A small specialty shop in an urban area has competitors. The owner spends many hours on marketing activities, including customer service. While stressed, the owner wants to implement ways to improve sales of the shop’s products. Is it possible to make these changes?
Often, yes. AI can take over some of the repetitive work, like answering common questions, drafting promotions, or deciding which customers should get which offer. But it works best when you start with one real problem instead of a pile of new tools.
The example everyone quotes is Amazon. Back in 2013, McKinsey estimated that around 35% of what people buy on Amazon comes from its recommendations. That number is more than ten years old now, and Amazon has never updated it publicly. Still, it shows how much money sits in simply showing the right product to the right person.
Key takeaways
- “AI in marketing” usually means one of three things: models that predict behavior, tools that generate content, or agents that carry out tasks. They solve different problems, so it pays to know which one you’re buying.
- Most marketing teams already use AI somewhere. Very few have connected it across a whole workflow, and that’s where the bigger gains are.
- Results from Amazon, Netflix, or Meta come from companies with huge data sets. Use them as a sign of what’s possible, not as a forecast for your shop.
- Start small. Pick one problem, measure where you are today, test for a few weeks, and decide in advance what result would justify going further.
- For most small businesses, the AI already built into their CRM, email tool, or ad account is enough. Custom work makes sense when your data lives in several systems that don’t talk to each other.
What Is AI in Marketing?
AI in marketing means using machine learning to understand customers, produce marketing content, or handle marketing tasks that used to take someone’s afternoon. People often talk about it as one thing. It isn’t. There are three quite different kinds, and mixing them up is how companies end up paying for a tool that can’t do the job they had in mind.
Predictive AI
This is the oldest and, honestly, the least flashy kind. It looks at what customers did in the past and estimates what they’ll do next: who’s likely to buy again, who’s drifting away, which product someone will want after the one in their cart. Amazon’s recommendations and most lead-scoring tools work this way. The catch is data. With a few hundred customers, a churn score is mostly noise.
Generative AI
Generative tools write and draw. Give them a prompt, and you get email drafts, product descriptions, ad variations, images, or a translated landing page. They’re fast, and that’s the problem too: it’s easy to publish a lot of text nobody checked. Prices, claims, and brand tone still need a human eye.
Agentic AI
Agents go a step further and actually do things. You give one a goal, say, “tell me how last week’s campaigns went and pause anything burning money,” and it pulls the numbers, writes the summary, and pauses ads within the limits you set. Because an agent can act inside your accounts, the permissions matter more than the model. Decide up front what it can touch without asking.
AI vs. Marketing Automation
These two get confused all the time. Marketing automation is a set of rules a person writes. If someone abandons a cart, send email A two hours later. HubSpot, Mailchimp, and Klaviyo have done that for over a decade, and on its own, there’s nothing intelligent about it.
AI changes what happens inside those rules. A model can decide which customers should get the cart email at all, and when they’re most likely to open it. A generative tool can write a version for first-time buyers and another for regulars. An agent can watch the results and adjust the sequence without anyone rebuilding it. For most small businesses, that’s the cheapest place to start: add AI to an automation you already run rather than ripping it out.
If you want the backstory, we’ve written about fun facts about artificial intelligence and how AI has evolved in business.
Statistics and Facts about the use of AI in Marketing
AI statistics are everywhere. The problem is that surveys, single-company results, and market forecasts get lumped together as if they measured the same thing. They don’t, so we’ve split them up.
What Marketers Say in Surveys
Salesforce’s Tenth State of Marketing report asked 4,450 marketers about their work between October 8 and November 17, 2025. Three in four said they’ve adopted AI. Only 13% said they use AI agents. Those agent users expect to get back about eight hours a week from jobs like making content variations and pulling data together. Note the word “expect.” Nobody measured those hours, and eight free hours only turn into money if the team spends them on work that earns it.

