Magnus Unemyr

With the vast amounts of data being produced in businesses today, there is no way for humans to sift through all of it to stay ahead. Relying on manual monitoring means important insights are missed often.

It would be nice to have a tool that monitors all of your business data and notifies you automatically when unexpected changes happen in customer behavior, demographics, and conversions. 

A tool addressing this problem is Outlier.ai, which takes a new approach to business intelligence. Instead of creating new dashboards and running queries when questions arise, Outlier solves this problem by sending an email each morning with the top few anomalies going on in the business.

The ideal use of AI: moving from drowning in historical data to being served with the right insights automatically.

By tailoring insights to each individual staff member, it is like giving everyone at your company a dedicated business analyst who is constantly looking for hidden problems and opportunities.

Sean Byrnes is the CEO and founder of Outlier.ai. He shares an interesting insight from his former career, which led him to start the new company:

“I met with hundreds of companies to discuss how they use data to make decisions. The most common question they would ask me is ‘What should I look for in all of this data I have?’ I realized that the last ten years of business intelligence innovation had centered on answering questions we already had, while the next ten years would focus on helping us ask better questions.”

This captures the ideal use of AI: moving from drowning in historical data to being served with the right insights automatically.

To offer this capability, Outlier and similar emerging platforms use different machine-learning algorithms that turn raw data into human-readable insights, understand what is normal, detect anomalous events, and filter through the clusters of events to select the few that best match the interests of each user.

Expect most business intelligence tools to be AI-powered in the future.

Welcome to a breakfast seminar on marketing automation and content marketing! We will talk about the latest in marketing automation and content marketing. This event is for you if you want to take your marketing to the next level.

Linköping 24th of September – register here

Göteborg 24th of October – register here

Event details

Many companies invest large sums on digital advertisements, and AI is well-positioned to optimize ad investments. In particular, machine learning can be used in optimization of programmatic display ads.

For example, AppNexus and PulsePoint both provide programmatic ad solutions, offering ad purchasing and sales based on machine learning. Other tools analyze ad spending and results across channels; not only your own, but also that of your competitors.

This is ad intelligence for the benefit of both buyers and sellers.

Machine learning can be used in optimization of programmatic display ads.

The digital advertising ecosystem is complex and changes quickly. Advertisers have a difficult time getting transparency on their own ads, not to mention what their competitors are doing.

Many layers prevent transparency, especially in the programmatic markets, where an ad impression can pass through a number of providers—from publisher to advertiser—before being filled.

Having insight into which advertisers are buying on which publisher sites and through which networks and exchanges helps everyone make better decisions.

Advertisers have a difficult time getting transparency on their own ads, not to mention what their competitors are doing.

Pathmatics is a company in this space that offers what can be described as a search engine for ads. They use machine learning to gather insights into ad spending and impressions, which can help companies respond to competitor activities or improve results in future campaigns.

I had a conversation with Kenneth Roberts, Head of Marketing at Pathmatics. He explained:

For buyers of advertising—brands and their agencies—timely information on their competitors’ advertising can make a major impact on the effectiveness of their own marketing and advertising performance. A brand, for example, can know if a competitor has launched a new marketing campaign or started spending for impressions on a new site (potentially a new target market). The brand marketers can compare themselves within their category and determine if they are being outspent. Advertisers can find sites that they are not advertising on—and the most efficient channel partner to buy through. They can also monitor brand safety and detect wasteful placements to cut or reallocate spend.

Pathmatics and similar companies can report on information such as desktop display, pre-roll video, mobile web, native, and social advertising.

Having insight into which advertisers are buying on which publisher sites and through which networks and exchanges helps everyone make better decisions.

For example, the Pathmatics pipeline for detecting ads, estimating impressions, and calculating potential spending uses machine learning models that are constantly updated with market data on prices, ad formats, site traffic, and indexed digital ad data.

It is easy to see how ad intelligence can be a marketing advantage over your competitors.

You can even let AI-based software automation tools manage and optimize campaigns autonomously—from purchasing ads to analyzing results. In essence, you can get a virtual digital marketer that manages and executes self-driven and self-optimizing digital marketing campaigns.

Let AI-based software automation tools manage and optimize campaigns autonomously

This is exactly what Albert Technologies does. You provide the creatives and goals, and their software tool (an AI agent they call Albert) uses machine learning to initiate and optimize ad spending across different media channels and devices over time.

It measures the results and adapts the ad investments automatically as the ROI changes.

