The Cost Curve: Building AI Value Without Chasing Frontier Models with Amin Mrini, Informa

Good morning and hello everybody and welcome back to another Able Leadership Series interview. I am welcomed here today by Amin Reinie who is the chief commercial AI officer from Informer, one of the world's largest events and media businesses. Thank you very much for joining me this morning Amin.

Thanks for having me John. So I'm going to slip straight into some questions. Most businesses globally now are using AI and they're using it to do the same things that they've always done but to try and do it in a faster or cheaper way.

You've argued previously the real opportunities in what becomes possible when the cost of execution collapses. What do you mean for that and and what does that mean and open up for kind of mid-market and other businesses? Yeah.

So I think the this idea of doing things for cheaper is incredibly short-termist and sort of mono dimensional as a way of approaching or or or trying to unlock AI value. I think AI uh is is a multiplying force for productivity and output per person in ways that helps organizations thinks about new forms of demand, new products and services. Um, and the goal should be dollars of revenue, AR, whatever the case might be, per head as opposed to trying to minimize the head per dollar profit uh, type equation.

I mean, obviously a cost impact is a one-off by definition, whereas a new product compounds over time. Um, and for those who are thinking of AI as only as a cost-saving lever, although I'm pretty sure we'll have to go through a phase where that is primarily the benefit because it's easier to measure. I'm pretty sure that in any market anywhere there's either a bunch of AI native companies or legacy providers who have a better up approach to AI who are trying to unlock its value in the ways that I would describe it.

And so ultimately being protective and thinking about cost only is a defense strategy that will ultimately affect your market share and your growth potential long term. Someone out there is trying to eat your lunch and he's thinking new products, new services, new demand, new ways of distributing content, information, products, you name it. And so that's something that you you have to be doing as well.

I I think it's a really interesting point. I mean we obviously speak to a load of people for different types of businesses in in a range of different sectors and there really is two distinct schools of thought here. there's what you've just described which is where's the opportunity?

how can we how can we use this disruptive new tech to to really go on and grow and scale at multiples that that would have been really difficult in the past and then the other school of thought is the one that you're kind of arguing against which is people going well okay how can I save some money you know how can I trim headc count how can I improve process whatever it might be and I think there are merits in both of those arguments but certainly if you have got a growth mindset then you surely are wanting to to to run at the opportunity that this has that have you got any examples of cool stuff you've seen people doing in that kind that field. Um I mean there's there's a very very long list. Thankfully the the cost saving and frankly I think the AI washing has dominated the media narrative.

So you hear a lot more around job ways primarily in big tech big tech which I think are more of an of an correction over hiring than than actual AI impact. But no, there's there's there's especially when you talk about mid-market, there's a there's plenty of small to mediumsiz companies who are uh if not AI native, AI nimble are thinking about talent and are thinking about developing the existing staff in exactly those ways of how do I increase output per person and so there's in functions like sales in functions like marketing primarily I would say there's a pretty clear blueprint starting to appear out there in terms of what does the the AI stack or the new AI enabled ways of working and what these look like. Yeah, I agree.

I mean, there's there's a lot of people out there who are trying really hard to create a really efficient and effective AI stack. And then there's load of other businesses out there who are just bolting any AI tool they can find onto existing workflows. What we find is that just pushes the problems somewhere else or it replicates the same problems that businesses already had.

Do you have you in your experience taken the time to take that step back and redesign the flow around the new tech? And is it is that something you see people doing or do you see it as just you know there's let's define a problem, let's find a tool that will solve it. What what do you think is the best advice you can give to people?

Yeah, I mean it's what half of my job consists of. Yes, I have. So I think we need to first of all we need to ban this AI use case term that people hear to hear everywhere.

What what's what's the use case here? So the taskbased approach of hey let's brainstorm a 100 different ideas and then build discrete AI tools which are mostly probably chat bots and then hope for the best and hope that we are more productive or I can take out 10 15% of cost. It just it just doesn't work right.

