build directly to API. But if you want that sort of system to system sort of general, it is huge for, for, for tying together an architecture together. 'Cause, you know, uh, where I think it's moving from now is not necessarily which tool to buy, it's how do you get them all to work together?

That's the, that's the hard bit these days, right? Yeah. And, um, [00:06:00] what about the user context piece? So when I think about like for like end user tools, MCP makes a lot of sense, right? Yeah. Because you have a logged in user. Um, but, you know, when I think about like API use cases, many times these are, these are jobs running, right?

That extract ETL data from one system to another. Yeah. Or maybe behind the UI, APIs will go access data and present it in a, uh, in a presentation layer. Yeah. So like what, um- What about the user context piece and how does that limit scenarios where MCP makes sense. Yeah. I, I think that that's where MCP is right now.

It's an extension of you as a user, which it makes sense, particularly when you're talking to, like, my background, document management systems. You need to know that users can't access something that they couldn't. So funneling everything through a user context is crucial. Big migration things aren't gonna work like that [00:07:00] when you wanna migrate things from w- from, from one to another, or some levels of reporting need agentic access.

But I think where we're at q- right now, nobody's quite figured out how agents can access systems safely when they're disconnected from the user context to give then information back from users. I can't wait for that to happen. You know, the, the, the real pending use case out there is, is firms have gotta move to a closed-down system for that exact same reason, because agents don't care about whether they should see things or not within a system.

So they'll... You'll do a search, and it'll say salary 2024.xls. You'd never open that as a user, would you? Never in a million year. Well, you might do, but you shouldn't. Agents are not gonna care. Not only are they gonna open it with your name on it, they're gonna give you the entire thing and tell you how much, uh, how much comp happened, happened last year.

So this is really important. But when you then move to a closed-down system, how do you get access to [00:08:00] knowledge within a system? And nobody's quite figured out to jump over the wall and to safely deliver the information back without breaking confidentiality, information, and all that type of thing.

That's the next thing that people are thinking about. So it's more about agentic access, which needs, like, this needs an entirely different access model than what we've got today, you know. So there's some interesting things happening in banking, in, in finance, but, but trying to bring that into a legal access to information, which is what people do a lot of the time, is, uh, is a really interesting thing.

Looking forward to it, yeah. Yeah. So, you know, we, we solved this problem internally at InfoDash using kind of a different angle. So, um, InfoDash is a internet and extranet platform, and we deploy in the client's tenant, and we have an integration layer, as you know, called Integration Hub, that has API access into all the back office systems that they use, and then it presents a unified security trimmed [00:09:00] API that respects ethical wall boundaries.

Yeah. And we do that by integrating directly with NTAP walls, or sometimes we pull the wall, you know, include, exclude lists directly from the source system, like financials. PMS is a good example. Like, many times those kind of ACLs will be embedded in the, in Adherent or Elite, and we can extract them there.

Yeah. So we kinda, we w- we had to solve for this before MCP existed. Yeah. And, um, but MCP does create a, uh- An external mechanism, you know. So where I see like the Integration Hub still playing a role is, again, in those wholesale operations. Like we have a client, and I can talk about this because it's public, um, Gowling in Canada, that they have a, um, a fairly well-curated corpus of documents, like maybe 50,000, that [00:10:00] represent like legislation, and then they have about a million rows of data that represent like annotations and, um, on that, on the, that corpus of documents.

And we just integrate with those data sources through, um, Integration Hub- Yeah ... and then, um, Azure AI Search accesses it through Integration Hub, crawls, indexes, vectorizes, and builds an AI index that can then be used in other places. Yeah. Like doing, doing that through MCP wouldn't make sense. Yeah. It's just there's, there's too much-- it, it's a wholesale operation.

Would you agree? You're, you're effectively allowing... You're, you're controlling the gate, allowing agents to go away, collect their information, and gating it on the way back in. Yeah. That's a tough engineering challenge, but at least a decent way, a d-decent way of doing it because it's speed, isn't it? At MCP, you gotta obviously pay for the tool call, spend tokens trying to figure out which tool to get.

You're then gonna go through limited API, which you may or may [00:11:00] not own, and then, uh, and then coming back again. It's a neat way of doing it. You could probably make that call via MCP that then goes out to spider through and, uh, and come back. And I'm seeing a lot of that, a lot of like MCP gateway architecture, where you have one route in fanning out to all of those systems and then coming back again, because you can actually measure what's going on.

You can measure who's doing what, you can measure token cost, you can measure all these things which is missing from so many people's architecture at the moment, is, is figuring out what's going on. And also the, the tool calls are gonna get expensive. Imagine if you've got 12 different tool calls. What are you doing?

Spending tokens figuring out which, which, which tool calls to get. You've got 12 different calls, all spending tokens, and then you've got to then cobble that back together on the way back in with no measurement whatsoever. That's, that's gonna get expensive. You then times that times 3,000 people all doing those same informa- all of the time, that is not a workable architecture.

And, uh, uh, [00:12:00] that sort of gateway approach is, uh, the... And they were like in the, you know, at Ildicon, they were just in the corner. There were a couple of people doing cool stuff like that, but it's, uh, but they were, they didn't have a, a booth the size of a football pitch, that's for sure. But- I know ... but really, like really important, interesting stuff, if a little boring to, uh, to describe to the general populace.

