New Call-to-action
6 Limitations of Marketing Artificial Intelligence, According to Experts
Blog Feature

By: Paul Roetzer

Print this Page

June 20th, 2017

6 Limitations of Marketing Artificial Intelligence, According to Experts

Artificial intelligence can have a transformative effect on marketing, making campaigns more personalized, productive and predictive. But the technology comes with very real limitations. 

You’ll currently need to rely on individual solutions to perform certain AI-powered marketing tasks. For instance, Narrative Science and Automated Insights provide natural language generation (NLG) capabilities, an AI-powered technology used to scale content creation. Albert uses AI to optimize paid campaigns. And solutions like Cortex, Acrolinx and Skyword offer various ways to personalize and optimize content marketing.

With no one-size-fits-all solution, but plenty of powerful specialized tools, it’s important for marketers to clearly understand the real possibilities and limits of artificial intelligence.

To help accomplish that goal, six artificial intelligence experts told us what they see as the biggest limitations in AI today.

1. You Need Rich Sets of Data

Screen Shot 2016-11-10 at 8.33.14 AM-2.png“On a practical level, artificial intelligence is limited only by the availability of data. Because there’s a direct relationship between the richness of data and the capability of an artificial intelligence application, data becomes the key to an AI system.” — Adam Long (@Adam_B_Long), VP of Product Management, Automated Insights


2. Reliable Prediction and Analysis Are Still Uncertain 

Screen Shot 2016-12-29 at 10.24.44 AM-2-1.png“In the 1980s there were ‘expert systems’ that were considered artificial intelligence. They had completely hard-coded rules, but we’ve now moved beyond that to statistical models to predict outcomes. AI now is basically just algorithms most people don’t fully understand yet. For example, ten years ago Google Maps used to be considered ‘artificial intelligence.’ Now it’s considered basic.

Reliable sentiment analysis (i.e. figuring out if a sentence is happy, sad or sarcastic) is really hard for artificial intelligence, along with reliable sentence parsing. Visually recognizing a teacup reliably is challenging for current machine vision algorithms. Folding laundry is another incredibly hard task for artificial intelligence powered robots. The human intuitions underlying what data to look for and what questions to ask are some of the biggest limits now.

“Another limitation is the ability to operate on very small amounts of evidence or data. For example, if you look at a lot of the bioinformatics applications, each piece of data is a patient’s DNA. In those cases you don’t have a lot of data, so it’s very difficult for researchers to figure out which genes correlate with illness.

“Another difficulty for AI is trying to optimize for multiple dependent outcomes concurrently. Let’s say you want to optimize two conversion rates simultaneously; that becomes very complex and difficult very fast. Imagine you’re A/B testing for a landing page, and there’s another algorithm that varies the ad creative that draws people to the landing page in the first place. The way to do that now is to do both independently of each other, because doing both at once quickly becomes computationally intractable.” — Pawan Deshpande (@TweetsFromPawan), CEO, Curata

3. It’s Really Hard to Replace Humans 100%

Screen Shot 2016-12-08 at 10.04.46 AM-1-1-1.png“The main limitation is that building systems where AI replaces humans 100% is really, really hard. We still don't have 100 percent self-driving cars, and when we do it’ll be the result of a huge effort by the biggest tech companies in the world.

“Luckily, there are lots of applications where AI can be used to empower humans rather than replace them. When AI is used to crunch data to come out with recommendations for instance, it still provides a lot of value if these recommendations are ‘right’ only 80% or 90% of the time—because humans can quickly determine the correct actions from the incorrect ones.” — Guillaume Decugis (@gdecugis), Cofounder / CEO,

New call-to-action

4. Your Marketing Stack Is Complicated and Disconnected

Screen Shot 2017-03-09 at 12.39.47 PM-2-1.png“When we look at the application of AI in marketing, the primary limitations lie in the constraints in the system. Those constraints primarily come from the inputs into the model, the algorithms available to the model and/or the actions the model is capable of driving.

“In marketing, the disparate disconnected systems that form the marketing stack can be a major obstacle in the first of these. Historically, you’ve had different platforms for content creation, analytics, content hosting, demand generation, sales, etc. Consequently, there is a lot of data loss. When you lose the connection between data points, you are only able to assess some behavior in aggregate versus individual use cases. This also means delays in gathering the data so you can’t act in real time or near real time.

“In our particular space, that meant marketers didn’t have the content, usage data and channel data—and couldn’t access direct results. We’ve tried to tackle this by creating an integrated platform with a range of data inputs that tie out to a single user.” — Tom Gerace (@tomgerace), Founder / CEO, Skyword 

5. AI Isn’t Human Enough

Screen Shot 2017-01-12 at 11.43.19 AM-2.png“Where we see AI as having limitations are in the obvious areas: emotions, feelings, subjective thinking. Humans are unique in their ability to feel in a very complex way and translate those feelings into emotional connections.   

“Limitations in artificial intelligence will also stem from the degree of precision with which technologists are able to replicate human ‘intelligence’ and decision-making.” — Or Shani, CEO, Albert

6. Some Have Unrealistic Expectations of AI’s Potential

Screen Shot 2017-02-02 at 10.35.30 AM-2-1.png“AI has come a long way in dealing with unstructured data as it applies to real world use cases and you can do a ton of cool stuff with that. Much of it works very well. That said, many capabilities are still fairly simplistic when it comes to real world applications, and AI is no crystal ball, superhuman force or singularity.

“And therein lies the second, more important challenge, at least from the perspective of an entrepreneur like myself. Many times clients have very strong opinions on AI, maybe they have done some proof of concept already and they think they are experts. Or, they are ‘armchair experts’ based on the latest TED talk they've seen and have highly unrealistic expectations when it comes to AI. AI has some very powerful, proven use cases for marketing and beyond, but it is important that customers, vendors, providers, consultants and everybody else stick to these value-adding, working applications rather than building castles in the sky and setting unrealistic expectations.

“I am not saying we shouldn't dream and be visionary when it comes to AI for marketing, but I think it is just as important that we deliver real results to prove the power of this technology.” — Andre Konig (@AndreMKonig), Cofounder, Opentopic

New call-to-action

About Paul Roetzer

Paul Roetzer (@paulroetzer) is founder and CEO of PR 20/20, author of The Marketing Performance Blueprint and The Marketing Agency Blueprint, and creator of The Marketing Artificial Intelligence Institute and Marketing Score. Full bio.

  • Connect with Paul Roetzer
Disclosure: Marketing AI Institute writes about and recommends AI-powered marketing and sales technology. In all cases, content and recommendations are independent and objective. In some cases, Marketing AI Institute may have business relationships with companies mentioned, which may include financial compensation, affiliate compensation, or payment in kind for products or services. View a list of Institute partners here and MAICON sponsors here.