So a simple example is okay I can create presentations more quickly but if the decision and the meeting during which it takes place is still a monthly meeting like you haven't saved time like you've just given you've just given people more time to either over overp polish an asset or or just do do other things that aren't necessarily going to be the the the things unlocking value. So this idea of the taskbased let's automate these tasks or let's find the AI use cases or frankly thinking that giving a clue to everyone is a strategy for AI isn't going to to to to anywhere to anywhere. You might be in very very tiny pockets making people slightly more efficient at one thing but then the value is only unlocked if you start chaining tasks which is why we start talking about workflows.

And then the reinvention piece to me comes from looking at the problem. the opposite way. So in the past companies have started thinking what can I help humans do more quickly or more effectively or more cheaply whatever the case might be.

The question that I ask is what do humans absolutely need to stay in charge of and then once you protect that or you take that out you you delegate and find ways of giving the rest to AI. What in this chain of tasks or sequences needs to be a human decision, a human approval or or needs to stay in the hand of someone to create an output and then handing that over and everything else to the to the extent that it makes sense for the workflow in question. Obviously there are accuracy problems, integration problems, compliance uh problems and so on so forth but then the rest to some degree can be will be handed to uh to to an agent to perform these tasks.

So look at the problem the other way around and simply thinking how do I speed up outputs in the way things were done before with the humans that were in charge of that task before is not where AI value is at all and and what would your advice be to people who are trying to decide where humans need to stay in this loop. So this is where this is where AI fluency comes in. So there's a there's a base layer of judgment and just knowing your processes, knowing the level of risk you're willing to accept, knowing where the uh approval gate is or what your governance sort of framework dictates number one.

And then there's not necessarily an intimate but just good enough understanding of what AI is good at and the extent to which it can perform the task accurately or not. And so then if you if you understand what BLM the limits of reasoning, the limits of accuracy, and then the what's good enough equation, you you're you're willing to accept if you're if you're I don't know if you're reconciling invoices, right? Where does where does the needle shift in terms of productivity?

If you're able to to do the the 20% of the hardest cases that are now being manually reviewed, 50 80% and then if you're left with just 20%, what does your team look like after that? So it's a combination of understanding the process, the needs for accuracy, risk profile of delegating to an agent and then a good understanding of what AI does well reliably or not comes from experimentation. It's not it's not something that you either have or do not have.

Just try it changes all the time, right? Because I mean even in the past 48 hours, Claude has released a new model which is slightly better than Opus 4. 8 eight was is Opus 45 seems to be a lot better at certain things and so it is it's constantly reviewing what the ability of whatever AI model whatever AI tool you're using has got as well what we talk about is we encourage people to use AI to do things that are repetitive administrative that don't require strategic thought creative thought or relationships because our belief and one of the things that we say all the time is that by using the tools that you now have at your disposal it allows the humans to do things they're really good at that machines aren't so good at moment.

I know for us comes back to strategy, relationships, creativity and social socializing is the other one, but that's kind of got a less business focus to it. Is that something you you would agree with? Yeah, I mean look on on your first point.

First of all, I think people shouldn't I don't think people should panic uh or read too much into the four to six week new model waves. saying they should look at capability improvement over a slightly larger window over say a year, 6 months to a year, not every cycle because I bet you that the the absolute vast majority unless you are in healthcare looking at uh drug pipeline, unless you are in advanced engineering, unless you're Airbus, I guarantee you that the absolute vast majority of your processes do not require frontier models. And so first of all that comes with a cost advantage and it also means that you don't need to worry about keeping up with the frontier and pay for state-of-the-art.

So models are increasingly are are I mean look it's looking like an exponential curve. Uh uh it's it's it's it's confirming to look like an exponential curve. Um but on a day-to-day basis for most of us and for most workflows you're you're covered.

