How Intel is leveraging AI to drive sales

The pandemic greatly accelerated businesses’ digital transformation efforts. This is particularly true in the marketing domain, where enterprises began to embrace automation and enablement technologies. When McKinsey surveyed 1,500 executives across industries and regions in 2018, 66% said addressing skills gaps related to automation and digitization was a “top 10” priority. Forrester predicts that 57% of business-to-business sales leaders will invest more heavily in tools with automation.

While Intel might be best known for its chip business, it’s among the companies embracing this automation and digitization. The company expedited plans to apply AI throughout the customer lifecycle over the past 12 months, particularly on the marketing side of the house, where the goal was to tap AI to help identify and solve selling pain points. Intel also sought to adopt predictive tech to give its sellers a competitive edge, ideally months ahead of a potential close with buyers.

As Jake Tatel, global director of sales enablement and productivity at Intel, told VentureBeat via email, Intel started its AI marketing tech journey about five years ago. The company’s analytics teams started collecting a wide range of data from prospect websites and social media accounts. Then they combined it with Intel’s own website activity and overlaid it with customer buying patterns to drive actionable insights.

“Our internal team has really challenged itself to look at use cases where we could utilize data that we’re able to amass from customer buying patterns, in addition to how potential buyers are engaging across all our properties — whether that be the website, our training properties, or other Intel-owned channels,” Tatel said. “While tapping the data we have internally is a key part of the equation, it was also important for us to weave it together with publicly available information, like prospect websites and social media data. We have a complex ecosystem, so it was critical for us to take a wide view of it all to figure out the best way to stitch it together for continuous, real-time scanning.”

Sales Assist and Autonomous Sales

Two applications sprung out of the analytics team’s early work: Sales Assist and Autonomous Sales. Sales Assist provides insights — referred to as assists — to account managers. These assists broaden managers’ opportunities to interact with customers so that they can cover more accounts. Meanwhile, Autonomous Sales creates automatic sales motions, which are actions that Intel offers customers and partners through emails, website ads, and newsletters. Autonomous Sales operates daily and automatically, without human intervention, and applies to all partners’ accounts — even if they aren’t covered by an Intel sales team.

Sales Assist now has over 1,500 users at Intel and has provided more than 17,000 assists relating to nearly 5,000 accounts, 80% of which account managers have taken action on. For its part, Autonomous Sales delivers a yearly volume of about 30,000 emails to more than 10,000 contacts within Intel’s customers, with an open rate averaging around 36% and a first-time purchase conversion rate of 16%.

Together, Tatel says that the applications are generating “significant” business value for Intel, with the incremental contribution of Sales Assist estimated to be greater than $100 million per year. Autonomous Sales is helping to create $30 million in new sales, he says. And in 2020, Sales Assist and Autonomous Sales together delivered more than $168 million.

“Salespeople won’t adopt any technology if it doesn’t actually serve their needs or help them do their job more efficiently. So, we made it a top priority to ensure that the ‘assists’ being delivered to the sales organization were actionable,” Tatel said. “We did this by building in a feedback loop that informed the development team on whether the recommendations and insights being delivered were helpful. This fostered collaboration between the sales and development team, and made it so we could actually increase pipeline via the assists.”

Sales AI

Intel created Sales Assist in 2017, built on the company’s broader Sales AI platform. Sales AI — which is made up of the modules Sense, Reason, Interact, and Learn — is designed to collect and interpret customer and ecosystem data and translate it into useful recommendations.

The Sense module continuously scans, mines, and collects data about Intel’s customers from a variety of sources. The data reflects interactions and engagements between customers including billing information, past opportunities, and first-party data engagements on Intel.com in addition to responses to communications and any affiliations with partnership programs. Sales AI also incorporates external data like a customer’s or partner’s website, news mention, social media information, and so on, regardless of their specific connection with Intel.

Tatel says that Sense runs on millions of webpages, tweets, sales transactions, and customer and partner engagements, transforming them into thousands of data points on over 750,000 companies. In 2020, Sense scraped 15 million webpages and monitored over 347,000 million tweets.

