The AI agent playbook: When, where, and how to deploy AI in your marketing strategy
By now, we’ve woken up to the reality of AI: It’s not scooping up jobs wholesale. Instead, the marketers using it are saving time, money, and headspace.
Still, there’s always the feeling you’re missing the next big thing with AI.
Many of us are already stuck in our ways. We use AI that gives us shoddy approximations of complete work, and we’re left cobbling subpar results together.
Successful AI implementations incorporate AI from the offset. They require intentional tool experimentation and selection, team onboarding, and routine auditing. What’s more, they require the right top-of-the-line AI agents for the right job.
Let’s teach you when, where, and how to deploy AI in your marketing strategy with the right fit for your team and industry.
First off: What’s the difference between basic AI and AI agents?
Current definitions for AI tools are like snowflakes: each definition is unique. Because it’s the Wild West of AI, nobody totally agrees yet on who is who in the AI world.
But for now, let’s roll with Dharmesh Shah’s definition. AI agents have these key elements:
- They're powered by Large Language Models (LLMs)
- They have access to tools and APIs (their own little utility belt)
- They can remember things (at least during their running time)
If you want to learn more about AI agents in general, we have an article for that.
For now, let’s learn how to use these robots to their best effect.
When to use AI agents (and when not to!)
AI agents are key — got it. So, where are marketers deploying AI agents? This is your guide to deploying AI that’s tasteful and painless.
Let’s start with some universal green lights and red flags to look for. Successful marketers choose their AI battles.
🟢 Here are the green lights for AI agents:
- High-volume, repetitive decisions: AI makes quick decisions, faster than humans. You can set AI to approve transactions or bucket support tickets without lifting a finger
- Real-time optimization needs: AI can make lightning-quick decisions in real time. This makes it ideal for something like optimizing ad bids, a task that takes a human more time to react to rapid changes
- Complex, multi-channel orchestration: Connecting the dots across multiple channels can take hours, days, or weeks of human work. But for AI, managing marketing campaigns across several channels is simple when it integrates with them
- Data-rich environments: Give AI some structured data, and it can extract and transform it in moments. After all, AI is trained on large datasets; it eats data for breakfast!
🚩 These are the red flags for AI agents:
- Brand-critical creative decisions: AI can’t write your positioning doc for you. Because this requires a creative perspective, it shouldn’t be left to an agent trained in other companies’ branding
- One-off campaigns: AI learns patterns from repeated actions, and we better learn how to use AI when we create processes for it. If you want to try something off-the-cuff once, best to do it yourself
- Low data volume scenarios: AI trained off of a handful of data points won’t make sound decisions. Remember: AI eats data!
- Highly regulated communications: Does your industry have strict compliance requirements? AI-generated content and decisions aren’t always 100% in compliance, so if you plan on using AI, tread with caution
You’re probably already seeing the cracks in some marketing suggestions with these red flags. Some marketers have been using AI to create ad copy for their cryptocurrency companies out of compliance and create new brand charters.
Let’s be the bigger marketer and learn how to apply AI agents for our industries without risking our reputation.
Industry-specific use cases: AI agents at work
These are the best use cases for a handful of big industries. By this point in AI’s lifespan, we better know what works and what doesn’t – see what you can apply to your job!
Retail & e-commerce
Best for: Inventory-aware promotions, cross-sell recommendations
What is an e-commerce storefront if not a cache of data ripe for extracting? Unsurprisingly, AI pairs well with retail because it’s more adept at sifting through dozens (or hundreds or thousands) of products and their details than human counterparts.
Retail has been using AI agents for years, ever since Amazon paved the way.
Every “An item on your wishlist is on sale!” notification is powered by AI. No, there isn’t some gnome who lives in each shopper’s computer that manages their wishlist. It was AI the whole time!
Similarly, there’s no gnome individually handpicking related products for each e-commerce item. Unless you consider AI a gnome:
Avoid: Brand voice, creative design
See also: Our red flag list of AI no-nos. AI agents aren’t so great at thinking outside the box; they’re good at thinking within the confines of their training data. Don’t use them to differentiate your brand!
Example: Dynamic pricing optimization
Dynamic pricing is another e-commerce example that’s blowing up. Amazon is notorious for this strategy — their product prices change every 10 minutes.
TikTok is also a rising star in dynamic pricing, cutting user deals with exclusive coupons if they revisit product pages or interact with ads. You can check the “My Offers” section of the TikTok shop for products on sale related to what you’ve bought or wishlisted.
Financial services
Best for: Risk-based engagement, life event marketing
Financial services can tailor their offer to users who may not qualify for a particular credit card or service due to their credit score, offering alternatives for their needs with AI-powered recommendations.
They can also target shoppers who’ve started a new business, are searching for a home, or are paying off student debt. While these referrals used to be manual (for instance, a bookkeeper recommending a legal writer to create a contract for your small business), intent-based data gives AI agents what they need to serve your products to the right people. Many digital advertising platforms employ AI for this purpose.