I would spend more time thinking about how I write the cost curve down than how I ride the capability curve up uh for a given for a given workflow. That's the first point. And then and then a good idea to start with what is repetitive codified has got the right context layer around it to make sure that you're making the right decisions or uh or performing tasks in the way that you want as it relates to strategy and creativity.

is I mean obviously creativity is intrinsically a human uh attribute so I don't know that it's even the right word but when it comes to strategy as a handsoff tell me what I need to do probably not but as a as a as an accelerant for humans same thing for ideiation brainstorming prototyping so it's not I wouldn't call that creativity but not far from it increasingly I find AI to be very very capable and there's and there's more and more things where I start with an idea and then I'm given the utmost sophisticated hyper refined version of the idea that it would probably have taken me a while to maybe not even find. So on those what is belongs to humans how long does that remain and to what extent I think ultimately it's not a AI non AI it's a can I comfortably outsource to AI and then there's something that requires an AI and human pairing more so than not AAI don't think there are many not AI fields apart from obviously well human contact relationships and so on so forth I like I like the way you've described it as an accelerant um that's certainly something that I do myself personally to use it to either ear off ground or to refine an idea and and move it through my thought process quicker than I would have normally done before. So, I quite like I like that description.

It's really good. A lot of there's a lot of noise out there about how do people measure the return on value of any any AI tools they're buying, right? And there's loads of stats out there from Deote and Forester and McKenzie and various other places about, you know, 94% of businesses aren't seeing any any return on value.

You've said previously that that that without measuring the baseline first, you'll never be able to prove that AI has delivered an upside. So for a thinking about kind of the smaller businesses here who may not have a dedicated data team or or a team that are in house to to do that kind of measurement, what would your advice be to them? Yeah, I mean look, it's not in a way like it's just paou, right?

I mean when you're launching a new product, you're you're accountable for how it performs or how it changes the customer journey, satisfaction, financials, you name it. So there's nothing new there. You need to have an ability to attribute a success metric and to track how KPI X Y and Zed evolve with the introduction of AI inside of a process.

So you're say you're you're uh you're you're you're automating content creation right for a certain flow for a certain persona you need to measure throughput you need to measure content interactions views clicks time on site repeat visits and so on so forth right so the baseline thing is make sure that you start with a clear understanding of where your KPIs are because otherwise how would you ever see an uplift if you don't know what the what the status quo looks like. So it's not there's nothing more sophisticated than what anyone introducing a new product, introducing a new service has had to do over the past wide number of years. Yes, there are attribution is is is difficult in some cases.

The there's there's a there's the causation versus correlation problem. In my experience, it's pretty clear whether it's working or not. you can't do without the numbers, but I haven't had a hard time getting comfortable with the numbers even when it's a partial view or even when you can't fully close the attribution loop because done well.

We're talking about I mean there's nothing I've done that was such a tiny incremental gain that we weren't quite sure it was it was worth it. Like there's so much to be redesigned, so much to be rethought. If you're starting with the things that are driving the extra 0.

1% and that you're not quite sure is within the margin of error, I would argue you're probably not starting at the right place. So it's not as complicated I would think I would presume than what you could make it sound like. Yeah, I agree with that.

And and and again I like what you said there. I mean if you bring any new tool, system process in you need to measure it. This is no different.

Just because it's more technical in nature doesn't mean mean it has to be hard. No value debate uh comes from uh an idea that is still very present which is yeah I have an AI strategy I've given everyone uh GBT or CL and so you're like okay so okay so everyone's burning tokens if you just say everyone's got access is a free-for-all then by definition you're not quite sure what you're trying to achieve what are you trying to achieve if your strategy is okay everyone be more productive here you go here's Here's a here's a chunk bot. Well, what are you trying to shift everything all at once or some intangible form of we should be more productive or we could do more with less or we could so that is impossible to right.

So the strategy of I I want to rethink this workflow and I want to increase the throughput of my sales team and bring the cost uh the conversion uh down. Uh I want to bring conversion up. I want to drive pipeline growth of x%.

I want to be able to manage a book of business with 20% less sales reps and then the other reps will be doing something some something else. That is something that you can measure. So people have approached it from a general purpose.

Let's have fun and let's trust people to what to sit there and just just reimagine their jobs by themselves as they do it. So it can't be it can't be everyone's responsibility to rethink. Now we everyone now says reimagine the workflow reimagine the workflow.