Sales AI employs a number of web-mining techniques to collect data. Leveraging natural language processing (NLP), the platform identifies cross-reference information about products, brands, advertisements, verticals, and other key variables. It then extracts data about the industries in which the customer is operating, such as automotive, communications, or health care. Thanks to pretrained language models including BERT and GloVe, Sales AI can understand the customer’s role (e.g., manufacturer, integrator, or reseller) and the technologies it’s using, according to Tatel.

“The Sales Assists AI engine automatically tags content and web pages using NLP and then notifies the seller whenever there is product intent of interest which is significantly different from a usual behavior pattern,” Tatel explained. “The approach imitates the way a seller thinks and acts, identifying potential opportunities and helps them to proactively engage customers at the right time.”

Sales AI also analyzes the text, website structure, links, and images in a customer’s website to learn more about them. When the platform spots mentions of a competitor’s product, it generates an assist, notifying the account manager than there might be an opportunity for conversion.

Mining for insights

Once the Sense module gathers a customer’s digital representation, the Reason module uses this data to begin mining, correlating, and generating business insights. An insight — a prediction of a likely sales motion — might be a suggestion that a customer needs an existing Intel product, for example, based on a new direction published on their website. Or a customer might announce an acquisition of a smaller company in a new business line, and an Intel salesperson could recommend what they need to grow this business line.

According to Tatel, Reason can deduce insights from one or multiple data sources by attempting to identify a customer’s goals, needs, organizational changes, and shifts in focus areas. This enables Sales AI to identify trends, analyzing customers’ purchasing history and product-related activities on Intel’s platform and monitoring information regarding product lifecycles, examining past purchase patterns to identify the customers that might be impacted.

“We’re combining the ‘human intelligence; of our sales team with the AI in a unique way,” Tatel said. “For example, Sales Assist can now spot when a customer has a product that’s end-of-life and can recommend an alternative product for a salesperson to suggest to the customer to aid in their transition or make sure they have the latest and greatest solution.”

For sales teams at Intel, Sales AI predicts topics of interest to a company by connecting between different webpages using links and users’ journeys to build a topic network. The platform can identify changes in this network over time, alerting salespeople with an insight when a customer shows a shift in interest toward a new industry or technology. And it looks for “unusual behavior,” such as a sudden increase in download activity from Intel’s Research and Development Center.

Recommending products

To assist with product recommendation, Sales AI offers a recommender system that considers multiple customer objectives. Combining three components — features creation, a recommender model, and an optimization model — the recommender system provides updated recommendations while focusing on revenue opportunities, according to Tatel.

The recommender model considers products that a customers did or didn’t purchase in the past, plus the customer’s expected volume of purchases and their internal priorities. This nets a ranked list of products, which is updated weekly based on the customer’s activities and characteristics. The optimization model refines the list by taking into account sales strategies and feedback, so that it doesn’t always recommend the same products.

The Sales AI Interact module takes over at this stage. Working with results from Reason, it aims to push recommendations to customers at the right time, way, and format, from web to email. The Learn module feeds information from customer and partner interactions into the algorithms powering Reason and the rest of the Sales AI modules throughout, allowing them to self-improve over time.

Looking ahead

In the future, Intel plans to broaden the applicability of Sales Assist by expanding the number of account types that receive assists. The company also plans to add the ability for Sales Assist to recommend specific actions to sales representatives, like sending a customer a link to products they recently viewed on Intel.com.

“Now that Sales Assist is rolled out more broadly, we’re continuously looking for new ways to add intelligence into our sales processes, and have started to look at off-the-shelf solutions … to help with content enablement,” Tatel said. “We’re going to be taking all the learnings from building our own internal AI-powered sales tool. These tools only work — and will only be adopted — if the insights they deliver are insightful and actionable. So, we’ll be building in similar feedback loops and beta programs to ensure they’re fine-tuned for success.”