HubSpot’s 2026 State of Marketing report found that 61% of marketers think AI is the biggest shake-up their field has seen in 20 years, and 80% already use it to create content. Different survey, different people, different questions. So please don’t add these numbers to Salesforce’s or compare them side by side.
What One Company Achieved
Michaels, the U.S. crafts chain, went from personalizing 20% of its email campaigns to 95%, according to McKinsey’s December 2023 write-up. Click-through rates rose 41% on SMS and 25% on email. Those are relative jumps in click-through rate, not percentage points, and McKinsey doesn’t say what happened to sales.

Where Analysts Think the Market is Going
Grand View Research’s current AI in marketing report, covering 2026 to 2033, puts the global market at USD 27.0 billion in 2025. It expects USD 35.4 billion in 2026 and USD 194.8 billion by 2033, which works out to about 27.6% growth per year. McKinsey goes further. It estimates that agents could eventually handle up to two-thirds of today’s marketing activities, and that companies building agent-based workflows can expect 10% to 30% more revenue from hyperpersonalized marketing. Treat both as estimates of what could happen, not results anyone has banked.
Want more numbers? Our AI marketing statistics roundup goes deeper.
AI Applications in Marketing: Practical Use Cases
Every use case below needs certain data going in, and each one should be judged on a single number coming out. The table gives the short version. Under it are a few notes on where each one tends to go wrong.
| Use Case | Input Data | What AI Does | Output | Metric to Monitor |
|---|---|---|---|---|
| Predictive analytics | Purchase history, site and app behavior, email engagement | Scores each customer’s likelihood to buy, churn, or respond | Ranked lists and propensity scores | Conversion rate of high-score vs. low-score groups |
| Customer segmentation | CRM records, transactions, product usage | Groups customers by behavior rather than demographics alone | Segments that update as behavior changes | Revenue per segment, campaign response by segment |
| Chatbots and virtual assistants | Help-center content, order data, past support conversations | Answers routine questions and hands complex ones to staff | Instant replies, routed tickets | Resolution rate without escalation, customer satisfaction |
| Ad targeting and optimization | Conversion data, audience signals, creative assets | Adjusts bids, placements, and audiences in real time | Automated campaign delivery | Return on ad spend (ROAS), cost per acquisition |
| Sentiment analysis | Reviews, social mentions, support tickets | Classifies the tone and topic of customer feedback | Alerts and trend reports | Response time to negative mentions, sentiment trend |
| Automated campaigns | Customer events, segments, content library | Chooses send time, channel, and message variant | Personalized multi-step sequences | Revenue per recipient, unsubscribe rate |
Predictive Analytics
Prediction needs history. If you have a few thousand customers and a clear question, like “who hasn’t bought in 90 days and probably won’t come back,” a model can be genuinely useful. With a few hundred, you’ll get scores that look precise and mean very little.
Customer Segmentation
AI is good at spotting groups you’d never think to define, for example, people who only ever buy during a sale, or who always order on Fridays. Before you build a campaign around a new segment, double-check it against what you already know about your customers. Sometimes the “discovery” is just a data error.
Chatbots and Virtual Assistants
Gartner predicted in 2022 that by 2027 chatbots would be the main customer service channel for roughly a quarter of organizations. Whether that happens or not, one thing holds: a bot is only as good as the help content behind it. Write the answers first, and make it easy to reach a real person.
Ad Targeting and Optimization
The big ad platforms now build AI into the campaign types themselves. In a May 2025 study of U.S. advertisers, Meta reported that businesses using its AI-driven ad tools got $4.52 back for every dollar spent. That’s return on ad spend, and it was 22% higher than the $3.71 average across all U.S. advertisers on Meta in 2024. It’s Meta’s own study and an average, so your campaigns could land well above or below it.
Google says that advertisers outside retail who switch to Performance Max see, on average, 27% more conversions or conversion value at a similar cost per action or ROAS. Keep in mind that Performance Max changes bidding, placements, and creative all at once. The figure doesn’t tell you how much of the lift came from better targeting alone.