But yeah, boring is cool again, as I keep telling everybody. 100%, yeah. And again, this, this space is moving so rapidly that getting your arms around all of the potential cost vectors, right? Um, you've got, um, all sorts of... Which is a good transition into kind of the next thing we were gonna talk about, which is How some of these systems are navigating access, right?

So these SaaS platforms, like there's been developments, like Salesforce initially came out with a fairly aggressive terms of service around [00:13:00] access to data that lived inside of their platform, and, um, because they have their own AI to sell, right? Which really caused an uproar. Like I heard people in really high places talk about, "Salesforce fix this or we're leaving."

Yeah. And, you know, some of these vendors are in an interesting position because if they just open the floodgates, A, they have cost that has to be managed, and B, when you go around the UI, they really just become a data store- Yeah ... that diminishes their value, right? Okay, yeah, you're the system of record today, but if you're not using all of this R&D investment that, the, that was put into the presentation layer- Yeah

the ROI on these SaaS platforms is impacted. So it's interesting stuff that's happening right now. It, it, it was, it was interesting back then. Salesforce is one of them, DM vendors were another. If your- Yeah ... moat, I don't really like that word, [00:14:00] but if your moat is the data that you hold, it doesn't come naturally to just give wholesale access to it, even though they've been doing it for years anyway via, uh, via the API.

But suddenly, AI access into their data seemed a bit... So, so they did, they didn't quite, didn't quite figure that out, so it was, it was, um, because then, then people, to your point, walled garden, they were being, they, they were being accused of. But suddenly they realized they opened the gates and allowed access to it.

It actually made them more sticky, paradoxically, because suddenly, okay, cool, I can now get access to my data. I know the data's over here. I know it's secured. I know it's stored in, in the right place, and it probably made them more sticky, 'cause all people are gonna do if you can't get access to it is, is it just naturally happens, is it just, you know, if you can't access data, things turn up in OneDrive, don't they?

They turn up in data warehouses and slowly just, just because you've gated access, your value diminishes. And I think people have figured that, figured that out [00:15:00] now and, um, and we're a lot better for it. What's really interesting now is pulling all those data sources together so that agents can now access it as one, as one call.

You know, we, we, we touched upon that before. We, we-- You certainly see a lot of data at the moment. The, it sort of reflects the org structure almost, doesn't it? You know, you've got experience data owned by marketing, you've got finance owned finance, you've got, uh, you've got CIO generally looks after the document management system, and all these different systems that don't quite agree with each other.

There's no sort of, to use a technical phrase, key that sort of... So an agent goes and visits all of that and goes, "What?" Right? "I, I don't understand any of this." So you just end up-- And, and, and agents need access to everything. They don't think, they don't care about the silos within a firm or the, all the different information.

You just need, you need access 'cause, 'cause that's how people think. They'll think of, you know, show me a similar [00:16:00] matter and show me the experience that we've got on all the pricing or whatever. And they would just go, "Well, why doesn't this work?" Oh, is this because 20 years ago we made a decision to have two different systems and the, the client numbers were slightly different, and we never quite got around to matching them because that was really, really hard, and we can sort of get away with it.

I don't think you can get away with that for much longer. So that, that having that sort of, you know, unifying platform I think is gonna be a big, big deal. More, more of a deal than, than, than a lot of other use cases that you see. Yeah. And some vendors, including, from what I understand, some of the legal DM vendors have implemented what feel like draconian, um, policies around this.

Like, I've heard that vendors are Asking for a rev share from adjacent vendors who access these systems Mm-hmm. And which that's a real problem, because if they went that direction with [00:17:00] InfoDash, we would just pass it along to the customer. Yeah. So the customer would ultimately not only just pay for the system, but any adjacent systems that leverage the data in there because, you know, uh, a DM vendor decided that they wanted a rev share from, you know, a, a, a, a partner, a partner, I'm using air quotes, um, company.

So like- Sure ... this, this is, um, that seems like it's gonna create a lot of friction. Yeah. It's, it's tough. I... You know, to be fair to them, the, the, you know, they can't-- just like you can't predict token cost, they can't predict API usage. And st- and API usage equals compute, so either they have a choice of, A, somehow protecting themselves against it, or B, slowing it down, which is what Microsoft do, and just throttling everything and upset everybody that way.

So it's, it's imperfect. But can you imagine being the buyer? Being the buyer of it. Like, so you buy a system and go, "Okay, right. So I need license count." Okay, so I've got this many users. I [00:18:00] think this many user are gonna use it. I hope they will. Storage costs, second thing. Okay, which modules to buy? Okay, I'm gonna figure that out.

Which package do I want? Okay. Then you've got AI usage costs within the platform, right? For the AI tooling, so that's a separate thing, which is consumption-based. Then you've got AI tooling from outside of the platform, where you've got token costs on, or credits on another platform coming into it. And now you've got to manage API costs as well, which may or may not be borne by your partner.