It can't be an individual contributor's responsibility day in day out. No one wakes up and thinks how do I how do I make do I rethink my job. So it's it's the approach and how AI was deployed that is creating this uh fake debate around AI value and also because people for some reason thought that burning tokens was was a sign of pride or something to that equates um AI fluency or literacy or just just just being so um the the the way of unlocking a value is clear.

Now I and I I hear the recipe is just so obvious it almost pains me to have to say it. But I still hear on some pretty big podcast out there saying now uh genius stroke of genius very small teams engineering capability sitting down with the functions and then embedding embedding AI behind the scenes inside of operations. So, so you bypass the whole adoption challenge, you bypass the whole change management challenge and you have uh uh and you preserve your ability to chain and and and and transform whole sequence it.

So you need you need you need technical fluency. It can't be everyone's responsibility. You need to have people accountable for that change.

You need to be have people embedded within functions for that change. That's why we have the whole FD praise and mania right now and all the deploy codes and consultancies rebranding themselves. So there's a clear way of unlocking value.

The debate just stems from a complete lack of strategy as to how AI was deployed. Cuz that that's really interesting. It's something that we've spoken about a lot with some other people and there are a variety of different ways that people are deploying it.

Sometimes it sits with the the CTO, sometimes it sits with the COO, sometimes it sits with the CHRO treating more as a a an HR kind of adoption challenge. And I think there's probably merits in all of those. and having accountability important when you're trying to do anything in business that that's an absolute given.

But your suggestion there that you break it down into segments um again very sensible in most most most schools of thought and then trying to go into the teams understand their challenges and then embed it on a piece by piece method seems very logical to me. Yeah. And so whoever it sits with it needs to come with the realization that really can't sit with one function.

Right. So technology is responsible for technology. So at the moment an AI team sits in technology and is responsible for reimagining processes and rewiring processes like technology is not responsible for the entire process wiring across functions in the business nor is HR nor is anyone.

So first of all the mandate the story told around AI the investment behind AI and what is trying what the company's trying to unlock the AI needs to come from the CEO. That's number one. And then number two, yes, the EI function is to sit somewhere.

Um, unless unless it's a direct CEO report, it it tends to be in technology, but it needs to come with the realization that okay, it's in technology because the engineering skill set, the product skill set, delivery skill set tends to be there. Fine. Um, but it needs to come with the realization that every function is responsible for the outcomes that AI AI engineering or re-engineering uh is unlocking and that uh that function is a partner.

it doesn't own the entirety of the the process business process outcomes across functions. It just can't work like that. And so there's a there's a level of AI fluency required in nontechnical functions.

There's a level of AI engineering capability uh that is required outside of technical technical functions as well. So there's a bit of a rewiring of the organization if not in the orchard at least mentally that's needed. Yeah.

Yeah. I I couldn't agree with you more and and that you you've mentioned it a few times, the idea of AI fluency across the board is is so important and it's how do you how do you enable people? How do you help people to become AI fluent?

And that comes back to the way you adopt it, training and development. Same stuff as always. You've mentioned previously a couple more questions.

You mentioned previously about AI squeezing the capable generalist. What do you mean by that? I think it was the pseudo generalist.

Um um uh I think I I think there's a there's a profile in quite common in in in most organizations of someone who's in charge of coordinating and managing without owning a domain expertise. Um so when the cost of execution was high and when domain expertise was quite siloed you needed the manager someone sitting at top coordinating scheduling meetings reporting making sure that the handover between those different silos was managed effectively. Those silos are getting closer to one another and in in some cases they are merging and execution isn't scarce anymore.

It doesn't need to be micromanaged as the rare occurrence in the business of it's so hard to make things happen that I need three layers of management to make sure that people know what to do and are doing it well. Those without a really deep domain expertise and those without the judgment to be proficient across domains to me are are going to lose value. And so either you're a hyper hyper deep domain expert who is able to challenge AI output to refine AI output and to see mistakes or inaccuracies or to help advance AI output or you someone who's able to to bring down the frontiers between functions.

