AI for Marketers by Hubspot Academy

The course in 1 minute

Understand how brands can now interact with customers by delivering hyper-personalised experiences through machine learning, deep learning and natural language processing. Marketers can utilise artificial intelligence (AI) for predictive analytics for nearly every step of the flywheel, for example, lead scoring, content production, and customer service, and most importantly, at scale. Organisations who enable marketers to use AI must have ethical considerations in mind, such as data privacy, transparency and biases in order to maintain customer trust and compliance.

Actionable takeaways and thoughts

Try out for yourself how generative AI tools such as ChatGPT and Gemini (previously Bard) can augment marketing processes or content, using the prompts this course has provided as a guide, in chapter 3.
Though there are many guides out there on the best prompts to use, but I find just tinkering with them on your specific use case is better. You almost never get the perfect answer on the first go, so just start and iterate as you go along.
Even if the output sounds persuasive, there can and often will be better answers if you get more specific. Use their answers as a guide, not as truth.
Ask the chatbot in multiple ways how they have come up with the answer. They often make mistakes, too.
Remember, if you’re using this tool, so are many others out there. How can you have these tools augment your thoughts processes or work, and give you an edge in this competitive market?

AI for Marketers summary

This is my online learning course summary for AI for marketers by Hubspot Academy. My summary and notes reflect insights from a seasoned marketer who is experienced in leveraging generative AI tools, although I try to encapsulate the entire essence of the course in bullet points for anyone new to this subject.

