Thank you, Adrian.

Thank you for having me along.

Thank you for coming on such a cold night.

Thank you to TechSpark for joining.

Introduction

Mark Riley, I'm the CEO and founder of Matheson AI.

We're a small boutique agency based up in Origin, which is in Barclay Square.

Primarily, we look after media and publishing clients.

We've recently expanded more into legal and risk clients.

Background and Journey into AI

So my background, how did I get into this?

So about 2013, 14, I was in New York working for Dow Jones,

Jones, the Wall Street Journal, headed up the innovation team.

And about 2015, my boss, Will Lewis, asked me to figure out what is AI and what's going on.

So I did an audit of the whole of Dow Jones, discovered pockets of AI.

And then I led the AI adoption across Dow Jones from about 2016 onwards.

and in the process probably met about 110 AI startups in the US and then

vendors that can speak to me and also launched a number of MVPs probably about

250 MVPs sorry minimal viable products for for Dow Jones in that time and

during that time I had to learn a lot about how do you select use cases and

this happens to me a lot in current clients and working for Hearst

From Innovation Labs to Use-Case Triage

magazines now which is cosmo men's health runners world and about six months ago we came up with a

list of 50 potential ai use cases so we had to do a triage exercise and figure out which ones

were most beneficial for her state forward so we're going to try and impart to you in 10 minutes

if i can is the kind of thought process that we go through to try and select the use cases the

most beneficial for you or your business a couple of apologies this section is meant to be futuristic

and societal impact.

I'm very happy to talk about consciousness and sentience and transhumanism

maybe next time, but that's not this talk.

Scope of This Talk

This is much more pragmatic.

The other thing I'm not

going to talk about is the AI bubble.

I appreciate when you're selecting a business or a startup or

use case, you've got to look at the bigger picture, the macro weather.

I could talk to you for half

an hour about the AI bubble, but maybe that's a conversation over a beer.

I'm going to take that

that forward.

Housekeeping and a Challenge

Quick prize for the end of the talk in 10 minutes.

If anyone knows or can guess where

the name Matheson comes from, you get a signed copy of my book.

And the last thing to say is

Setting Up the Innovation Lab

that when I set up this innovation lab, Will Lewis, the CEO, gave me $5 million for a year

and a team of five, six.

I said, what am I going to do?

What use cases do you want me to explore?

And he said, well, just do whatever excites you.

I mean, that was my brief.

And I think I'm going

Personas: Who This Is For

going to give you 10 principles right now but to be honest the most important one is do whatever

excites you so in terms of personas this talk could go in one of four directions as I say if

I'm clever enough to do it on the fly so we've got four use cases sorry four personas here we've got

the 10 crows of vibe coding just building stuff yourself or maybe for something to use at work

you've got the AI wizard who's the internal MLOps dude in the office building code presenting use

use cases for the office and for your colleagues internally.

You've got the AI consultant, which is me,

which is building a product for clients.

And then you've got the AI entrepreneur, the crazy person,

the one that's stupid enough to launch a business

in this environment, of which I'm also kind of there.

So who, quick show of hands, who's a tinkerer?

Cool.

Who's an AI wizard, in -house wizard?

Any consultants?

Of course.

and then any entrepreneurs yay crazy people any or none of the above and he's just ai curious

Show of Hands

Credibility: What We Built

cool so just in terms of credibility matheson we built these tools so as example cases so

we built a self shelf image .co .uk we built this app sits on your phone you take a picture of your

bookcase or your kindle and it'll then come back with a personality insight it tells you what sort

of personality you've got and then gives you further reading recommendations so that's just

purely for fun we built a tool for american institute of physics which talks their entire

archive allows them to surface physics papers relevant to the research you're doing we built

sally which is contract review tool for an insurance company in the docs here and they've

reduced the review time from four hours to 45 minutes then my own startup which i'm happy to

to talk about over a beer, it's called Hannah and it's trying to reinvent local news.

It's

a research tool for doing research and building newsletters.

We've got a Bristol edition and

an Oxford edition.

Market Landscape and Trends

So just to talk about the money first, let's chase the money.

This is

quite recent from a blog called Rule of Thumb.

You can see the spread of use cases.

Most

of these will be familiar.

Some are more robust than others.

Some of them are just GPT wrappers

but it's one to watch the next six months.

But this is the most interesting graph for

for me, so different to the dot -com bubble

where most companies were just useless.

They had no revenue, they just had a dot -com

on the end of their web address.

They had a web address, that's all they needed.

This is phenomenal.

These businesses are making substantial amounts of money.

This is for real.

I think this is out of date already.

I believe that Lovable's on 100 million ARR after a year.

This is phenomenal progress, this is for real.

And then this one's interesting to me.

Revenue Traction vs. the Dot-Com Era

this Harvard Business Review brought this out in the summer, use cases from 2024 to 2025.

2024, a lot of it was just kind of work research, looking up documents, trying to figure out, using it like a bit of a Google.