Then on top of all of that, you've then got to get your crystal ball out and predict adoption patterns, and you, you're never gonna know what they are. You know, you can try some sort of benchmark. You're gonna be wrong. Uh, and is it, and it's not... Is it a good thing if the adoption's lower than you expect?

No. Is it a good thing if it's higher? Sort of yes, but, but, but, but, but, but not really. You know, I can see this being... You know, like remember Microsoft licensing back in the day, like how complicated that got and, like, you needed dedicated [00:19:00] consultants to come in just to tell you how much to pay Microsoft.

I can, I can well imagine this being like a standalone skill of figuring out all of that, because that is just, you know, a simple question, how much is this platform gonna cost me? Like, what a time to be alive, Ted, eh? Yeah. Well, and you know, Microsoft is, is back to really opaque Uh, pricing again. Now you've got consump- you know, Copilot credits, for example- Yeah

that can get consumed from different places, agents that are built in Copilot Studio, um, Cowork. Um, you know, there are-- Uh, it is really difficult in the Microsoft space, which is where we play- Yeah ... to tell customers, um, like, uh, the, the Gowling example is a good one. Like, you know, they had a very logical question, like, "What is this gonna cost for Azure AI Search?"

And the reality is we had to test it. Like, it actually had to go live to [00:20:00] see the usage patterns- Right. Yeah ... because it's a client-facing system. So we couldn't model it internally. We had to essentially Uh, find out after the fact, which is undesirable. So like h-how are firms gonna navigate this? You need to just start like right now measuring things, exactly what people use stuff for and why.

Like you've just, you've got to start that at first principles, measurement framework, figuring out your use cases, getting your use cases and systematizing them into sort of firm owned processes. So at least you know what people are doing and how much that particular workflow is gonna cost you. Added benefit, you can now tell your clients exactly how many times you've used, uh, you've used it on their matter.

You tie it to a matter, you then you may not choose to do it, you can pass some of that cost back onto the client if the hours are coming down. You stand no hope of doing anything of any of that unless you're measuring things properly, and that's both from a human level and a, and a, and, [00:21:00] and, and sort of ambient collection of, of, of, of metrics to sort of try and, to try and do that, you know.

So it's, it's trying to make things as, as, as, as predictable as possible without getting in people's way, you know. Which is why I'm just, we talk about this, don't we? Like, uh, you know, own the skill, rent the model type thing, because that's what it is. It's just breaking down the process into its constituent parts and putting the right parts to the right model at the right cost, so that when something like Jev happens, it's happened this week, you know- Mm

you can react to that, and you can bring your cost, you can, you can bring your cost down because you've granularly gone through those repeatable processes and you own them and you know them and you understand them and you've got ownership of them and you've got champions constantly optimizing them.

That's sort of what thi- That's where things are going/should be going. If you've got all of that, all of that stuff sort of in a proprietary platform [00:22:00] that you can't get your hands on, then you just have to ask the vendor, "Are you using..." I mean, Jev is a, one of many examples, you know, "Are you using a model?

What's it doing to my credit price?" And you have to hope that they give you a decent answer on that, you know. It's, uh- Yeah ... yeah. And to your point, like the world is changing, so Jev is a, a great example of that. I mean, this is a new paradigm, right? It's not a, it's not a language model in the traditional sense, right?

No. It's a, for those that aren't aware, it is a, um, it is an AI model that gives back probabilities of answers. It does not generate language, so to speak. Yeah. And it is about 400 times less expensive to, um, uh, I think there are no output token costs and- Yeah ... it, so it is-- That's what I mean, this world is moving so fast.

Like if you're the architect of these systems- Yeah ... it's like Jev came out of right field, at least for me. Um, and it has a very [00:23:00] compelling value proposition that if I were a, a systems architect, I'd be thinking in great detail about the processes I have deployed and how I can leverage a tool like that to reduce my overall costs.

Absolutely. And, and, and you've, you know, as I say, if you understand the process deeply from start to finish, you can just swap out the models as they come in. You know, models change all of the time, don't they? You need a posh model for some hard thinking tasks. You need a stupid model for others. You need a d- and, and sometimes you don't need a model at all.

You need a deterministic process that's just- Workflow ... just a workflow, right? Yeah. And call that via MCP, just to complicate mat- matters further. But, but you know, the user, the end user pressing the button shouldn't care about any of that. You know, you should just have the right guardrails, the right tuning, the loop processes.

But underneath the hood, it's our job, if you will, to sort of abstract that all away and just have, have the best blend of quality and cost, and you only do that if you break that [00:24:00] process down into its constituent parts, I'm repeating myself here, and understand each, each bit of it. And then it's way less of a big deal to swap these, swap these things out, you know, way less of a big deal.

Yeah. Yeah. You know, I ha- I have a theory, and this theory has evolved over time. You know, when, when AI, I, you know, kind of the, the moment for me was Opus 4-5 that came out in December, November, December timeframe, that I really saw a big step change in capability and, you know, it, it, it shifted my thinking and now I'm kind of shifting back.

