Marketing and the AI Maze: How to avoid failure

Part One - The challenge

It’s hard to avoid the feeling that across both B2B and B2C marketing we are still in a period of uncertainty and caution. Efficiency, demonstrating business value … these are the topics that consistently top CMO surveys - especially in Europe and the UK.

But amidst the dreary headlines, and in contrast to almost all other areas of marketing, serious money is being plowed into AI, and I can’t help but feel maybe not all of it is being spent wisely …

This isn’t a bah humbug attack on the technology that underpins Gen AI. We should start by acknowledging that it looks set to have a disproportionate impact on our humble profession - see Figure One below.

And this feels different. I never could get a concise elevator pitch for how Web 3.0 would transform Marketing. With Gen AI it’s different. The promise of click-of-a-button asset trans-creation, immediate and perfect translation, the ability to check your assets for brand consistency with a trained AI ‘brain’ in real-time and without any need for time consuming governance or going to central Brand team, even original video and image generation, it’s all on the table.

In fact if you work in the marketing team of a large organisation it’s likely today that to design, produce and launch a global multi-market brand campaign it will take roughly 9-15 months. Just with some of the AI enabled changes I mentioned above, it’s easy to see that cut to a few weeks. Maybe days.

So, unlike with previous technology-push hype bubbles that offered complex technology solutions to next non-problems – yes I’m looking at you blockchain and metaverse – Gen and predictive AI will shake up the marketing value chain. Anyone can today tailor a Gen AI model with their data and start seeing results. Take a look at Amazon Bedrock if you haven’t already.

And yes, we have clearly over-hyped what it can do right now, but nevertheless this is revolution not evolution. Gen AI is here and ready to shake things up. It already is. However, the speed with which Gen AI landed, the subsequent hype bubble, combined with it’s obvious potential to upend many traditional marketing value chains, leaves many marketing leaders we’ve spoken to feeling like they’re standing at the entrance of a maze. At the centre is the promise of unimaginable riches. However, the routes are seemingly endless and multiplying by the minute.

The big question then is how do we navigate effectively? This article is going to try and give some answers. In short, we want to share some insights around the question of AI investment for Marketing leaders. To (hopefully) give some tools for working through how to identify the Gen AI assets that will actually deliver profitable results for your organisation, and to deploy them in a way that delivers value. In other words, of all the shiny AI things we could go after, how do we rapidly and efficiently spot the very few that will translate outlay into revenue &/or margin?

The good news is that we’ve just described a problem old as corporations themselves. We find ourselves faced, once again, with the Innovators Dilemma. Do nothing and risk being slowly drowned as competitors bring more powerful AI products to bear, or do too much and risk a huge hole in the balance sheet with nothing to show for it and, at worst, an extremely uncomfortable chat with the CFO.

The Challenge - Adding some colour & numbers

Even as the Gen AI Hype bubble is (finally!) showing signs of deflation, real money has been committed and continues to flood in at a mind-blowing pace - 62% of Global Business technology professionals say that their organization plans to significantly increase AI investment over the next year[1]. Double-click into marketing and the numbers only go up - 72% of marketing leaders are actively prioritising spending on AI over the next 12 months, with over 67% putting up to 30% of their marketing budget towards the technology.

To bring that to life a bit, let’s try and translate those percentages into hard cash. First, lets be conservative and say that that ‘30% of budget’ ends up being a more muted 10%. Next, let’s isolate just the US market. With our parameters set, we can factor in our inputs as follows:

An average of 7.1% of revenue was spent on marketing activity by US organisations with Revenue of over $10Bn[2] – which is essentially the top 300 revenue generating businesses in the US ( the top 304 to be exact).
We know that the top 300 US companies[3] generated revenue of around $12.23Tn last financial year

The result of ($12.23Tn X 7.1%) X 10% - We’re looking at a very conservative estimate of $8.5Bn being spent on AI over the next 12m by the top 300 US marketing departments alone. It could easily be more than double that.

That’s a lot of cash and, as the AI hype clouds part to reveal the fog of doubt and misty drizzle of practical issues - like Gen AI habitually making things up, legal uncertainty etc - it’ll be extremely easy to either waste the investment, or worse, not make it at all. And so we arrive back at the same big question set out before – i.e. how can I make sure I get the best value from my (very large) investment?

