The Augmented Agency: Four things we have learnt deploying Generative AI across EssenceMediacom

Five months ago, a small group within EssenceMediacom set out with an aggressive roadmap of Generative AI deployment and the ambition to become the first Generative-AI augmented media agency.

Today this belief has turned into a deep conviction that our agency and our industry are about to transform for the better.

MediaCom and essence were both leaders in the application of technology and AI to communications. essence’s expertise in the area of data engineering, campaign automation and custom algorithms, alongside MediaCom’s applied Machine Learning and modeling chops are an incredible technology and skillset foundation to build on.

Our EM product and tech team therefore caught the AI wave in 2022 and sought and obtained very early access to Microsoft’s private GPT models when it was the only game in town.

Then came a whirlwind few months during which we built, refactored, tested, deployed, advocated, learnt and experimented.

Today, more than one thousand Essentials have access to a private and secure AI assistant, code-named emChat. We have embedded AI capabilities and routines within some of our most sophisticated tools. We have plugged AI into our million-strong audience survey. We have also built “learning and development” bots, trained on client best practice to educate and inform teams on what good looks like. 

And much more is on the way.

While we will no doubt look back at this moment in time as “the first few moments”, the impact and potential we are seeing is already enormous and the “magic” is far from having faded.

Here are four things that we have learnt, so far…:

Generative AI will impact every role in media agencies

Much of the current focus of discussion of generative AI in industry publications is around a small set of creative-led use cases like generating more assets faster and writing search copy.

While the benefits of Generative AI in these areas are real and immediate, they are only scratching the surface of its potential. What we have found in giving wide swathes of our people access to these tools is that every single Essential, no matter their team and level, can benefit from the technology.

An analyst will use the tool to quickly turn some “R” econometric code into Python. An Account Executive will have AI write complex excel formulas to merge two documents together and expedite a bit of competitive reporting. A strategist will use AI to power its brainstorms and act as a contrarian to its ideas. A platform specialist will use it to query a tech spec database. All will tap into it to learn and grow in and across specialism.

The benefits of this technology are incredibly broad, and only unlocked through a democratic and broad approach to deployment and testing.

AI makes people more important, not less

AI hallucinations are a well documented phenomenon. They are pervasive when using these tools and while they can be minimised, it’s currently impossible to eradicate them completely.

Moreover, Generative AI is brilliant at being plausible. Many of the wrong answers it provides, including made up links and sources, are more than credible and require scrutiny and knowledge to identify and root out. Getting good, non-hallucinated answers - also requires fluency in “prompting” - an overhyped concept that is a lot less like “engineering” and a lot more like “knowing the answer you are after when you ask a question”.

Enter our people and their unique expertise. Their knowledge and capability is the key to truly unlocking value out of AI without having false positives guide strategies and answers. AI will make foundational skills like critical thinking, diligence and strategic approaches to problems core to every organisation looking to seriously scale its use.

Appropriate use of AI requires a code of ethics and diversity

Let’s not beat around the bush: Generative AI as it is today is biased. It was built by a small group of individuals in a single location (mostly) belonging to a single demographic. It is trained on a body of data and content that represents, in most instances, the majority view of humanity over the past centuries. It is also more often than not likely to answer questions with the most (lowest?) common denominator.

This being said, Generative AI is surprisingly good at adopting and representing diverse points of views. It, however, needs to be instructed to do so by the user.

An ethical and transparent approach to AI that leads to true and fair representation of diverse consumer behaviours and ideas that address all - not only most and with more of the same - therefore needs to be defined, enshrined and trained into users. 

In its absence, AI will perpetuate systemic bias in the work that we do, which itself reaches billions of people every day. It is our responsibility as industry leaders to ensure it is used with ethics in mind.

AI is not a silver bullet. It will not lead to obsolescence

Do not believe the social media hype. AI might evolve to solve all of humanity’s problems and make marketeers - alongside everyone else - obsolete, allowing us all to live a life of leisure. It is not there today, nor will it be tomorrow or the day after.

What we have learnt from our experiences so far, with both Microsoft’s GPT-4 and Google’s new and excellent PaLM family of models, is that AI will be brilliant at augmenting human thinking for most tasks within agency workflows, and will absolutely shrink the time needed to deliver some of our services - like programmatic, or excel macros, did in their days.

It is however nowhere near good enough to serve as “auto pilot” to campaign planning and does a consistently terrible job when doing so. 

Humans are needed to prompt, assess, elevate, enrich, question, include and all the other amazing things that only we can do. 

For now and the foreseeable future.

What is true, however, is that a generative AI-powered agency, with true distributed availability of the technology embraced by all teams, will far out perform an agency that leverages the technology for gimmicky uses. 

We made a leap and EssenceMediacom is quickly becoming a better org for it. Are you iterating as fast as you possibly can?