So a product manager who's flu who's technically fluent enough to be able to prototype to be able to look at design to be able to look at research to be able to look at launch planning for example why would you need a UX designer a product designer an engineer a sales enablement specialist and a marketing manager today to launch to launch an AI to I'm pretty sure I mean there are I mean that's how I've built my team so you could you could compress those five or six roles into into two or three if if not more. So I think we're entering the age of the the hybrids. I think AI augments those exactly those profiles and those are the people who will thrive.

Those who have combine one deep technical domain expertise with the ability in the synthesis in the synthesis to operate across domains. And so by extension those whose those who aren't particularly hands-on first of all those who aren't individual contributors and those who are just the lubricant between layers of an organization that is collapsing that will flatten out over time will lose relevance. So I mean I've heard this I can't remember what what it was about 101 15 years ago um I was working in the sector and and people were talking about the demise of middle management which is similar to what you're talking about here that lubricant layer and and I think you could be right this time I would imagine so you've done this so great question to ask you I imagine hiring genuine hybrids at this moment is is quite hard is that right or am I wrong about that no it's very hard it's very hard I had the uh I wish I had the recipe and uh be more successful um at it.

Look, the the good news is we're about to find out, right? You have companies like block for example who have decided that they could do the same job with 40% less people or middle managers contacts and information layer entirely managed with AI which is effectively is the equivalent of replacing that uh coord that layer of coordination of meetings of information gathering and information sharing. So and they're not the only ones.

So we will find out pretty quickly whether the thesis is right or not. So both things can be right at once, right? We could see a a rise of the new hybrids without losing those u those middle management positions for well we'll find out.

But to come back to your question around I mean there's nothing unfortunately there's nothing new there and and importantly in terms of hiring good people hybrids there's already there's always been hybrids out there augmented hybrids is a much better hybrid hence the point u that I'm making having one having one helps you find the second and then having two helps you find the third and so on so forth the profile so the the the profile of the hiring team to me is the biggest differentiator in terms of uh finding the next best person for that and that fits the bill because people recognize people with that skill set want to work with people that that have it recognize it value it obviously it doesn't solve the so and and then N plus one fine but it doesn't solve the first in most cases you won't find uh those people and if you find them they'll be they're quite rare so ultimately it's a lot more of an upskitting training how do I uh uh assess performance as it relates to AI usage and how do I upskill my own team on AI usage to transform your people into these hybrids? Like just you won't just start clean slate and find a new full team of hybrids out there. If you find a small number of them, use them to upskill your team.

And ultimately, I'm pretty sure that upskilling is more is is more important than hiring just purely based on volume and time that it takes to to to churn uh and then replace. So I don't I don't have a a magic recipe the the ways to interview the channels for hiring that work well of mouth being constantly out there interviewing meeting people whether you have open positions or not to build a pipeline because you never bump into the right people when you need them unfortunately and then trusting then once you're able to create that culture and bring that skill set great attracts great yeah great advice So look, we're out of time, but I'm going to ask you the final question I ask everyone on these, which is there's someone watching this, they are at the start of their AI adoption journey. They are a senior leader in a in a mid-market organization.

They're not as advanced as you. Um, what would be your single piece of advice to them about how they how they take on this opportunity and challenge? I mean, it's hard for me to pick one.

So, start with simple but transformational goals, right? So there there's there's a lot if you're starting from scratch by definition there is a lot of lowhanging fruits that can transform productivity in terms of either your commercial performance, your product performance, you name it. But simple, clearly stated and measurable goals.

You won't transform the organization in one go. And again, giving people individual tools isn't going to achieve that. So do not underestimate the engineering challenge that that represents.

invest in that technical capability and then embrace that uh which is now commonly referred to as the FTE model which is embedding a very small number of capable engineering and product talent within functions and then drive the change from within that function and think about AI behind the scenes. The less AI you see the better for productivity. You're not building tools for the masses and then everyone spend their day on four or five for every task of course calls for a new tool type thing.

So in terms of ways of thinking, how to approach, how to deconstruct the problem and then the right model for unlocking AI value, those would be the the main advice. Wonderful. Well, look, thank you so much for your time today.

I mean, we really appreciate it and wishing you and inform the best of luck as you use AI to forward your business even more over the coming months and years. My pleasure. Thank you very much.

Thank you.