But this isn’t a simple ROI problem. AI for Marketing is, at best, a wide fuzzy and rapidly shifting landscape. How Marketing leaders generate ROI by adopting the right technology and adapting to make it work is an innovation problem. This is important to point out, because innovation is hard, as we’ll show in a bit.

Part Two - Our solution in brief:

I’m finding some success in answering ‘the big question’, both with clients and within our own organisation, by applying our Lean-Agile Innovation Framework to AI. Yes, I know I know, that was a seriously jargon-heavy sentence. So, to combat the consulting bullsh*t language I’m going to try and build a good mental picture of just what that means.

To do this, first imagine a funnel. My grandfather had a big old stamped-metal funnel covered with peeling gloss-red paint in his tool-shed for filling the lawnmower. That’s the kind of funnel I’m imagining.

Now, at the top of this funnel, picture the hundreds of AI tools and potential applications that your organisation could use pouring in. At the bottom, we want to see the two or three that are going to deliver business value magically flowing out. To make this work, between the top and bottom there are going to have to be some filters to control the flow, and to make sure only the best stuff is making it through.

So far so good, but filters alone are not enough. The image you’ve got in your head is useful, but it’s static, and innovation, including AI innovation is not static. For the funnel to work it needs to come to life as a living, working and evolving system. To do this it needs a way of operating – this is to say that it needs to be configured with a mix of people, processes and technology that collectively make the whole thing work.

In my case, I’ve been using the principles, methods and tools from the domains of Lean Manufacturing, Agile Software Development and Product / Service Design to make the funnel run. To make the system work.

Okay, so that’s it. That’s our summary paragraph / mental picture we want to get across. However, we wanted to also give you some more insight into the framework and, unlike most fluffy consulting articles / blog posts we read, actually give real practical tips we’ve found make this all work. If you don’t have the time, then stop reading here. If you want to learn more, the rest of the paper will un-pick the above paragraph in increasing detail.

Part Three - De-Jargoning:

At the start of part two I lobbed in the terms Lean and Agile. It’s worth explaining what they mean and why they matter.

What’s an innovation funnel?

In industry, it takes about 3,000 raw ideas to produce one successful commercial product (fig 2). For some categories it’s way more – in pharma only 1 in 5,000 compounds makes it to the pharmacist’s shelf, of which only 1/3 will recoup their R&D costs. To find that 1 in 3,000[1] diamond we essentially have two choices.

Choice one - take an educated punt. You probably have the bandwidth to pick 10 ideas, so get your people in the room and pick the 10 you like the best. The problem, as every VC investor I’ve talked with has told me, is that it’s almost impossible to spot a good idea from the outset, and there are always far more ideas than resource to execute them. It’s why VC’s spend as much time evaluating the team as they do the idea.
Choice two – build a disciplined but fast and effective system for systematically sifting through the idea-mountain. This is the innovation funnel. It’s a system – i.e. a value chain that combines people, processes, tech to consistently find and scale diamonds fast.

What about Lean and Agile?

Lean is a body of work that sought to codify the success of Toyota in a way that other manufacturers and industries could pick it up and try to copy. It can trace its intellectual roots back to probably the most famous innovation organisations ever created – Bell labs - and its most famous globetrotting alumni Dr Edwards Demming. There are numerous books on Lean. If you’re interested in learning more I recommend the short and highly effective This is Lean: Resolving the Efficiency Paradox. For now, sufficient to say it took the resource utilisation focused models that emerged in the West at the start of the 1900’s and completely ignored them.

Agile software Development – to use its full name – originated when a group of software engineers, who were strongly influenced by Lean, got so fed up with the horrifically inefficient and ineffective way software was built that they published a four-point manifesto + twelve supporting principles. These have since been translated into numerous methodologies and supporting techniques. You can see them here if you haven’t already, and you’ll likely immediately spot their applicability outside the world of software - https://agilemanifesto.org/. Like Lean, Agile looked at the way things were being done and completely ignored them.

Design thinking came about at a similar time to Agile software development. Tim Brown of Ideo fame was fed-up of the linear and bureaucratic processes often applied to product development. Design thinking was the solution.

We’ve found these bodies of work to be extremely compatible. Not only this, but the methodologies, techniques and tools that they have evolved can be extremely powerful when applied to the innovation problem. Remember we said that the innovation funnel is really more a system than a model? Well we built ours using scaffolding provided by Lean, Agile and Design Thinking.