Sentiment Analysis
Sentiment tools earn their keep as an early warning. If complaints about late deliveries suddenly double, you want to know on Tuesday, not at the end of the month. They still struggle with sarcasm and mixed reviews (“great product, awful packaging”), so have someone read before you respond in public.
Automated Campaigns
HubSpot, Klaviyo, Salesforce, and most other email platforms now include AI that picks send times and message versions inside the sequences you set up. Try it on one flow first. Cart recovery is a good candidate because you can compare it directly with what you send today.
Generative AI in Marketing
This is where most marketers started with AI, and it covers more than writing. Copy is the obvious one: emails, ads, product descriptions, social posts, and as many variations as you want to test. You’ll save time. You’ll also need someone to check every price, offer, and claim, because the model will happily invent a discount you never approved.
Visuals come next. Product shots on new backgrounds, campaign images, the same ad in six sizes. Check the license terms of the tool you use, and don’t generate pictures of real people.
Localization is quietly one of the best uses. Adapting a campaign for another market can take hours instead of weeks. For anything high-profile, though, get a native speaker to read it. Tone and idiom are exactly what machine translation gets wrong.
Finally, brand review. You can have AI check drafts against your style guide, banned words, and required disclaimers before a person signs off. It only works if your brand rules are actually written down somewhere.
How to Use AI in Marketing: Start With a Small Pilot
We usually suggest a pilot of six to eight weeks, aimed at one problem. You measure where you are before you start, and you agree in advance what result would be worth scaling. It sounds slow. In our experience, it’s the quickest way to find out whether AI actually helps your business or just makes more output.
- Pick one problem and write down today’s numbers. Choose something that costs you real time or money. Answering the same customer questions over and over. Writing the weekly promo email. Losing carts at checkout. Then record how it performs now: hours a week, response time, conversion rate, revenue per email. If you skip this, you’ll have nothing to compare against later.
- Look at your data honestly. Where does the information the AI needs actually live? Customer records in the CRM, orders in Shopify, past campaigns in Mailchimp? Note the duplicates, the gaps, and whether customers agreed to be marketed to. More pilots stall here than anywhere else.
- Try what you already pay for. Your email platform, CRM, ad accounts, and help desk probably have AI features switched off right now. Start there. A custom integration only makes sense if the data is split across systems that need to work together, or if the process is unusual enough that no off-the-shelf tool fits.
- Agree on the rules before launch. What can the AI do on its own, and what needs a human to approve it? A sensible default: AI drafts, a person approves anything a customer will see, and nothing touches prices, discounts, or budgets without sign-off. Put a name next to the job of checking.
- Test it properly. Run the AI version alongside your current process, not instead of it. Split the email list in half, or alternate weeks. Give it at least four weeks so a single good or bad week doesn’t fool you.
- Count everything, then decide. Add up the subscription or development cost, the setup time, and the hours your team spent reviewing and fixing what the AI produced. Output volume doesn’t count as a result on its own. Compare against your baseline. If the gain is still clear after those costs, scale it. If not, write down why and pick the next problem.
For a wider view, see our guide to implementing AI in your business and our tips on choosing a marketing tech stack.
AI in Digital Marketing
For Diwali 2021, Cadbury employed AI to help small Indian businesses “hire” Bollywood actor Shah Rukh Khan. AI technology was used to generate images and video of Shah Rukh Khan speaking, and each advertisement included a personalized message directing consumers to visit a particular local business. By leveraging dynamic creative optimization and AI, the campaign reached more than 500 pincodes and 2,500 small business owners.
Cadbury benefited too: according to the agency’s case study, the share of homes that had bought Cadbury chocolate reached a three-year high of 9.4% during the campaign. The campaign received a gold award at the 2022 Clio Awards. It shows how one celebrity asset and AI-generated variations can produce thousands of local ads that no team could film one by one.

