Artificial Intelligence | News | Insights | AiThority
[bsfp-cryptocurrency style=”widget-18″ align=”marquee” columns=”6″ coins=”selected” coins-count=”6″ coins-selected=”BTC,ETH,XRP,LTC,EOS,ADA,XLM,NEO,LTC,EOS,XEM,DASH,USDT,BNB,QTUM,XVG,ONT,ZEC,STEEM” currency=”USD” title=”Cryptocurrency Widget” show_title=”0″ icon=”” scheme=”light” bs-show-desktop=”1″ bs-show-tablet=”1″ bs-show-phone=”1″ custom-css-class=”” custom-id=”” css=”.vc_custom_1523079266073{margin-bottom: 0px !important;padding-top: 0px !important;padding-bottom: 0px !important;}”]

Microsoft’s Modular AI Journey at its Global Demand Center

This article is co-authored by Dr. Arun Shastri (ZS Associates).

Modular AI approaches allow for more flexibility in picking the most burning or highest impact issue first so that later iterations can focus on secondary issues and process improvements. It also releases solutions faster.

We hear a lot of stories about companies that have successfully implemented AI at scale.

But that doesn’t mean it’s easy. It’s not just difficult to develop the algorithms. The process dependencies are a tangle. Building data pipelines and getting organizational alignment and adoption require a herculean effort. You can respond to this complexity by deconstructing the complex business process into simpler problems that can be solved independently or releasing a simplified model of the whole process and then over time consider adding more and more complexity.

Both modular techniques result in incremental development and deployment. The advantages are numerous: less drain on resources because there isn’t a grand scale at the start, more responsiveness to industry and organizational changes and greater adoption, thus greater impact. The AI journey that we’ve taken at Microsoft’s Global Demand Center (GDC) over the last few years is a good illustration of a modular approach.

Daily AI Roundup: The 5 Coolest Things On Earth Today

Microsoft’s GDC is a large scale, rapidly growing customer acquisition and lifecycle engine powered by an advanced set of marketing and sales automation processes and tools. One of its primary goals is to generate leads and provide insights for the sales team. As with any connected marketing and sales revenue generation engine, the challenge is to generate high-quality leads that convert to sales.

The processing costs of the lead and the number of salespeople available to follow through on them are material considerations. We needed a scalable, effective solution to prioritize and qualify leads for the sales team. As we considered AI-driven solutions to improve the lead qualification process, we needed to be mindful of how critical this function is. Radical moves that could put business continuity at risk were not an option. We also needed to consistently demonstrate impact to sustain funding and executive support given how much effort and attention had been invested in solving this problem in the past.

The system was ripe to benefit from Modular AI. Over a few years, GDC had accumulated data on millions of converting and non-converting leads. Initially, the organization relied on a rule-based lead scoring system, with scoring modifiers set by experts. GDC had improved scoring models via ad-hoc explorations of data. Analysts would adjust the modifiers based on their findings. These efforts were generally unsuccessful. Sellers spent roughly 70% of their time trying to get hold of customers. Their lead-to-opportunity conversion rate was below 4%.

Moving from a rules-based approach with humans analyzing hundreds of lead attributes and thousands of possible sales actions to an AI-driven solution where lead performance data would be continuously analyzed by a self-learning algorithm was the biggest improvement opportunity.

We saw positive momentum around AI and the initiative enjoyed strong executive support. We needed a carefully planned, modular approach for our AI transition journey to keep the momentum while not putting business continuity at risk.

Eighteen months into this journey, we’ve deployed six modules. We began with two foundational modules and when they were successful, we expanded the approach:

Module 1: Predict conversion to opportunity.

With this module, we trained an algorithm to score lead conversion by leveraging a history of leads that converted to sales. The resulting model considers 80 features constructed from lead attributes and actions and outputs a probability for each lead to convert to an opportunity. It uses these scores to prioritize a salesperson’s leads. The machine-learning algorithm learns from each new processed lead and updates itself every month.

Related Posts
1 of 433

Read Also: Mansfield Energy Advances Digital Vision Through Microsoft Partnership

Module 2: Validate customer intent to talk to sales.

An AI-driven email bot was developed and deployed to contact the higher scored leads to validate their intent to talk to sales. Leads that indicate interest convert the easiest, so we wanted to prioritize these leads higher if we could confirm their intent.

Module 3: Develop estimate of deal size.

Next, we needed to predict the most probable range of potential opportunity size for each lead. Smaller organizations were historically easier to convert. The first module learned this and started to deprioritize bigger organizations, resulting in significantly decreased opportunity size. The third module was designed to address this and, with the help of the next module, reoptimize the model to provide leads that maximized expected revenue.

Module 4: Combine measures to better prioritize.

Because leads vary in value, we needed to combine the capabilities of the first two modules to help sales teams predict the probability of a lead converting and its revenue potential. This model combined probability and deal size into a single prioritization score, optimized for the expected opportunity size.

Module 5: Individual lead score explanations. 

To get better buy-in from our sales people, we needed to do a better job of providing explanations. The old, human, rule-based model had a transparency that enhanced users’ buy-in. This module generates text that explains the top influences that went into each lead’s score. Having this transparency dramatically improved how the model was perceived by the sales community and increased trust in the AI.

Module 6: Better prioritization of open sales leads.

Finally, to improve adoption further, we addressed prioritization on an ongoing basis. The sixth module considers ongoing sales engagements to date (phone calls, emails and their outcomes). It expands the training dataset for our model from just marketing to both marketing and sales interactions and increases the accuracy of prioritization for our most important leads, the ones that sellers are already working with.

What was the impact?

Adoption of these modules led to a threefold increase in lead-to-opportunity conversion and equally impressive growth of booked revenue. A modular approach to AI also gave Microsoft the flexibility to tackle the most impactful, low-hanging fruit first, and then solve additional problems to enhance what was in place or adapt to business process changes. GDC continues to evolve its AI as it moves to calculate customer lifetime value and cover more stages of the end-to-end lead process.

It’s reasonable to start simple. Automate current efforts or augment current methodologies with AI to make your business a bit smarter. In many cases, even simple changes drive a lot of positive impact. Keep the discussion centered on how you gradually improve your AI to address a broader set of issues and investment centers for more efficiency and effectiveness. Modular AI approaches allow for more flexibility in picking the most burning or highest impact issue first so that later iterations can focus on secondary issues and process improvements. It also releases solutions faster. And the faster your teams adopt them, the sooner you can reap the rewards.

Leave A Reply

Your email address will not be published.