Um, but you know, I thought, man, the, the way the, the innovation curve on these models is so steep and s- and steepening, you know- Yeah ... not just the velocity, but the acceleration- Mm-hmm ... um, that we're gonna be able to deliver work with minimal human in the loop time sooner than we anticipate. So my immediate [00:25:00] reaction to that was this is going to be, AFAs are going to, um, start to gain traction sooner than we expect.

But I'm actually, I'm kind of shifting course on that back to where I was originally, that it's gonna be more of a slower burn because I'm seeing how much human friction there is, not only on the, on the sell side, i.e. law firms, but on the buy side with level of comfort, um, with ev- w- even with knowing the right questions to ask.

Like we're gonna talk about Chinese models in a minute, but like, you know, are, are GCs and CLOs issuing OCGs that mandate that law firms let them know when a Chinese model starts delivering work? Even if it's hosted inside the vendor SaaS platform, there are biases that as a consumer of the output of that, I would wanna have awareness and I [00:26:00] would wanna understand what compensating controls exist within the vendor to, um, mitigate those biases.

Yeah. But at a minimum, I would wanna know, maybe even I'd want permission, and like that risk framework doesn't exist. Like it's just- Yeah ... this is brand new, right? So like- Yeah ... lots to unpack there. Well, it, it is, and again, it's just you've got to tr- I'm just gonna keep repeating myself. Measure, measure these things against the client, no matter what model was used, what process was used, who did it, when, how much, what was the token cost.

And, and it's got, that's got to be There's got to be configurable per client as well, 'cause some clients are gonna go, "No AI, AI whatsoever." Some clients are gonna go, "You can use a frontier model, but you can't use a Chinese open way one." Some are gonna say, "You can use it, you can use a local model." That's cool.

If you want, you can do it on your own infrastructure. That's another thing. Some people will say, "You can use whatever model you want, but anonymize all my data before [00:27:00] it goes out." And there are, there is levels of what you're gonna do per client. That's just-- It's not definitely gonna happen, but it probably is, isn't it?

You do start thinking, thinking quite-- It, it's a whole new skill to, to, to cater for, you know. And I'm, I'm a, I'm an extreme optimist about that type of thing 'cause this, that's Jevons paradox like right there, isn't it? It's like new roles being created to, to, to put all these things together, which will inevitably feed into the, the core business model of most firms that we're dealing with, you know, so.

Yeah. And that's, that's where Jev gets its name, um- ... from Jevons paradox, which is interesting. Um- I'm gonna pretend I knew that all along, so, uh I just happened to read. I've been paying attention. I've been reading about it. Um, but, you know, if, if it is true that the billable hour is going to die more slowly than many of us predicted or even hoped, the pass-through costs [00:28:00] ha-- right?

If you're working on an hourly basis, you can't absorb all of the tech costs associated. Those have to get passed through if hours go down, right? Yeah. And neither the buy nor the sell side is really ready to go full bore AFA. That- Yeah ... you will hear clients tell, "We're ready." Uh, you're not, 'cause you're still asking for shadow billing, and there's not a level of trust between law firm and client where- Right

clients feel comfortable just saying, "Okay, charge me a million dollars and we're good." Um- Yeah. So how do- Yeah ... how do firms think about Um, attribution, right? Like, not only the client number, they need to know the matter number, right? Uh-huh. And potentially the task code that's being worked on- Agreed.

Absolutely ... so that the client can audit it. Like, how does that happen? Yeah. I mean, you know, with, with billing, there's a l- with so much talk about the billable hour, it's just, it's all it is, is a pass the parcel of risk, isn't it? From one , from one person to the other. It's just like, "Who's gonna absorb the risk and how?"

That's all it is, 'cause [00:29:00] both, you know, law firms gotta make a margin, and corporates don't wanna paying, pa- paying, paying o- o- over the odds. So, however you wanna package that up, it's still, the, the people are a fixed cost. That doesn't, that doesn't go away on the law firm side. You need to make a margin on every single matter that you do, or you're gonna try your hardest.

However you package that up, could be billable hour, could be AFA, could be whatever. But technology cost will come into it. You know, it w- it, it, it sort of has to because it's just so... The, the cost is, well, depending on who you believe going up or down, over going down over time. But, you know, irrespective of whether you're gonna charge back for it, you do need to know what it is per task as well, not per matter, per client, per phase, per task, per, per everything.

You gotta know, then at least you can make a decision. And, and clients that I, uh, on the corporate side that I speak to, they're really starved of information of, of how law firms are using AI on [00:30:00] their matters. It's like they're d- not, not least because they wanna sort of know that they're not being overcharged because they're under pressure to reduce bills.

A story as old as, old as time. But they wanna learn how this stuff is being used as well. You know, they, they, they're trying to figure out their own strategy and they're looking to their law firm clients who they perceive of being quite far ahead of the curve. Some are, some aren't, you know. And, and, and they d- they, they, they wanna learn and absorb actually how you do that themselves so that, you know, the clients who can go, "Okay, we've used AI on this particular matter.

We use it for this task here, here, and here. We found actually that on this particular task, we had to use a posh model, and the verification was horrendous, therefore it wasn't worthwhile doing. But on these tasks here, here, and here, it was great. We now have a roadmap to bring in these particular models and try this to optimize this a little bit.