Notable AI in Marketing Examples
Definitions only go so far. Here’s how real companies have used AI in their marketing. We’ve tagged each one with the kind of AI involved, and marked the older cases so you know how fresh the information is.
Spotify: Personalized Playlists
Spotify watches what you play, skip, and save, then builds playlists like Discover Weekly, which launched in 2015, and Release Radar. You get something new every week that feels picked for you, which is a good reason not to cancel. In a market where switching services takes two minutes, that matters.
Amazon: Product Recommendations
Amazon’s “customers also bought” and “recommended for you” sections draw on your purchases, your browsing, and what similar shoppers did. McKinsey’s 2013 estimate that about 35% of Amazon purchases come from recommendations is still the most quoted number. Amazon hasn’t published a newer one.
Sephora: Virtual Artist
Sephora added Virtual Artist to its app in February 2016. Built with ModiFace, it used the phone camera to map your lips and let you try on more than 3,000 lip colors live. Eyeshadow, lashes, and tutorials came in a 2017 update. According to RetailDetail, the feature was used 45 million times by 1.6 million people in its first eight weeks. If you can see the shade on your own face, you’re less likely to buy the wrong one.
Netflix: Content Recommendations
Netflix recommends titles based on what you’ve watched, and it even changes the artwork you see for the same show. In 2016, two Netflix executives estimated that personalization and recommendations saved the company about $1 billion a year in cancellations it would otherwise have had.
Coca-Cola: Social Media Monitoring
Coca-Cola uses AI to scan social posts, including photos, that mention or show its drinks. That helps the team reply faster and notice trends early enough to shape the next campaign.
Starbucks: Personalized Offers
Starbucks uses data from its app and rewards program to tailor offers. If you order the same oat latte every morning, you’re more likely to get a deal on it, or on something similar, around the time you usually come in.
Nike: Personalized Ads and App Content
Nike draws on activity in its apps and on its website to decide which products and ads each person sees. Browse running shoes a few times, and you’ll start seeing new running releases.
Heinz: “A.I. Ketchup” Campaign
In 2022, Heinz and the agency Rethink typed “ketchup” into the DALL-E 2 image generator. Again and again, the pictures looked like a Heinz bottle. Fans then sent in stranger prompts, like “ketchup in space,” and the best images became social posts and print ads. It was a clever way to prove brand recognition without saying “we’re the best.”
ClickUp: SEO Content
ClickUp used NLP-based content tools that suggest headlines, keywords, and topics to cover, and applied them to its blog. For smaller teams, the point is that these tools help a solid SEO plan, but they don’t replace one.
BuzzFeed: Personalized Quizzes
In 2023, BuzzFeed started using generative AI for quizzes that write a custom result based on each reader’s answers. Quizzes were already one of its most popular formats, so it put AI where readers were already showing up.
Alibaba: FashionAI Store
In 2018, Alibaba opened a FashionAI concept store in Hong Kong. Smart mirrors recognized the clothes a shopper picked up and suggested pieces to go with them. It was an early attempt to bring online-style recommendations into a physical shop.
PepsiCo: Consumer Insights
PepsiCo uses AI to analyze consumer and market data, and the findings feed into product and marketing decisions. Less guessing, more evidence.
Zoom and Grammarly: Customer Communications
Zoom’s customer-facing teams use Grammarly’s AI writing help so support and sales messages come out clear and consistent, without adding another round of review.
Mercedes-Benz: Social Listening
Mercedes-Benz uses AI-powered social listening to follow what people say about it and its competitors. That way, the team can answer complaints and compliments quickly and see how the brand is perceived.
