A responsible response to AI

As a provider of B2B marketing services to clients around the world, align.me has an obligation to keep up with all tech including the rapidly evolving field of artificial intelligence (AI) and to work out what this means for our customers. With the sudden interest in ChatGPT and other AI technologies, we understand the need to get our heads around this technology on behalf of our clients.

Our goal is to identify the areas where AI can bring real value to our clients and to provide guidance on how they can best leverage this technology. At the same time, we are mindful of the potential pitfalls of AI and are committed to helping our clients avoid them. We believe this is a critical aspect of our responsibility as a trusted provider of marketing services.

We're not coming to this entirely cold. At align.me, we bring a wealth of experience and knowledge to the table, having built and executed over 6000 go-to-market (GTM) plans for customers. We understand the value of data and the role it plays in making informed decisions. As a digital marketing agency, we send tens of thousands of emails and present millions of ad impressions every month, giving us rich experiences that we leverage every day. However, we recognise that this experience has not yet become 'data' that can inform our use of AI. We have a way to go.

In the next sections, we will delve deeper into the state of AI, where we think it can be used, and our plan to master AI for our clients.

Emerging Technologies in Marketing

While AI has been around for several decades, it has experienced a surge of energy and interest over the last decade. According to a report by McKinsey, between 2015 and 2020, investments in AI have grown at an annual rate of over 30 percent. One catalyst for this increased interest is the sudden visibility of GPT3, a powerful learning model that has the ability to generate text that is often indistinguishable from that written by humans. ChatGPT, an interface to this learning model, has also been gaining attention for its ability to mimic human-like conversation. However, it is important to note that these tools are still in beta and can be wrong in important ways. Additionally, while GPT3 is currently free to use, the environmental impact of these models is not yet fully understood.

At align.me, we have 30 digital marketers who are eager to master AI, but we are still in the early stages of our journey. We have some data, but it is not being managed as an asset, and our knowledge on the subject is spread across various sources. We use tools that rely on AI every day without thinking about it. Some of us have experimented with ChatGPT, Watson, and other tools, but we have not yet centralised what we have learned.

To better understand the potential applications and limitation of AI in our marketing, we have obtained a legal opinion on the use of ChatGPT. The opinion concludes that there are no restrictions on our use of ChatGPT, but that our customers cannot copyright AI-generated works. It also suggested that it is best to declare if works are wholly or mostly AI-generated. We are also in the process of developing an AI manifesto that outlines how we plan to use AI responsibly and reasonably, and that declararation is already a part of the manifesto.

While AI is still in its early stages, it has the potential to greatly increase productivity and transform our work for clients. We are committed to staying on top of this emerging technology and leveraging it to better serve them. In the next section, we will discuss where we believe AI can be most useful in the marketing industry.

Near-term uses of AI

Here are some of the more evident potential use cases for AI in our context:

1. Ideating and research: AI can help generate ideas for content topics and conduct research on specific industries and audiences, providing valuable insights that can inform our marketing strategies.
2. Writing on topics about which we are not well-versed: AI can be useful in creating content on topics that may not be within our team's expertise. For example, if a client in the healthcare industry needs content on a technical medical topic, AI can generate a first draft that can be refined by our team of subject matter experts. 'Composable ERP', 'the security risks associated with containerisation', and 'suboptimal fluid flow' are not topics beyond the reach of our highly-skilled writers, but they sure take time to write well for.
3. Basic copywriting: AI can assist in creating basic copy such as best practice guides, product descriptions, and ad copy. This can save time for our team and allow them to focus on more complex tasks.
4. Versioning: creative work is highly subjective. You might love one version of an article, and I might prefer another. By generating a handful of variants of a well-written article, clients can allow their subjective preferences to inform selection.
5. Refreshing existing copy/docs: AI can help update existing content with current data, ensuring that it remains relevant and up-to-date.
6. Icon creation: AI can be used to create icons and other design elements that can enhance the visual appeal of our marketing materials.