And then 2025, the trend on the personal use case is much more towards therapy and companionship.

I think that's a really interesting trend

especially amongst Gen Z

are now using

ChatGPT as a friend or as a counsellor

as a therapist

so there's definitely an opportunity space there

Sector Momentum and Where AI Works Now

so just trends across the industry

I won't dwell on this just to say

it's like electricity now basically

AI is no longer to one vertical

it's like everybody is using AI

what's most interesting to me is healthcare

and drug discovery where most of the money is going

for the long term bets

a lot of regulation a lot of caution in that sector so the real impact we're

seeing is in finance especially around things like fraud detection and that

media which is my world was seeing a lot of it for content creation legal

advisory we're seeing a lot of that coming up the inside track now the

startups like Harvey and then robotics is the one to watch in terms of what's

gonna make real progress in the next five to ten years

sales and marketing is interesting sales and marketing was the first early

adopter.

Most of the use cases a year ago or two years ago were chatbots, customer chatbots,

customer service bots.

Very heavily invested in, but that's now kind of hitting maturity already.

The more interesting spaces for me are healthcare and media.

And I'll keep coming back to this

theme, but play to your strengths and stay in your lane and use what domain expertise you've got.

Ten Principles for Selecting AI Use Cases

1. Start with Customer Problems

So principle number one, 1start with the customer problem, not your skills.

identify a problem that you can solve for your colleagues or your clients forensic analysis of

a workflow going to deep deep research into how the workflow works how people's days are spent

and then figure out where the friction points are all in AI startup is looking at a task and turning

it into a prompt and then getting that prompt to work better and better against evals so if you

If you can work out the pain points, you can solve for your customer.

And boring is beautiful in this world.

The more boring a task, the more likely they're going to come to you to find an automation.

2. Play to Your Domain and Data Strengths

Play to your domain and your data strength.

So if you're very lucky, you've got a company which has proprietary data.

This is very true of healthcare and defence.

If you haven't got proprietary data, you can go and find user data.

It's what we call data exhaust.

Everyone's phone is putting out data exhaust.

Also, Duolingo is using user data from their platform.

Strava is using running data from their platform.

Character AI is learning about how people interact with these bots.

If you don't have that, then you're going to have to go to public,

and you can beg, borrow, or steal.

Perplex has just went out and stole as much as they possibly could

and put it into their cache in their servers.

You can go and beg it off climate companies.

You can look for it in the council website.

There's plenty of places you can go and get public data.

data.

3. Stay in Your Lane: Pair AI with Expertise

Again, stay in your lane.

The more knowledge you have of these domain sets, the more likely

you'll succeed.

AI really, really works well when it's coupled up with domain expertise.

The best success we had at the Wall Street Journal was when we sat the data scientists

down with the journalists and actually learnt what their workflows were and the marketing

teams and they could get synergies that way.

4. Build for Two Years Out

Build for two years out.

If you think the

models are as good as they're going to get now and you build a solution that's as good

as GPT or Gemini is going to be now, you're going to be blown out the water in two years.

So you're going to need to plan for where these models are going to be.

You need to

build features that are accommodating for increased intelligence, probably means more

reasoning and more memory, more capability and longer context windows.

Gemini 3, I mean,

frankly is it going to kill Cursor?

Potentially.

Bizarrely, Cursor is a customer of Anthropic

and ChatGPT and the rest.

So these companies, these frontier models, these hyperscalers

are actually killing their customers

in the process of getting smarter.

Does anyone use Fixer?

British startup.

Very cool.

Annotates your emails,

tells you which ones to ignore,

which ones are marketing,

which ones to respond to.

Gemini 3 does that.

I mean, poor old Fixer.

They raise tens of millions.

Claude Codan in Obsidian does that.

Right.

So there you go.

So, quote from our lord and master, Sam Altman.

I think there are two fundamental strategies.

is you either bet the technology is about as good as it's going to be,

or you bet the technology is going to get massively better.

And I'll share the date with you so you can watch this YouTube video.

5. Navigate Legal and Compliance

Navigate the legal and compliance frameworks.

Are you up to date with the EU AI Act?

Does your tool or your solution or your idea or your product,

does it reflect the framework that your company is operating around?

How do you mitigate hallucinations?

Very important to my media clients.

The editors hate it if bots lie on their websites.

It goes against their religion.

So you need to have a human in the loop to mitigate against any kind of hallucination

problems.

I'm going really fast, sorry, I can't keep up.

Five minutes.

6. Rethink Your Revenue Model

Rethink your revenue model.

SAS.

Good old SAS.

Companies like Box, Dropbox, they had a 90 % margin and zero cost of sales every time they

created a new customer.

Fantastic business models.

AI is not SAS.

It takes time to train, it takes compute.

you.

A good margin on an AI SaaS business is 40, 30, 40, maybe if you're lucky, 50%.