Why These Examples Matter
Here’s a side-by-side look by task and approach. The results are measured in completely different ways, so read them as what each company got, not as a ranking of which approach is best.
| Company | Marketing Task | AI approach | Reported result | Lesson for a smaller business |
|---|---|---|---|---|
| Amazon (2013 figure) | Product recommendations | Predictive | About 35% of purchases come from recommendations | Start with “customers also bought” features in your store platform |
| Netflix (2016 figure) | Retention | Predictive | About $1 billion a year saved in reduced cancellations | Personalizing what returning customers see can matter as much as finding new ones |
| Michaels (2023 report) | Email and SMS personalization | Generative | Personalization rose from 20% to 95% of emails; CTR up 41% (SMS) and 25% (email) | Personalize one channel first and measure click-through against your current version |
| Sephora (2016) | Online try-on | AR and computer vision | No public figure | Interactive tools work when they remove a real buying doubt |
| Cadbury (2021) | Local advertising | Generative video | Ads for 2,500 small businesses across 500+ pincodes | One asset plus AI variation can serve many local audiences |
| Heinz (2022) | Brand campaign | Generative images | No public figure | Generative AI can support brand storytelling, not only production |
| Meta advertisers (2024 data) | Paid social | Predictive ad optimization | ROAS of $4.52 vs. $3.71 average | Test the platform’s AI campaign type against your manual setup |
Benefits and Limitations of AI in Marketing
AI can make a small team punch above its weight. It can also create new work, new costs, and new ways to embarrass yourself in front of customers. Every benefit below comes with a catch, and it’s better to know both before you sign up for anything.
Personalization, if Your Data is in Shape
The upside is real. McKinsey research found that 71% of consumers expect companies to personalize their interactions, and 76% get frustrated when that doesn’t happen. AI makes it possible to tailor offers and messages for thousands of people instead of a handful of segments.
The catch is data quality. Personalization built on duplicate records, old addresses, or a wrong purchase history doesn’t feel personal. It feels creepy or careless. Clean the data first, and make sure you have consent to use it.
Faster Content, but Someone Still has to Read it
Generative tools can turn out a week’s worth of emails and posts in an afternoon. That’s the part everyone notices. What gets missed is the review time. Someone has to check every claim, price, and promise, and fix the tone when it drifts. If your editor ends up spending as long fixing drafts as they used to spend writing, you haven’t saved anything. Track review hours from day one.
Automation That Runs While You Sleep
Agents and AI-driven workflows can run campaigns, reports, and follow-ups without anyone pressing a button. For a small team, that’s a lot of capacity. But connecting AI to your CRM, store, and ad accounts costs money and setup time, and things will break. An agent that pauses the wrong campaign or sends a half-finished email needs a fallback. Plan for error handling and an easy way to switch it off, not just the happy path.
Better Decisions
AI is good at finding patterns in data you’d never get through by hand, like which products are bought together or which customers are quietly leaving. It’s also capable of confident nonsense. A pattern can be a coincidence, and a forecast is only as good as the data behind it. Use AI to raise questions faster, then check the answers before you bet a budget on them.
A Competitive Edge That Won’t Last on Its Own
Early adopters do get an advantage. It fades as soon as competitors switch on the same features in the same tools. Using AI isn’t a requirement for success, and it won’t guarantee it. The lasting advantage comes from your own data, your knowledge of your customers, and how well you fit AI into the way your business actually works.
The Future of AI in Marketing
The next few years will probably be less about new tools and more about connecting the ones we already have. Right now, most teams use AI for separate jobs: a draft here, a forecast there. McKinsey and Salesforce both describe a shift toward agents that carry a task from start to finish, like planning a campaign, building the audience, writing the variations, launching, and reporting back.
We’d expect a few practical changes. Customers will talk back more, since AI makes it realistic to reply to every email and message instead of sending “do not reply” blasts. Search keeps moving toward AI-generated answers, so brands will need content that answers questions clearly enough to be quoted. And marketers will spend more time setting goals and guardrails and reviewing results, and less time on production.
None of that removes the need for people who understand customers. It just changes where they spend their time. For more on what’s coming, follow our blog.