And dear Mr. Customer, your bill is now X. My pleasure." You know, that's the conversation that, uh, that both sides want to do. And if you get it right, it's a win-win. It's a [00:31:00] positive flywheel like that, that, that, that, um, that, that, that whole thing. But it just, it just starts with just, just breaking down things and, and, and sort of, not being such a buzzkill that you take all of the, all of the, the, the, the, the thinking out of it.

But, you know, at least you sort of understand what's going on. And, and the, the most interesting work I do is that, is, you know, that's legal engineering per se, isn't it? You know, it's, uh- Yeah ... it's that. Um, you know what's, what's interesting, um, is we are grappling right now at Info Dash with that exact question around transparency f-from clients to law firm activity around how is AI being used, how is it impacting the e-ec-economics of the relationship.

So our extranet platform, like I said earlier, deploys in tenant inside the law firm. So we have hooks through Integration Hub. We can expose anything that they have in their systems. Mm-hmm. Um, historically, we have run [00:32:00] into friction with law firms not wanting to share. Um, I'll tell you a quick story. Uh, a buddy of mine who runs a law firm, he's also a client, one of our bigger clients, and, um, he-- I was telling him some of the benefits.

He, he-- They're, they're an extranet customer, so they can, they can provide a high level of transparency around what they're doing. Mm-hmm. And he told me about a dynamic that exists in his firm. He's a listener. Uh, I think he listens to every episode, so he's gonna know I'm talking about him here. So during the conversation, he said, "Ted, on our invoices, we used to have year-to-date billings at the, in the footer of the invoice, and a partner came in and said, 'Take that off.

I don't want the clients using it, you know, as leverage in the conversations.'" Yeah. Like, and, and that is really-- It's like, dude, they know, right? Like - They know ... they, they write, they write the checks. They know. They know. [00:33:00] It's not like they haven't got legal spend systems and whole AI looking at every single thing and doing entire benchmarking of their panels every single, you know, every, every They know.

There was a, there was a sort of fear, isn't there, when you sort of got the, like a high billing client. Just don't push your luck because somebody's gonna notice at some point, aren't they? And, uh, and, uh, reduce your spend or kill you off or go and find somebody else. Th-th-there is, there is that just trying to sort of Flyer a little bit under the radar, but it makes no sense, does it?

No. There's a lot more transparency in terms of, in terms of what happens. And I think it's, it's-- there's a human element to that, isn't there? You want to-- You're the expert. I'm the expert right now talking to you. I feel rather uncomfortable if you ask me a question in my sphere and I didn't know the answer, and I just...

I'd probably just make something up, right? But I shouldn't. I should, I should just say, "I don't know, I don't know the answer to that, Ted." I should, you know. But, uh, but, and it's, it's, it's that when you're [00:34:00] presenting yourself as, as, as, as, as the expert, you don't want to show what you don't know, because that's just the way, that's just the way we all think, isn't it, when we're here.

Well, you know, and, and it really speaks to the, um, you know, and I have to be careful because, uh, law firms are my customers, but, um- Absolutely. Yeah ... I've built my brand on speaking candidly, so I'll, I'll say this, is law firms have been very, um, not done a very good job focused on the customer journey to really being client focused.

Like, for the top relationships, absolutely, you have a relationship partner that when the client calls and says, "Jump," they ask how high. Yeah. But broadly speaking, and I know this as a consumer of legal services, been an entrepreneur for 32 years, um, they are not, they have historically not been, uh, customer focused l- the way other industries are.

Yeah. And there hasn't been a level of trust in the relationship between [00:35:00] corporate teams and their law firms. It's a very interesting relationship. Like, when you get them together- Oh, man. Yeah ... like at a, a pricing conference, like at the highest levels, very polite. You go one level down, like legal ops and, and pricing, the gloves come off.

Um- Like I was at a pri-- I was at a pricing conference last June, and we had a panel of three, I think they were like deputy GCs of really big companies. Yeah. And they were telling the audience that they wanted to have predictability. And these pricing guys were like, "You don't even know what you're asking for when you ask for predictability."

And like, oh, dude, I thought- ... I thought a fight was gonna break out. But, um, it's, it's really interesting. It's very polite at the most senior levels. Yeah. And then but you go one or two layers down the org, and the frustrations start to really become visible. So it's-- I, I, I feel like it's a, there's a, there is some focus that needs to happen on that relationship dynamic.

Yeah. Yeah, yeah. I, I, I, I do. [00:36:00] There's no such thing as an NPS score, is there? Things that, you know, customer... There's no such thing as Customer success. Now, part of that is the, is where the relationship sits and who owns that relationship, and it's the sort of the partnership model. You know, I've been at, been at other firms where there's a f- where you have a sales team, uh, bolted onto an existing big consultancy.

Who owns the client? There is a real... "It's my client, thank you very much. I'm the one who sends the invoices. I'm the one that does that. No, I don't want a customer success person in there, thank you very much. Don't want a pricing person in there. Do not want a anybody else, because this is, this is my client and I have the conversation.

I own it. And I like-" You get that, don't you? But there are some things that fall by the wayside as, as a result of that, and it's just, it's just that who owns that relationship is, is what drives all of that. So there's the, the antibodies really come out when, when can you imagine there being just like a hardcore sales team in a law firm?