Or Shani, the CEO and founder of Albert Technologies explains, “Albert is autonomous, or self-driven, meaning that he can create digital marketing interactions unassisted, using the results of multivariate tests and deep-level analysis to make better decisions moving forward.” He also notes, “Albert can run thousands and thousands of tests across hundreds and hundreds of variables in a very brief window of time.”

Whereas traditional solutions simply analyze data and then wait for a human to make a decision about it, AI tools like Albert can analyze the data, determine the best course of action, execute it, and then continually optimize itself in real time based on what it learns.

While Pathmatics and Albert are general tools for working on multiple channels, you may want to consider specific platforms only as well.

If you are interested in Google Adwords more specifically, AdScale offers algorithmic bid and budget management based on machine learning to help optimize advertising performance. StrikeSocial does something similar, but on YouTube.

It’s one thing to optimize the campaign at large, but another to optimize the ad creative itself. Dynamic creative optimization is the process of adapting the creative to the person seeing it in real time.

Dynamic creative optimization is the process of adapting the creative to the person seeing it in real time.

A huge amount of data can be mined to predict which creative will trigger someone to convert.

Some tools address ad optimization from this angle, for example by predicting the creative quality of your different ads, the optimal number of ads you should use simultaneously, and the best creative refresh cadence to avoid ad fatigue.

Have a look at Refuel4 or Spongecell if you are interested in this. Additionally, Adobe is now starting to add AI capabilities to Photoshop and their marketing cloud products as well.

More and more specialized tools are becoming available, which promises an interesting future for marketers. Only time will tell how much AI will be able to automate graphic design in the future.

Setting the right price is a key factor in business success. Still, many companies use gut feeling rather than facts when determining the cost of their products. With predictive pricing, we can apply machine learning to the problem of setting the best price and improve revenue optimization.

We can change the price dynamically too.

Many companies use gut feeling rather than facts when determining the cost of their products

Dynamic pricing is often divided into two categories: revenue management, which is driven by supply like hotel rooms or seats on a flight, and supply driven, which is propelled by competitor price-matching or out-of-stocks.

Customer acceptance for dynamic pricing is generally quite high for products with fluctuating availability and demand, where the product by nature vanishes after a certain time.

Take a global car rental company with thousands of locations as an example. They may change millions of prices each day to optimize their revenue based on a huge number of factors, from the weather, to local events, to the prices of their competitors.

Customer acceptance for dynamic pricing is generally quite high for products with fluctuating availability and demand

This can clearly not be done manually, and prediction algorithms can help find the optimal pricing for each car, in each location, at every hour.

Luckily, readymade solutions for revenue optimization exist, and they can predict the best price of most products—provided you have enough data to crunch.

One example is Perfect Price, which offers a product that covers both revenue management and supply-driven situations. I had a conversation with Alexander Shartsis, co-founder and CEO of Perfect Price.

He says that “Today, consumers accept and even expect price fluctuations. Frequently, they are able to time them better than the companies setting the prices—costing companies billions annually. It seemed clear that there should be a simple service to automate the complex data science required to make dynamic pricing possible for any business.”

Perfect Price leverages a core platform based on a number of AI technologies, including supervised machine learning, reinforcement learning, clustering, and more.

This platform generates a demand function that is accurate down to the micro-segment level, meaning it is specific for very small groups of customers. It then can calculate an optimal price based on business rules and other configurable parameters in order to optimize the customer’s objective—profit, revenue, etc.

It is obvious this type of technology can benefit many companies, in particular those where the pricing is changing often, as is the case with flight tickets, rental cars, and hotel rooms.

Prediction algorithms can help find the optimal pricing

Additionally, AI can be used to predict the most likely selling price for certain goods that are never exactly the same, like the price of a used car or a house on the real-estate market.

In such cases, data about the sale of similar products are gathered over time, for example the selling price of used cars, along with data attributes like the type of engine, stereo, color, mileage, equipment, the owner’s neighborhood, and more.

Given this data, a price prediction algorithm can guess the most likely selling price for the next used car to be sold, even though it has never seen exactly the same combination of attributes before.

A price prediction algorithm can guess the most likely selling price

This works similarly for real estate properties and other categories of products where there is only one unit of the same configuration available. Regression algorithms are suitable for this kind of functionality.

An unexpected example of a solution like this comes from GoDaddy, the web hosting and domain name supplier.

Their domain value and appraisal tool uses natural language processing to understand the word combinations used in a domain name. It then predicts the price of the domains using those words on the domain name aftermarket.

That’s why when you go to buy a domain, the one you want may be expensive, while similar ones without the same pull can be quite cheap.