Should you buy or build AI? Considerations for marketing teams

The race to implement reliable, scalable AI is on, and you only get one chance to beat your competitors.

Should you buy or build AI? Considerations for marketing teams

It’s 2022, and the debate over whether or not brands need artificial intelligence to grow their business is over. The biggest companies already have it. So do many mid-market brands. For everyone else, the question has moved from whether to invest in AI to how they’ll implement it. For marketing teams, which have a ton to gain from using AI, that question is only growing more pressing.

Generally, marketers have two options: buy AI software from a vendor or build it yourself with a dedicated data science team. Both routes have their pros and cons. Read on to familiarize yourself with the key components of an AI pipeline, then learn which situations warrant the DIY approach and which ones call for outside expertise.

Key components of AI

There are a few things every AI pipeline needs. Whether you’re building out these capabilities yourself or getting access when you purchase AI software, these are the core components you’re looking at:

  • Identity graph: Compiling the consumer data you need to power AI
  • Data science: Creating and supervising your machine learning algorithms
  • Systems integration: Delivering predictions to the rest of your tech stack
  • Operating expenses: Ongoing cost of maintenance

Let’s break them down one by one.

Graph going over the time and cost to buy or build AI. DIY AI costs 8 times more and takes 10 times as long.

First, the identity graph. At its core, AI models discover patterns in historical data and make predictions based on those patterns. It requires a tremendous amount of data to train an algorithm, making data an absolutely critical item on your punch list. You’ll need both first-party data and rich consumer data licensed from responsible sources. Then you’ll resolve all that data in an identity graph to link it to individual consumers.

If you’re creating your own identity graph, you’re looking at a 1-3 month timeline and at least $500,000 to license reputable data and get it off the ground.

Next, data science. Once you’ve got data, it's time for algorithms to come into play. There are a whole host of sophisticated options to consider, from random decision forests to k-means clustering, and it will take a couple experienced data scientists a minimum of 6-9 months to build and supervise them.

Keep in mind that data science salaries start at $150,000, often running much higher, and demand for skilled data scientists is outstripping supply by a long shot.

Systems integration is another must-have. Your predictions will be meaningless if you can’t deploy them to the rest of your tech stack—imagine going to all the trouble of predicting, say, who your most valuable leads are, and then being unable to access that list in your data warehouse or CRM.

For this step, you’ll need your engineering department to build out an ETL pipeline, which will take 1-3 months and roughly $250,000.

Finally, operating expenses. If you’re extraordinarily efficient, you can get all of the above off the ground in a year and a half. But bear in mind the operating expenses—if you're working on a one- or three- or five-year roadmap, you’ll need to account for infrastructure, salaries, data management, and so on. Over five years, those tend to come out to around $1.5 million in ongoing maintenance costs.

These four components—identity graph, data science, systems integration, and operating expenses—are the most basic requirements of a successful AI pipeline. When you put it all together, you're looking at a $2.5 million price tag.

When you should build AI

There’s no way around it: AI takes an enormous amount of time and money to set up yourself. However, in some instances, it’s still worth doing in-house.

One such instance is if AI is central to your product offering. You naturally need full control over your predictive models here, and tailoring your AI pipeline to your unique business needs is crucial. Uber is a great example of an “AI-first” company, where they use algorithms to match riders to drivers and optimize driver routes in addition to optimizing marketing spend and allocation.

Enterprise brands that oversee massive operational systems are also great candidates for in-house AI. Take Walmart, which recently developed a proprietary machine learning platform to automate last-mile deliveries. Once your scale of operations hits a certain threshold, not only do you require the flexibility of your own custom AI technology, but you’re also more likely to have the resources to execute it.

When you should buy AI

Given the massive investment AI requires, it’s rarely worth re-inventing the wheel. The technology has made leaps and bounds in the past decade, and now that it’s fully mainstream, there are plenty of sound off-the-shelf options for every use case. Typically, these options take just 1-3 months to onboard and cost a fraction of one data science salary per year.

For the same reason retailers turn to options like Stripe to set up payment systems, it’s in most brands’ best interest to take advantage of existing AI solutions, freeing up much-needed time and money for other initiatives.

Plus, make no mistake—the race to implement reliable, scalable AI is on, and you only get one chance to beat your competitors. You may well have the money, talent, and resources to build your own AI pipeline, but by the time yours is up and running, brands that bought the ready-made option will have been using their predictions for months. You’ll need to ask yourself if you can afford to be six to twelve months behind from the get-go.

Or why not do both?

It’s important to note that the “buy or build” debate doesn’t have to be completely binary—for some brands, “buy and build” makes an awful lot of sense. Companies with robust data science departments and ample engineering resources may reasonably want to focus on core operational AI, and won't have enough bandwidth for the type of AI that sales and marketing needs.

In these cases, opting to both buy and build is a win-win. Your data science team is free to focus on foundational AI needs without being spread too thin, and other divisions or departments can buy ready-made AI solutions that allow them to hit the ground running in just a couple months. These off-the-shelf options can range from embedded AI functionality for SaaS companies to audience targeting solutions for direct-to-consumer brands.

Put resources where they matter most

For decisions around AI implementation, it all comes down to where you want to invest your time and money. For some brands, the upfront investment is crucial to unlocking growth in the long term. For others, it pays to free up bandwidth and get actionable AI in months, not years.

Should your team buy, build, or both? Connect with a growth expert to discuss your specific goals.