Can you imagine what, [00:37:00] can you imagine what, what that would look like? You know, like, uh, that would be- Yeah. Well, you know, that's a, that's a great point. So I, I am also a believer that the go-to-market thinking for law firms has to change. So historically, how law firms have gone to market, the way they differentiate themselves are people, um, practice depth- Mm-hmm

global footprint, brand, which is a kind of a combination of those things. Yeah. Like, those things are still going to matter in the future, but for some of the more, um, operate the company work that is largely automatable or, um, fully automatable- Right ... the go-to-market motion is going to look different.

Your brand and your people aren't gonna be as important as your how you've structured learning loops into- Mm-hmm ... your practice, how organized your knowledge and wisdom is that lives inside the firm, [00:38:00] your pricing- Yeah ... your speed to delivery, and the quality that you output. Those are very traditional elements that normal companies, not law firms- Yeah

have to build their go-to-market motions around, and that's a very foreign concept for law firms. Yeah. And we're entering into a phase where those things are gonna have to be incorporated into their strategy. Yeah. I really should have had a look at some of the AI native law firms and see what their go-to-market is.

Ones that were good at go-to-market or certainly a lot better, depending on your point of view, is the, is the, um, B2C firms, billboards on the, uh, billboards on the highway. You know, your, uh, trip and slip stuff, which is their-- That's their whole deal, isn't it? Because they're just, it's just all about new, new logos, if you will, the en- the entire time.

I've not... Have you seen how much the, you know, your, your Crosbys, your Norms up there, is their go-to-market changing over time? The messaging I'm seeing is like, "We're not like the big bad old law firms," which is which is fine, but, you know, where's the sort of value prop? [00:39:00] Where's the, uh, you know, where's, where, where's that?

I d- I haven't seen it, but admittedly I've not been paying that much attention to it, but, uh- So interesting. So I saw a, um, one of these AI native firms, I won't say which one, that they were at a conference, and their shirts that they were wearing in the booth said, "Fire big law." That's, that, that's their go-to-market.

Um, and you know, I had- It's been so long, isn't it? It's just like- Yeah ... fine. Well, I... And, and you know what? I had, uh, Mike Schmidtberger, I think he's the chair at Norm, on the podcast a couple of weeks ago. Okay. And he was very clear about, um- Their, their go to market is not in terms of execution, but just kind of overall broad strategy, the segment of the market that they're targeting, it's very different.

Like the first generation of AI native firms like the General Doubt Legals, the Garfields, the Crosbys, that first generation was really focused on lower value, um, h- highly automatable [00:40:00] work. The Norms, the Irving Technologies, like they are, those firms are going after high value work. Yeah. Fund formation, like maybe lower, um, s- strategic M&A advisory.

Like they have real, like Norm has real like- Yeah ... talent from big law. They can execute on this. So it's like, I f- I feel like this is really an evolving space. Yeah, absolutely. I mean, you-- And what's their go to market Norm in, in terms of, in, in, in terms of, in terms of what they're doing? What's the-- If I were to go on their website right now, which I could but I won't.

What's it, what's it- Yeah I, I think it's largely relationship driven. Um, I'm, I'm, I'm making some-- I'm reading some tea leaves there. Yeah. But the people that they've brought over have been in big law executing this work. And l- I'll tell you another, uh, go-to-market element that I know for a fact- Yeah ... is their cap table.

Yeah. Uh, Blackstone is their biggest capital provider. That's right. That's [00:41:00] go-to-market right there. Yeah. That is gonna happen once they've, uh, once they have a miss in a, on a particular quarter. You might find some salespeople popping up in, uh, in, uh, in those firms. That, that will be an interesting watch because they are absolutely in- not only incentivized to do it, but they're gonna have to, aren't they, in order to, to sort of get that market share.

So, you know, guarantees on turnaround time would be something I would do. I'm just making this up now. You, you know, guarantees on turnaround, money back without quality. These are the sort of things Alex Hormozi would sort of tell them, wouldn't it? Just, just Totally. He, he would, you know. What are, what are those guarantees?

What's the offer per se, and how do you, how do you package, package that up? And then how would-- And big law are probably gonna have to react to that. Yeah. You say, yeah, it's, uh, yeah. So, um, there's one topic that I, I really wanted to, to touch on with you. I had a post this morning, so we're recording in late September.[00:42:00]

Um, this will be in the review mirror by then, but I think the, the, the topic it will be relevant in the future is, um, there was a Bloomberg article this morning that talked about Harvey's net margins went from 50% to negative 50%, um, after they kind of released their new agentic tools in, in the spring.

Uh-huh. And that was the motivation behind Tenent, which is, I think it's Kimi 3 that's been like post-trained, um, and now they're very much recovering because the token cost got reduced by what- whatever it is, 90%. Yeah. So this is the question that it brings up. As a client, are you gonna wanna have a visibility and maybe even some say so in how these open weight models get used to deliver legal work?