Our Recommendation: Where We’d Start
If you asked us where to begin, we’d tell you to skip the big AI strategy deck. Pick the one marketing task that eats the most hours or loses the most money, and run the pilot described above.
For most small and midsize businesses, that first task is one of three things: answering repeat customer questions, producing routine email and social content, or following up on abandoned carts and cold leads. All three have clear before-and-after numbers, they use data you probably already have, and the risk is low if something goes wrong.
What would justify scaling? Two things together. First, a measurable improvement over your baseline after at least four weeks, like faster response times, a higher click-through rate, or more recovered revenue. Second, the gain has to hold up once you subtract tool costs and the hours spent reviewing AI output. If only one of those is true, keep testing or try a different problem.
When is the software you already own enough? Most of the time, at the start. If your customer data sits in one CRM or e-commerce platform, its built-in AI features will cover a first pilot, and probably a second.
Custom integration is worth it when the off-the-shelf route keeps hitting walls. That usually happens when customer data is split across a store, a CRM, a support desk, and spreadsheets that don’t sync. It also happens when you want an agent to act across several systems, or when your process is specific enough that no packaged tool fits. That’s where engineering effort pays for itself.
How LITSLINK can help
LITSLINK is a software development company. We build AI software and connect it to the marketing tools you already use. That might mean linking your CRM, store, and support data so personalization actually works, building an AI agent with sensible approval rules, or adding AI features to your own product.
A couple of examples from our own work. One B2B client had sales reps going through about 14,000 inbound leads a month by hand, and a typical reply took 26 hours. We built lead scoring software that ranks each lead the moment it comes in and tells the rep why, in plain words. Their strongest leads now get a reply in around 4 minutes, and each rep saves roughly 9 hours a week.
The other was an e-commerce platform whose merchants had thousands of products and no time to write about them. About 60% of their product pages didn’t even have a meta description. Our product description generator writes one in roughly 40 seconds from the product data, and it won’t mention a spec it can’t find on the sheet. Merchants in the pilot who refreshed their old listings got 38% more organic impressions over four months.
If your pilot has proven the idea and your current tools can’t take it further, contact us. We’ll look at your data and systems and tell you honestly whether custom work makes sense.
FAQ
How can a small business start using AI in marketing?
Start with one task, not a platform. Turn on the AI features already in your email tool, CRM, or ad account, and point them at a single problem such as customer replies or weekly emails. Measure where you are now, test for four to eight weeks, and only expand if the numbers improve after you count your time and costs.
What’s the difference between AI and marketing automation?
Marketing automation follows rules you write, like sending a reminder two hours after a cart is abandoned. AI makes decisions inside or around those rules: who should get the reminder, when, and with what message. Plenty of automation tools now include AI, but a simple rule-based workflow isn’t AI on its own.
What can AI marketing agents actually do?
An AI agent can carry out a multi-step task toward a goal you set. For example, it can pull campaign results, write a summary, pause ads that pass a cost limit, or draft follow-up emails for approval. Agents work best with clear permissions and a person reviewing anything customer-facing or budget-related.
When do you need custom AI development for marketing?
You need custom development when your data lives in several systems that don’t connect, or when you want AI to act across those systems. It also makes sense when your workflow is too specific for packaged tools. If your data is in one place and your needs are standard, built-in features are usually enough to begin.
Is generative AI safe to use for customer-facing content?
It can be, with review. Generative tools sometimes invent facts, prices, or claims, and they can drift from your brand voice. Have a person approve anything customers will see, and keep your style guide and legal requirements written down so reviewers have something to check against.
How do you measure the ROI of AI in marketing?
Compare results against a baseline you recorded before the pilot, using one main metric such as conversion rate, response time, or revenue per email. Then subtract tool costs, setup time, and the hours spent reviewing AI output. What’s left is your real return.