However, it is important to note that there are also potential pitfalls to using AI in the our marketing. An obvious risk is the possibility of using AI to generate bad copy. While AI can assist in creating content, it is not a replacement for human creativity and expertise. It is important to ensure that AI-generated content is reviewed and edited by a human copy team and also by a marketer who understands the domain to ensure that it is of high quality and meets the brand's tone and voice. See my comments in the acknowledgements at the bottom of this article.

Another potential pitfall is the accuracy and reliability of the data used by AI. While AI can quickly generate insights and recommendations, the accuracy and reliability of these insights depend on the quality and relevance of the data used. It is important to ensure that the data used by AI is relevant, accurate, and up-to-date.

While there are potential risks to using AI in the B2B marketing industry, we believe that it can be a valuable tool when used appropriately. At align.me, we are committed to staying on top of emerging technologies and leveraging them to better serve our clients. We will continue to explore the use of AI in our marketing strategies and find ways to use it responsibly and effectively.

A plan to get on top of this

Again, we recognise the importance of staying at the forefront of emerging technologies in order to provide the best possible service to our clients. I'm sure you do the same in your own field. As we discussed earlier, artificial intelligence (AI) has the potential to revolutionise marketing, and we are committed to exploring its potential applications. Our clients would insist that we do, and that we do so 'on our dime'.

To that end, we have developed a plan to explore and experiment with AI over the next year. Our plan consists of two learning phases: a month of uncontrolled exploration followed by a year of purposeful experimentation; then a third codification phase.

Go wild and report back: During the first phase, we will encourage the entire team to engage in uncontrolled exploration of AI. There will be no rules, and team members will be free to explore any AI tools or techniques that interest them. The purpose of this phase is to allow team members to become familiar with the capabilities of AI and to identify potential applications in our work.

Throughout this phase, team members will document their findings as they go, allowing us to compile a comprehensive database of early knowledge on the subject. This database will serve as a foundation for the second phase of our plan.

Go fast but with purpose: In the second phase, we will move from uncontrolled exploration to purposeful experimentation. We will identify the domains we want to explore further, such as SEO, copy creation, copy review, ideation, and Funnel Plan for go-to-market planning. For each of these domains, one team member will lead a cross-functional group to explore a use case that we believe will bring value to our work.

During this phase, we will be focused on identifying, validating, resolving, and operationalising use cases within a month. We recognise that decisions made during this phase will therefore need to be rapid and light rather than ponderous and complete. Any operationalised use cases will be subject to ongoing change and refinement if we're to land anything of value within a single month.

Lock it in: As we move through that second phase, we will operationalise our learnings every month. In the final phase, before this becomes BAU and not the subject of a discrete initiative, we will codify our knowledge into training courses and evolve them ongoingly. By doing this, we will ensure that our entire team is equipped with the skills and knowledge they need to use AI effectively in their work.

In summary, our plan to explore and experiment with AI over the next year will allow us to identify the most promising use cases for AI in our work and to develop the skills and knowledge needed to use AI effectively. We believe that by doing so, we will be able to provide even greater value to our clients and remain at the forefront of the marketing industry.

Acknowledgements and reflections

The align.me team allowed me to co-opt one of our Friday Experiment sessions (weekly shared learning) to assemble the inputs for this plan. Most of the ideas contained in this article came from the workshop with the whole team - half in the Melbourne office and half remote (even many of the Melburnians). We had five or so of the team on leave and not able to contribute. Thank you, team, for your thoughts.

Our legal team at Elementary Law drafted our legal opinion. In briefing them, I asked ChatGPT to create a first draft which it did beautifully and quickly. Also inaccurately. The response from ChatGPT turned out to contradict clauses in Open AI's own terms of service, as discovered by the diligent professionals at Elementary Law. Thanks, Jason and Rachel. We love your work.

Our AI Manifesto - such as it is - is wholly copied under a creative commons licence and is free to be recopied on the same basis - from them or us. You can find it here , although, frankly, there's nothing much there from us yet. In time, we'll put more original thought into that manifesto as we become clear on where we think the lines ought to be drawn in our own use of AI.