Because a- again, even if they're inside the four walls of the vendor, um, you, you protect [00:43:00] data leakage, uh, and privacy, um, events that way. But what about the biases? Like, can they be-- I, again, I'd wanna know what the compensating controls are. Like how-- Now, are those gonna start showing up in, in OCGs? Are they showing up already?

Like it, it really brings some interesting questions to the table. Abso- absolutely it does. I mean, they're what you're one event away, whatever event may be, and there's been plenty of them. You're one event away from that being a real thing, you know, in terms, in terms of you need to know not where your data is going and how it's being processed.

Processed. Processed. I mean, that's just, that's a, a tale as old as time, isn't it? You know, from GDPR in, in, from, from Europe onwards. You've just, you've got to know that so you can, you can verify the quality of the app. I guess if you're, if the concern is, is biases within those, w- within those Chinese models, you just have to overlay a human on it to read every word, which most people do, at least, at least, at least for now, and that's gonna [00:44:00] sort of get you out of trouble because it doesn't matter what, where, where it was processed, the person delivering the work product to the client is you, the human being.

It isn't the model. That'll get you so far. You go one step down any sort of automation, you wanna build a product around it, make that client-facing to automate things with some level of bias in there, that's gonna be problematic. Is, is, is it, it's a bit strange that all of the open weight models just come out of China.

There's not many there. I know there are some outs- outside of that, but you would have thought there'd be some sort of ge- more of a geographical distribution of, uh, of, uh, uh, uh, of those models. And there's some really interesting studies in terms of, in terms of what they do, in terms of the decision making process for agents in- inside those models.

And ag- it's very, very... What, what's gonna happen when, when, uh, when th- this event happens, whatever that might be? You've got to handbrake turn and choose another model, and you better be, have [00:45:00] everything decomposed and, and, uh, and abstracted away in the right way so you can move quickly 'cause you're gonna, you're gonna have to do.

And I imagine Harvey are pretty probably well placed with their, uh, with their valuation and funding in order to, to, to, to do that. But that, I mean, that's gonna be a real thing. You know, t- like we were talking about before, the different layers of security and, uh, and, uh, client confidentiality and, and thoughts around it that you, you've got to start thinking it, thinking like that.

Totally. Yeah. There was, um, there, there was one other thing that we touched on in our LinkedIn Live that I wanted to just get out on the table because there's been a development since we chatted. Um, we talked about Morgan+Morgan. Yeah. Um, they're, they're a $2.5 billion revenue company with something like 1,100 lawyers.

And- Yeah ... what's happened since our last conversation is they announced $1 billion in R&D or, you know- Yeah ... around AI. You know what's super interesting, and we touched on [00:46:00] this a little bit, and I don't think, I don't know that my audience has a full appreciation for this dynamic, which is who has more experience with outcome-based pricing than the PI firms?

Yeah. The answer is no one. No, no, no. They don't- And go to market. Oh, all day long. Like in 1999, my first ever job, and like I was, you know, it was, uh, in the IT department, and it was for a, uh, PI firm in Manchester in the north of England. 100 user firm. We were doing document automation back then. That was like a whole thing.

You know, just bringing in people just to, just, just, just, just low cost, low cost graduates coming in, doing that first pass review. All of their work came in via an agency, and that agency is just smashing the phone in call centers all day long, delivering leads to a law firm. The law firm then does, uh, does the, does the qualification with cheap resource, so graduates at the time.

And then we automated as much as we possibly could out of, out of, [00:47:00] out of the process. We w- we had Dragon Dictation, which didn't work particularly well. We had this like really old cobbled together, um, document automation thing, but you know, we got it working. And, uh, and they were just moving things.

That's the whole deal, isn't it? Is just- Yeah ... just margin, margin. Every single In fact, my next door neighbor runs a, uh, runs a flight compensation firm, and he's been doing it for 20 years, and he's, he, he says it's been drilled into him for 20 years, quarter turns of efficiency across the entire organization because every, it all matters in a, in a conditional fee, fee arrangement, you know.

And their processes are absolutely sacrosanct to them. Yes. We, we, we wouldn't let us in to, to, to build a... A lot of them would just say, "Yeah, no, I don't want an external consultant coming in, building my process, thank you very much," 'cause this is, this is my business, is this process, and the automation within it.

That's like the whole deal. So, so, [00:48:00] so we're so-- it's a closed shop. You don't, you don't get in there to actually see what, see what they're doing 'cause that's, that is their IP. There's a lot to learn, as you say. Yeah. It's a, it's like a Formula, Formula 1 pit crew, right? Where every, you know, 10 milliseconds translates into competitive advantage.

Yeah. And what's interesting about this that I find slightly humorous is how frowned upon the billboard lawyers have been in the big law circles. Uh. And now they need to look to this end of the market to, to really, you know, people who are much more experienced than them delivering outcome-based pricing and have a mature go-to-market strategy and execution.

It, it's-- I don't know, I kind of- Oh ... I find it funny for some reason. I do. I also find the announcements of we're gonna spend X number of dollars over, uh... I find that funny. You know, it's just- Yeah ... everybody sort of loses their mind on LinkedIn, don't they? They said, "Oh, look, look what's happening." Nobody's gonna go and check, are they?