We also learned that meticulous co-ordination, empiricism, and alignment around clear strategic objectives are critical – we really don’t recommend you setup and ‘innovation lab’ and treat it as some pirates and mavericks playing with shiny toys. Innovation requires the same clear-eyed discipline as any business initiative.

We’re hit to impact the top-line, bottom line or both. We’re not here to sit on bean bags and demo some really cool but utterly unrelated or impractical technology – “here’s our latest augmented reality app that lets you scan a building material and match it to our product catalogue”, yeah really cool, but have you met many tradesmen who have ever used Augment Reality and who don’t know what types of brick they need? Yes, that’s a real example …

So, what’s our solution then?

Part Four - Our AI Innovation Process in more detail:

**Making innovation work:**

A Disciplined, Creative and Iterative Process
Supported by cross functional teams
Measured on business value

We setup a value chain that runs across six stages …

1. Meticulous alignment

2. Understanding the Needs to be Met - Gemba / Empathise

3. Horizon Scanning

4. Ruthless Prioiritisation

5. Rapid Iteration

6. Business integration for scale out

We embed the right people – Cross functional teams that include…

· Strategic Analysts – the people who focus on the money and business case

· SME’s – the people who focus on deeply understanding the AI tools

· Researchers – the people who understand how to test the tools in a scientific way

· Service Designers – the people who understand how to adapt tools into the parent organisation

· Change Managers – the people who understand how to get the parent organisation ready for scale-our

· Decision Makers – the senior powerful coalition empowered to make decisions fast and derive scaling across their organisation

We drive a culture that focuses on:

Results over theatre – This is about generating business value, not having a shiny ‘Innovation lab’ that impresses clients, but that struggles to generate return.
Science over Emotion – We believe in emotion, but we are amazed at how often companies that promote iteration and experimentation don’t deploy expertise in how to setup and run experiments. An experiment to us means what it does at CERN, it’s about rapidly testing a hypothesis and using data to determine validity, not about trying to justify a pet-project. However, unlike at CERN, we look at qualitative as well as qualitative data. We don’t fall into the business-case trap of using mathematical elegance to paper over large unfounded assumptions and limited actual awareness of how real people will react to the AI products we’re testing.
Integration over isolation – This effort needs to be integrated into, and communicate constantly to, the parent organisation. Want to be ‘pirates in the navy?’ Okay, but the idea here is to scale valuable AI capabilities across the org fast, not attack it. The thing about pirates is that everyone, except the pirates, hates pirates.
Dexterity over following a plan – Innovation is creative as well as scientific. We sometimes find people mistake our scientific mindset as somehow anti-creative. We prefer the yin-and-yang analogy, you need science and creativity working in harmony, and they really can. Our methodology is about guiding, supporting and protecting creativity, not eradicating it. Put it this way – do you want a creative team to come to you and say ‘here a 5 great concepts, pick one’, or would you rather that same team come and say ‘here are five great concepts. This is what our early research data tells us, and that is why we’re recommending you take forwards 2 & 3 into the next round of iteration – here’s the plan for that’ …

Part Five - A Deep-Dive into the six-stage process:

Stage One - Meticulous alignment:

Antler is currently tracking 209 start-ups in the field of generative AI alone – Check out their list here: link. Many of these companies offer multiple products.

That’s a lot of complexity to keep track of, and that’s before we even consider how to manage the team(s) we’ll need to be working on this internally.

How to manage complexity is not a new question. Project management is not a new profession and there’s no shortage of frameworks designed to conquer complexity with process and governance. However, before you re-assign your PRINCE2 qualified project managers to your innovation effort, we’ve learned that Innovation, especially technological innovation, traditional project management methodologies are not optimal. Instead, we’ve found success by borrowing the Lean concept of Jidoka.

Jidoka is a somewhat abstract principle, but in short it means making everyone in the innovation effort aware of what is happening all the time so that they have a clear basis for decision making. We like the football analogy in Niklas Modig and Par Ahlstrom’s book This is lean – A team of football players who couldn’t see the pitch boundaries, the goal, other players, the score or how much time is left - and who couldn’t hear the whistle, or their coach - wouldn’t be very effective. If we equate our AI Innovation effort with this game of football, far too often we see innovation teams playing independent games to different rules with different goals and no communication. This is the antithesis of alignment, and it WILL cause failures.