Artificial intelligence refers to the simulation of human intelligence in machines, performing tasks that normally require human cognitive abilities. For example:
- Visual perception
- Speech recognition
- Decision making
- Problem solving
Traditional programming refers to predefined rules and instructions to execute specific tasks. Its output depends solely on inputs and rules, and it does not have the ability to learn or improve through new data.
Since the 1950s, AI has improved dramatically due to increased computing power, availability of large datasets, and improvements in algorithmic techniques.
What can marketers do with AI? Marketers can combine all the below approaches all at once for maximal efficiency.
- Recommendation engines provide suggestions at scale. ML models analyse huge amounts of customer data, grouping them based on attributes then try to predict their behaviour or responsiveness. E.g Netflix uses deep learning to provide personalised recommendations and targeted promotional content to individuals
- Predictive lead scoring predicts which leads will convert, improving sales reps efficiencies.
- Sentiment analysis uses NLP to understand the tone or sentiment behind text to classify them positive, negative or neutral across multiple channels.
- Computer vision AI enables brands to identify and understand objects in visual media to evaluate engagement and brand awareness. They are used also in programmatic advertising to deliver highly targeted ads based on analysing image content.
- Highly customised ads, websites, emails, recommendations to individuals using deep learning algorithms
- Using predictive analytics, AI learns the customer behaviours and patterns to forecast needs, predict churn, anticipate purchases and improve upcoming campaigns.
- Synthesise market research using AI, understanding the big themes with customer interviews, insights and providing feedback to the product team.
- Personalise and optimise websites, ads, pricing, email, search, social media and others
- In customer service, common customer queries can be answered using NLP capabilities, such as personalised chatbots. At Sephora, they can provide product recommendations, schedule appointments, even complete transactions. In beauty, it can match makeup colours to user photos. They save time for customers and provide instant brand interactions.
- Generative pre-trained models can produce code for blog posts, ads, websites based off prompts, allows for search engine optimization (SEO), scalable content production, dynamic personalization, continuous testing and consistent branding enforcement through multiple touchpoints.
- Natural language generation (NLG) helps with scaling content production so marketers can do more strategic work.
- Cross channel orchestration by coordinating messaging and customer interactions through touchpoints for omni channel experience.
- Automating complex data analysis to adjust campaigns in real-time.
- There are many others listed such as revolutionising SEO, or analytics related tasks e.g tracking bounce rate, engagement rates, optimal social media posting times and SEO rankings, many of which a seasoned marketer would already be using in tools such as Google analytics, Jasper.ai or SEMrush.
- The best thing mentioned however, is that AI can uncover insights that a human has not thought of to ask. This can be really powerful; feed it data and let it tell you what it finds.
Ethics and responsible use in AI for marketing must be front of mind.
- Data privacy, transparency and bias are key considerations. Companies must uphold ethical standards and build trust through thoughtfully addressing these.
- Increase AI literacy in organisations, educate on risks and considerations while using AI.
- Data privacy
  - Only collect necessary information to deliver a personalised experience
  - Strong cybersecurity measures to protect customer data from breaches, e.g encrypting data and limit internal access.
- Follow California Consumer Privacy Act and GDPR practices on personal data
- Transparency
  - Provide customers with transparency and control over their data. Enable easy and fast opt out options or delete info.
  - Understand what the third party AI tools are doing with the data that you input.
  - Companies must clearly communicate their use of AI to customers in clear language, or it can erode trust. Publish an ethics framework if you can.
- Biases
  - Data used to train the AI models must be diverse, and representative of the target audience, including various demographic groups, geographies, customer segments. Historical biases must be corrected and output fact-checked and continuously monitored.
- What companies can do
  - Privacy impact assessments to identify risks and collecting and using customer data
  - Create consent flows to explain to customers how their data will be used and provide opt out options
  - Vet ai software providers carefully, select only ethical providers and test them extensively before deployment to uncover issues
  - Build cross functional teams including marketing, legal and IT teams to monitor marketing campaigns for compliance with internal ethics policies and external regulations.
  - KPIs can be identified to measure against the risk of using ai tools
  - Ask customers to provide feedback on the replies and then promptly fix and address
  - Add prominent notices on website that you re using ai for advertising
  - Establishing strict protocols to scrub training data and proactively test for biases
  - Add humans in the process to review campaigns for fairness
An MIT study showed that using ChatGPT can decrease the time it takes to complete a task by 40% and output quality rose by 18%.
Scalable content production is here. AI can create thousands of unique content pieces. Examples include product descriptions, blog posts, social media posts, newsletters and press releases.
Creative experimentation – input a specific tone of voice and explore creating different content styles. Examples include ad copy testing, content repurposing, tone variation, translation and localization, interactive polls/quizzes. Doesn’t stop here.
Image generation and editing can include creating original content from scratch or modifying colours in a product/logo/image.
Video content such as brand spokesperson videos, product demos and virtual influencers with little human intervention.
Automated design, much like having an inhouse graphic designer, creating infographics, banners, social media graphics all to your brand aesthetics and principles.
Four elements you need to pay attention to when working with AI. 1) Know your audience well 2) Optimise for SEO 3) Measure performance to guide future efforts 4) Ensure it maintains a human touch to resonate with your audience.
Prompts used must be specific. DICE framework can be used. D: Define (the goal) I:Identify (target audience) C: Craft (the prompt) E: Evaluate (results)
8 prompting methods for text; comprehensive, iterative, layered, feedback loop, guided, parallel, flipped interaction, alternate approach.
I found the examples of each prompting method very helpful. I’ve learnt that being very specific and explicit is important.
Prompting tips for images/videos: Research the specific prompt syntax, capabilities and community norms of your chosen platform.
- Describe angle, distance, colour, style, or other visual qualities
- Describe artistic style, such as impressionist oil painting, illustration, film still, hyper realistic, cubist, surrealist, renaissance, even artists like Frida Kahlo or Salvador Dali
- Describe angle or distance, such as aerial view, fisheye lens, macro, x-ray vision, split screen, side view, closed up, zoomed out, shot from above or below, point of view from bystander
- Use emotive descriptors, such as melancholy, serene, isolated, longing, wonder
- Consider lighting such as natural sunlight, overhead, backlighting, spotlights, lens flares, warm gold hues, vibrant neon, calm gentle mood
Know that generating humans can be challenging due to biases towards certain demographics. Improve realism and accuracy by being explicit. Add the attire, ethnicity, age, body type and other characteristics. This helps the AI not to default to its training bias.
Think critically about AI. It has limitations on what it can do. It can provide wrong information; always verify. Avoid over-reliance, preserve human agency, use AI as a tool not as a substitute for decision making. You can ask the AI’s reasoning, how did it get to this answer?
One should ensure you are doing the following when using AI
- Leverage A/B testing for experimentation
- Automatically improve recommendations based on user feedback
- Review analytics dashboards to fine-tune personalisation
- Analyse customer data frequently to update segments and personalisation models
Best practices for organisations to take advantage of AI predictive modelling:
- Accessibility to dashboards and modelling tools, so decisions can be made on all levels not just for a select few
- Cross-functional training and ensure collaboration on how to interpret and apply AI insights, inculcate data-driven decision making
- Create templates to analyse model outputs so you can facilitate scenario planning
- Fine-tune models as you go
- Invest in change management
Understand that AI provides suggestions, humans make decisions.
This course starts off strong by providing a good overview of what AI can do in marketing, but slowly tapers off by the third section by being repetitive and I had a sense that the course writer was bulking up the course towards the end. As someone who likes learning by reading, I did the course entirely by reading transcripts and found it engaging enough to complete it. I think this course is very helpful for someone who isn’t a marketer to understand how to build an AI-literate department. Additionally, this can be great for business leaders on how to build a solid marketing team by understanding what AI tools can do to generate hyper-personalised content at scale.