Avoid: Compliance-sensitive communications
Because many AI agents use the data you give them for further training, you shouldn’t give them sensitive customer data. There are also notorious stories of chatbots giving users sensitive company information because they’re jailbroken. Be sure your customer-facing AI agents are secure!
Example: Next-best-product recommendations
A next-best-action model can be applied to financial services — to help them sell the next-best product to a user. Since financial services normally require an extended research stage for customers, companies in the financial sector want to be proactive in guiding customers to the next step.
For instance, you can identify customers who frequent a particular store and offer them a co-branded credit card. Or, you can offer payment deferment options for large purchases. Many banks give you the “Pay Over Time” offer next to purchases over a certain threshold:
Travel & hospitality
Best for: Ancillary revenue optimization, loyalty programs
AI agents can be used to upsell ancillary products. For instance, a hotel website uses an AI agent to recommend spa services or restaurant reservations from their in-house stores.
You can also send shoppers emails about their loyalty programs. Use this for notifying them about points close to expiring, what they can afford with their points, and specific bonus points offers tailored to them.
Hilton Honors does this right with tailor-made offers for your rewards tier:
Avoid: Crisis communications
Nothing reads “apathetic” quite like an AI-generated response to a crisis. It’s better to send nothing at all.
Steer clear of genAI and formulaic responses to disasters, and leave that to a human communications specialist!
Example: Dynamic package bundling
If a traveler is buying tickets for Aruba, maybe they also want a hotel in Aruba? Tours in Aruba?
Travel+Leisure gives an example of personalized package recommendations based on the travelers’ purchase behavior. This data model recommends vacations to members based on where they are in their planning process by combining four key datasets: transaction history, inventory, demographic information, and web activity. More data, better recommendations.
Media & entertainment
Best for: Content recommendations, engagement timing
The OG AI recommendation engine: “People also read…”
News and entertainment media have been using this for years. That’s what kept 2010s teens trapped on BuzzFeed for hours.
You can also use AI for SMS push alerts tailored to your users’ interests. Apple TV notifies you when an episode of a show you watch releases:
Avoid: Original content creation
BuzzFeed flew too close to the sun. Recommending content was one thing, but then they started writing content with AI.
The AI generated content hasn’t paid off, and though BuzzFeed implemented it in response to setbacks, ad revenue continued to fall.
Many media companies are still using AI…covertly. But if you’re caught, the repercussions are significant. Nobody wants to consume AI content branded as creative expression yet, so you’ll potentially damage your brand permanently.
Example: Churn prediction and prevention
Entertainment companies can use AI to stop churn before it starts.
For instance, bundles reduce streamer churn. In response to this information, streaming services are recommending service bundles to users who use one service under an umbrella (think Disney+, Hulu, and ESPN+).
Implementation essentials
AI agents are widely applicable. You just have to choose the right one for your use case!
Here’s your AI agent implementation starter pack.
- Quality data: AI agents depend on data. If it’s high quality, the more the better. Establish centralized repositories to store and manage data, like a data warehouse or a CDP
- Consolidated data: AI agents need consolidated and cleaned data, not inaccurate or outdated or inundated with duplicate content. This makes integration with existing CRMs, analytics tools, and the like simpler if those data sources are cleaned (and AI automation can help you with that, too!)
- Clear metrics for success: To steer the ship, choose your guiding star. What you should measure depends on the AI agent you’ll deploy, but some useful stats to check for are the accuracy of responses, task completion rate, average response time, cost per interaction, user engagement, and error rate
- Careful consideration: A lot of teams patch holes in their workflows with low-cost AI plugins. What would save them more time is selecting one AI agent that would significantly improve their workflow. Don’t make the mistake of a plugin for everything.
Future-proofing your strategy
AI can be future-proof; Amazon’s product recommendation engine has been around for over 10 years, and it’s still kicking. So even though some AI agents are a fad now, the technology has been around for decades and will carry on.
Futureproofing your strategy just requires you to build on the right foundations. AI isn’t your “get rich quick” scheme; it’s your index fund. Choose carefully, and it will pay you back. The quickest wins are usually the smallest, and that applies to your AI agents’ foundations, too.
Your AI agent should be scalable, taking on greater demands without the risk of losing performance. This is because, ostensibly, your team will grow and your customer base will grow. Select AI agents with that in mind!
Like all good processes, you need documentation to cement an AI agent as a permanent fixture at your company. The best implementations of AI over the past couple of years have been heavily documented so everybody on the team can use them. Otherwise, you’re all individually reinventing the wheel when you work it. Disseminate your information!
Keep these in mind, and your AI strategy will outlast the hype cycle.
Conclusion
Despite all the cautionary tales we’ve shared, it’s really not that hard to get started with AI agents.
Your first step is to find a problem that you can automate with data and machine learning. Maybe one of the examples above piques your interest, or you can read another one of our guides on AI for personalization or AI for customer loyalty.
Then, use your favorite research tools to spot an AI agent that solves your problem. We’re marketers. We love research!
The teams that will bake AI into their processes for the long term find agents that are tailored to their specific problems, not glamorous new products that everybody is using for the hype. Use your marketer’s eye to spot the difference, and you’ll be on your way to using AI like it’s second nature.