CES 2020 could have been called the Great AI Festival. There was a home system that signals when appliances need maintenance or are being misused; a retail body scanner that records shopper demographics and moods as they try on clothes virtually; even a digital billboard that projects individualized video and images at vehicles. Naturally, marketers got drunk on the promise of artificial intelligence. For many, the hangover will be spectacular.
Why? Because the data on which AI relies, perhaps like no other software, is a mess. I've seen:
Multimillion-dollar marketing campaigns generating thousands of qualified leads that salespeople never see, because marketing automation and sales contact software programs don't share the same designation of what a lead is.
Customer files filled with old names and addresses and single points of contact when there are 10 or more decision makers on the account—this last problem worsening with two-year management rotations that create musical chairs within the world’s largest marketing companies.
Salespeople operating from email contact files (especially Outlook) invisible to sales and marketing automation software. Worse, non-marketing departments increasingly publish content in PDF format, which can’t be tracked, so the most important part of customer engagement online yields no useful information at all.
AI puts unprecedented pressure on every detail, which can quickly become campaign killers if lacking. Before we can get anywhere with AI, we need to get our data ready. Here’s how:
Deep-clean first party data. Ironically, AI depends on manual verification. We have to check data point by point. If there are 15,000 files in customer and prospect lists, we have to go through each one individually, then across databases attached to martech tools. It starts with the basics—all the decision makers on the account, location, contact information—then moves to transaction history, particularly what’s missing (there are always holes AI will trip over).
AI can reveal cause and effect only if we have a clean, complete record of where customers are in the life cycle. Otherwise, dirt in the data can become its own bias in AI. We need to include how the company has performed over time, from product sales to overall revenues and how budgets have been set relative to performance. Then we need to have the discipline to maintain data cleanliness.
Get all your data in one place
We can drive traffic to individualized conversion points only if we create the right linkages. That means CRM, lead scoring and marketing automation data need to match up with owned and paid media activity. Typically, I find these customer records disintegrated by as many as a dozen different databases and departments, if not outside agencies.
Tools like IBM’s Watson Discovery can import all first-party data from private servers and software programs (e.g., salespeople’s Outlook files) and relevant third-party data (e.g., segmentation studies, target lists, media usage) into one virtual database. Then we can make sure it all matches up accurately, complies equally to governance requirements (e.g., CCPA, GDPR, HIPAA) and follows the same security protocol.
Once the data is all together, we have to make it consistent or we will build marketing automation on a broken foundation. AI needs one set of definitions. For starters, "prospect," "lead," "qualified lead," "sales qualified lead," and "customer" each need to mean a single thing.
Go all the way back in history
If customers are buying a software tool, the conversion cycle might be seven days. If they’re choosing a new ERP system, it could be a year. And if they’re long-term customers buying fleets, it could be continuous over decades. Track every step in the journeys—from clicks on media to website visits to salesperson interactions—and update frequently. That’s the foundation required for AI to model and predict. Specifically, modeling is a balance of activity and time. The whole trajectory allows us to identify periods that are “just right”—showing a clear pattern of behavior over time without too many changes.
Integrate data on environmental factors
The real prize from AI will be a real-time understanding of customer dynamics and motivations. That presupposes vision into what’s happening in consumers' worlds on many levels, which can nourish the analysis of patterns in the physical, economic and social realms. Customer histories need to include what’s happening in the stock market, weather, politics, policy, and dozens of other determining factors B2B doesn’t typically consider.
Now, for the hard part: Marketing needs to loop IT into new technology it is exploring, and both need to respect the twin goals of innovation and system stability. And agencies need to give marketers a lot more than impression-based reporting. 
Marketers need a dashboard that shows where and when customers are seeing what content, and what actions they take in response.
There’s no question artificial intelligence can make insight development function like Google: continuous, instantaneous and comprehensive. And, given the right data, it can help make marketing the precision and proof engine business craves. But before they can get into predicting the future, marketers must seize hold of their pasts. Otherwise AI will be another expensive trophy that fails to deliver the transformation it promised.