And the reason is you're paying for the inference as you scale, your inference costs are going

to go up.

It's a very, very different business model, which impacts on pricing, which is

a fascinating conversation, how these models are going to charge out their usage.

Most

of the vibe coding tools are operating at a loss because their VC money is compensated

for the inference costs that they're charging the customer so how do we price this you can no longer

do it on a seat model do you do it on a on a time model do you just charge a flat rate for a year

or do you build an inference cost you pass that cost on to the customer uh she's hilarious by the

way if you want uh i'll send you the debt she really slags off the ai business okay moat uh

7–8. Build a Moat and Don’t Give Up at 60%

i'll only do talk about two of these so niche is lovely if you've got a niche and you're you're an

an expert in a niche, as the Americans say, just go deep, not broad.

Don't try and come

up with a new browser.

Just go very, very deep on one solution, transcribing meeting

calls, for example.

And then I think this one's kind of super relevant.

People give

up on a lot of AI startups because it doesn't get better than humans because they can't

be bothered to fine -tune their prompts against evals.

When it's 60%, they'll give up, so

it's not as good as the human.

You've got to work night and day for months to get your

prompts to behave well enough so the model knows what you're trying to do, so you start

hitting evals of 80 % or 90%.

Then you've got a business that you can cash out for millions.

9. UI Is as Crucial as AI

UI is as crucial as AI.

It's a lovely UI.

Slagged off perplexity.

It's a thing of beauty.

they are product people.

Gamma that I'm using for this deck is a beautiful, beautiful UI.

Moving on.

10. Define ROI Early

This is kind of the meat of the talk.

How the hell do you define ROI?

It could be many

things.

It could be increased engagement.

It could be increased subscription.

It could be

increased retention.

Or it could just be money.

It just could be cash.

Define your ROI.

Or it could

just be an efficiency.

It's time saved.

Can we reduce headcount?

Are we more productive?

Well,

I'm telling my clients, please, stop thinking about AI as a purely efficiency tool.

Please

start thinking of it as a new revenue tool.

If AI is just efficiency, half of us are going

to be unemployed in five years.

It can't just be efficiency.

It's got to be new businesses

as well.

And it is super cheap now to start these startups.

It's a very elementary tech

stack.

So it's easier to get in the game, which means it's harder to stand out.

So you're

going to have to, as Glenn said, you've got to be cracking at your go -to marketing.

And my rule

of thumb is spend as much on marketing as you do on dev.

Frameworks and Tooling

The Three Boxes of Innovation

Three boxes of innovation is a nice tool.

Box one is just an incremental fix.

Box two is a new product.

And box three is what I call the

WTF box, something completely out of your comfort zone.

You could think of it as box one as assists

a human, box two replaces a human, and box three is superhuman.

So you've got GPTs here, you've got

fine -tuning and personal tutors.

You've got robotics and Neuralink over there.

I try and

get my clients to come over here, get really ambitious, and then I put them back into box

two, which is where they should be placing their bets.

Build, Buy, or Borrow

Build by Borrow, there's an AI

for that.

It's a fantastic website.

It's got thousands and thousands of, well, 70 ,000 to

be precise, AI tools in there that you can do some research, see if anyone's done it

it before.

Chances are someone has done it before, so building it from scratch is not

always the best idea.

Reality Check: Impact on Jobs

And then I don't know whether to use this slide, and I didn't really

want to leave on a downer.

But it's an important point.

It is the elephant in the room, and

it is something we need to be conscious of.

If you're reasonably successful, you're going

to make a tool that makes everyone's life better.

Their job's going to become more fulfilling,

you're taking away all the tedium they can do more interesting work that's what

we're being sold if you're really successful you're going to put a bunch

of people out of work I'm not going to sugarcoat it now if you're in this

business you're going to have to front this and there are lots of mitigating

factors if lawyers suddenly get cheaper law is democratized more people can

access highly specialized professional legal services however in the meantime a

a lot of lawyers might get put out of work,

say with accountants and consultants.

So it is worth reflecting that if you're in this game,

who are you going to impact?

If you're very successful,

you've got to impact competitors or your colleagues,

and what are your mitigating thoughts

and how you're going to present that?

And there are loads of defences against it,

which are well rehearsed,

but it's just something to flag at this point.

Conclusion and Next Steps

Ikigai: Choosing What to Build

Finally, Ikigai.

I don't know how many of you know Ikigai.

If you still don't know what to do,

this is the best place to fall back on.

So what you love, what the world needs, what you can be paid for,

and what you're good at, and you find your icky guy.

What will you build?

That's it.

Thank you.

Plugs and Community

Some quick plugs.

I've set up a supper club for AI practitioners in Bristol.

You have to be a practitioner.

No recruiters allowed.

And we meet once a month in a pub in Clifton and talk AI philosophy.

I've got a newsletter called Medium Wolf,

and I've got a book on Amazon called The AI Profit.

and the name does anyone guess that's where it came from thank you