We could write an entire other article here about the methods and tools we use to achieve Jidoka – for example our Lean Portfolio Prioritisation system - and probably will in the future. However, to keep this paper within a respectable word count we’ll finish by underling our key learning – before you start setting up an AI innovation effort, spend a lot of time putting the scaffolding in place that will support you to build in the right direction. Be clear on the goals, how you’re going to track them and how you’re going to track and display operational + performance data in a way that is near real-time accessible to all.

Some common pitfalls to avoid:

Different teams in different project-management tool-stacks – some use Jira, another team uses Asana and a supplier won’t work in your instance of MS Teams. It’s tempting to bodge your way around these issues. Don’t. Get everyone in the same tool-suite from the start. Invest in keeping the tools well-structured and up-to-date. Make information transparent and simple.
Too many chefs & part-time cooks – For those of a slightly older generation there was a popular film called Highlander in which a repeated line was ‘there can only be one’. We have what we call ‘The Highlander rule’ when it comes to Innovation Ownership. There must be a single senior and empowered decision maker. They must be near full-time dedicated to the role. We do not have time to do decision by committee or, even worse, try and manage four or five people all under the impression that they are in charge. Given the overlap between the CIO, CMO and CDO when it comes to AI, the CEO has role to play here in ensuring roles and responsibilities are clear and understood.
No standardised objectives and measurable key-results – ‘Ding cool stuff’ is not in and of itself valuable. Being clear about what you need new AI products to deliver is critical. We’ve seen many innovation labs fail at this point – great at making impressive things that sound like they should be great, but bad at turning that into metrics that matter – lead-time reduction, revenue uplift, margin improvement etc.
Over-reliance on virtual tools – Visual management works. Period. Miro and Mural are great places to master information, but having your team stood in a Obeya room all looking at the same information is one of the most powerful alignment techniques we’ve ever found.
Narrow skill-sets – In our experience people who’ve spent their lives becoming amazing experts in a field aren’t so good at understanding what others bring to the table. Designers tend to think that if they just delight the customer enough then profit will magically appear. Finance tend to think that if there’s enough forecasting and data modelling then profit will magically appear. Technology think that if the code-base is neat and high-quality then profit will magically appear. Marketing think that if you spend more time on amazing impactful creative then profit will magically appear. Operations think that if your operating model is lean and effective then profit will … you get the point. They’re all right, and also all wrong. Innovation requires all these skills working together in an environment where they can collaborate effectively and are focused around a common well-understood goal. Designers benefit from having a qualified management accountant working with them day-to-day, and vice-versa.
Rush to get started – Maybe the biggest trap. We believe in being extremely disciplined and meticulous in shaping and testing our innovations when it is cheap to do so. We go slow when the cost of change is low, and we make sure we get to a low-level of detail before we’re ready to start making expensive decisions about what innovations to scale and what to drop. Conversely, once we have the detail, then we sprint. Just as going too fast into expensive building and scaling of AI is a trap, so too is getting bogged down in long innovation build / delivery phases. Paradoxically, going slow is what then allows innovation teams to go really fast.

For more detail on steps Two - Six contact me using the form page. It’d be great to chat.

Sources and referenced works:

[1] This data is cited in Gary Pissano’s highly recommended book Creative Construction. Wherever we’ve looked this data replicates. It also ties with our lived experience. If still sceptical, we also recommend How Big Things Get Done by Professor Brent Flyvbjerg and Dan Gardner to get a cross section of alarming failure rates in most sectors, as well as the recent article Keep your AI Projects on Track by Iavor Bonjnov in the Nov-Dec 23 HBR magazine which shows failure rates >80% in AI innovation

[1] https://www.forrester.com/report/generative-ai-what-it-means-for-content-management/RES179750

[2] https://cmosurvey.org/wp-content/uploads/2022/09/The_CMO_Survey-Highlights_and_Insights_Report-September_2022.pdf pg.21

[3] https://docs.google.com/spreadsheets/d/1IeI-gIZmxIKh5JoAME8WWLi6OIhMgTZbGRB7RPWxzlA/edit#gid=0

Marketing & the AI maze … or ‘how not to waste a ton of money’