[00:49:00] Exactly. I thought I, I might do an announcement just to see, just to see. I'm gonna spend a trillion. I don't know. A trillion over 100 years. You go get 'em, Dan. Yeah. And I love that when people just like, th- all they do is, like, make an announcement like that, say nothing else, and then the whole of LinkedIn just explodes with what they've done.

It's good marketing. It's great. It's- Yeah. It really gets them in a conversation. It's, uh, it really, it really makes me laugh when those, when those things come out. I know. But yeah, how do you-- I mean, are, are, um... Is big law gonna be able to bring themselves to- To ... to really exploring? Like I had a, I had a, I can't remember her last name, Lucia.

She used to be at, um- What was her firm? Um- She's a dentist ... she was at Decker. Yes, exactly. Yeah. And she had, she's s- she's such a great thinker. She had her dentist come in and talk to her law firm [00:50:00] about customer service, and she goes, "I actually look forward to going to my dentist because of how this guy delivers the service."

Yeah. Nobody looks forward to going to the dentist. So thinking, like, out of the box like that, like Chick-fil-A here in the US brought in Ritz-Carlton in the late '90s. Yeah. And, like, the other fast food chains laughed at them like, "What are you doing, bro? This is fast food." Yeah. And look at how that turned out.

Yeah. So, like, are law firms gonna be able to tap into these other, like, l- unconventional sources of inspiration? O- once the incentives lines up, they surely will. You know, they- Yeah ... they, they surely will. You know, they, they... I do find it, it's a bit, that R&D, they don't call it R&D, they call it innovation, you know.

Mm-hmm. They, they have this, this still, still, I mean, probably less now. There was this whole idea of fee earners and fee burners, you know, just like, just, just, yeah, this whole... And it's just, it's just, that's not gonna be what the business model looks [00:51:00] like in not that long, you know? Right. It's, it's, it's about, you know, process engineering and, and, and, and, and having that sort of mix match.

It's g- Got to be like that. You know, of course there will always be relationship-based, highly complex, high stakes lawyering that needs to happen. That will look the same. Um- Yeah ... you know, and the rest of it will just look different, and there'll probably be more of it as well. And, um, you know, that's, it's, uh, yeah, it's...

I just love this inflection point we're in. Something happens every single day. You gotta think on your feet. It keeps you intellectually stimulated, doesn't it? It just, just, there's just another problem to solve, another thing to think about every day. I love- So it's, it's great. I love it. I- And- I enjoy, it's like, you know, a stock market that is flat, it's hard to make money.

Yeah. If there's movement in either direction, you can, i.e. change- Yeah ... you can, you can figure out how to, how to make money. Yeah. But, um- [00:52:00] Yeah. And it's a, it's a difficult, m- different way of thinking that you've gotta do, which is not easy. It just doesn't- No ... we didn't, we've been, we've been running a firm in the same way for 20 years, changing that, not only, not, not only the thinking, but the systems underneath it and how the whole firm is built.

You know, I, I run a time and materials-based professional services organization myself. That is not even... Yeah. It's hard, isn't it? Imagine- Very ... and then even if it's like, I had like 200 people, not 6,000. Yeah. So every single measurement, every single KPI was all based on the billable hour, like unwinding all of that, all the incentives.

It's, uh, 'cause, and it drills into people. People think like that 'cause they've been doing it for 20 years. They're just thinking bill, bill, hours, hours, hours, hours, hours because they- It becomes part of their identity. Absolutely right. Yeah. Yeah. And, um, and, and it's, it's, it's really, really difficult to change.

I don't give people a hard time for, for not suddenly turning up. And, and when I see these sort of, you know, everybody sort of bashing big, Big Law, and I go, "Okay." [00:53:00] They're doing all right, aren't they, still? Yeah. You know, and, uh, and there's some really interesting work going on at, uh, firms of all sizes, you know, just- Yeah

yeah. Well, we're, uh, we're out of time. This has been a fantastic conversation. Oh. All of our conversations are. Um, before we go, how do people find out about s- the work you do at Signal? Maybe the first thing you should do is spell it, because it's a little different. And, uh- Yes ... tell us, tell us where they can find out more.

We need to change it at some point. But, but find me on LinkedIn, David Malkinson on LinkedIn. There are two, exactly two David Malkinson's on LinkedIn, and, uh, I'm the one from, from Signal that has lots of posts about legal AI on there, so you'll spot me a mile away. So, uh, so yeah, hit me up on LinkedIn, give me a follow, like and subscribe, whatever you do the, the, these days, and, uh, and if I can be of any help, then give me a shout.

Awesome. Well, David, thank you so much for the time. It's been, uh, it's been a great conversation. It's, it's been, uh, been-- I've long [00:54:00] since, uh, coveted a spot on this, uh, podcast, so thank you, Ted. It's been great. You can check that box, man. We're done. All right. All right. Take care. Good to see you. Bye. Bye-bye.

Thanks for listening to Legal Innovation Spotlight. If you found value in this chat, hit the subscribe button to be notified when we release new episodes. We'd also really appreciate it if you could take a moment to rate us and leave us a review wherever you're listening right now. Your feedback helps us provide you with top-notch content.