The original text of this article was generated by ChatGPT, but not without lots of effort. Of course, I provided as inputs my instructions about intent and all of the notes from the workshop, but I also had to instruct clearly how authoritative, how long, and how candid I wanted the tone. We already knew from early experiments how iterative (chatty) you need to be to get the most from ChatGPT. That's how it's been designed.

And it made lots of errors. I had to say a dozen times that we style our company name in all lowercase (align.me, not Align.me). ChatGPT apologised profusely each time, indicated a very clear understanding of the error, and then diligently repeated the error.

Generating a 1500-2000 word essay proved beyond its ability - or mine to instruct, perhaps. After several attempts that generated a 600-word summary of our notes, I felt too much important detail had been lost. Asking ChatGPT to make the article longer resulted not in a more detailed version formed from my inputs but a long version formed from its own summary.

And to do that, it made stuff up. The resultant copy suggested that we'd consulted external experts for the workshop - we had not - and other lunacies caused by trying to turn 600 words into 2000 rather than recreating a 2000-word article from the team's inputs and my instructions.

I finally decided to explain the overall intent and intended structure, then feed the flow of the article into four separate sections.

As a result, the final work came out well, if a bit 'stilted' as ChatGPT had created somewhat self-standing mini-articles with intros and conclusions to each section that killed the flow a bit. I then fed the whole article back to ChatGPT and asked it to tweak modestly for flow. Here's my literal instruction:

"Here is the article as it stands - end to end. Having been pulled together in pieces, it reads a bit repetitively and looks a little like it was written by different contributors. It doesn't need a lot of work, and I don't need many changes, but maybe adjust a little to maintain its authoritative style but read a little more fluently. Feel free to insert headings if they are needed as that is a common element of our writing style."

ChatGPT crashed on this task four times. I then upgraded to 'plus' (ChatGPT's paid version), thinking that perhaps I was asking too much of a free tool. After the upgrade, I found that it still crashed but did so faster.

But it did other things really well. Like when I asked it to fact-check an earlier assertion that I had made without any evidence that AI has really accelerated over the last 10 years. I had no idea whether that acceleration was more accurately described as a few years or 20. The answer provided was clear and useful but it proved difficult to convince ChatGPT to weave its wonderful fact check into the article, so I just copied that response and wove it back in myself by hand later.

The final article was manually proofed for flow and format, although by me on a Saturday, not by one of our copy team on Monday, as perhaps should be our mode. I initially chose not to fix the stilted copy that was a consequence of generating this in parts so you can see the result. The copy style also reads more like a brochure than an article, with too-frequent use of phrases like "At align.me, we..." which we might normally use only once in such an article. I originally left those in as markers of the limitation of the tool. Later, when I decided to share the post more widely, I couldn't do so without cringing, so removed some of the more obvious transgressions.

I used Grammarly - another AI tool we use - to proof the AI-generated copy and my added acknowledgements section. The image is also AI-generated using Dall-E 2, another offering from Open AI - the makers of ChatGPT. I am less comfortable with the fine art of instructing and evolving via Dall-E than ChatGPT, perhaps reflecting my relative comfort with matters visual versus verbal generally. I like that the image is a bit weird, too - evidencing the opportunity and the pitfalls all at once consistent with my intent in penning this article.

Although our designers will likely insist that we replace this image with one that they create, and they would be right to, I kind of liked the idea of a theoretically quick AI-generated image actually took a long time due to my clumsiness with Dall-E. Doing so from home on a Saturday rather than something perfect next week from professionals feels compatible with this article, even if not good practice going forward.

This article is meant to be a bit raw. A bit transparent. My hope is that the candour paints a picture of a company ready to learn and worth backing to do so rather than one hopelessly behind the eight ball. Certainly, many companies are ahead of us in their use of AI and likely some in our own field. I am generally comfortable that we've not been early adopters but do plan to be fast followers. We are told to be our authentic selves, and I am old enough and ugly enough to be reconciled with the risks of being vulnerable. And you? Would you have preferred a more resolved position from us?

I remain excited about the potential of AI generally and ChatGPT specifically. It will not replace us but allow us to serve our clients better. Sometimes 'better' will mean more fully; sometimes more affordably; other times less predictably (if that's what they want).

Fun times.