AI: Attempting to extend human intelligence
The use of artificial intelligence (AI) or machine learning (ML) is increasing across all industries to augment data and rule-driven tasks including automation, deep analysis and decision making.
Simply put, AI systems are trained with defined rules from a known sample to make predictions for a new, unknown sample. The technology is the advancement of simple decision-making software that now has significantly more on-demand compute power with large, complex data sets.
However, today’s artificial intelligence is still error-prone and unable to evaluate context beyond data sets — and has a long way to go before it can completely replace certain human activities.
The inherent inaccuracy of early-stage AI
AI systems learn from a defined set of “rules,” or classifiers, set by an IT team. These rules allow the technology to pull in certain data, evaluate it and create an output based on that evaluation. Since AI is unable to perform logical reasoning and interpret context beyond these established rules, results are often skewed.
An AI program, or even a product feature with AI, quickly becomes too advanced for organizations lacking a large team of data scientists to support it. This team must continuously train data models, run hypotheses, validate results and remove false positives and false negatives to ensure accuracy. As referenced in an article from labsix, false positives are common even in more advanced AI systems.
With all this in mind, larger corporations with the proper infrastructure — like IBM, Microsoft and Google — are taking a step back from their AI solutions.
As stated in a white paper from cybersecurity company Cylance, “the efficacy, interpretability, and robustness of the [AI] model all hinge on the features.” Features refer to the number of properties taken into consideration by a model. Most companies claiming to use AI or ML are working with first generation models, which contain over 1,000 features. Even large organizations doing AI “right” are only in the third or fourth generation (1M-3M+ features). Fifth generation, the highest possible, has unlimited features — but no one has achieved this yet.
In summary, it takes a significant investment in technology and people to effectively leverage AI beyond having a technology in your architecture.
AI in the sales enablement space
For the last several years, sales technology providers have been drawing in customers with claims of using artificial intelligence for things like forecasting and lead scoring. Sales enablement vendors use these claims as well, with platforms “powered by AI” that automate processes for sales video assessments and content recommendations.
Many enablement solutions largely use AI to relieve sales managers of the task of providing feedback for their team’s practice pitches. The AI evaluates a video pitch, and is programmed to award points to sellers who perform certain actions or use certain words; however, there is no context for them. For instance, a rep could have a great smile, tone, and cadence but use technical terms out of context with no business value or coherent meaning, but still score highly by the AI system. Enablement managers and data scientists still need to review and interpret the data gathered by the AI and make changes. And even this simple use of the technology is mistake-prone. For video recordings, AI is only as good as the transcription services, which continue to improve, but still have a 1% error rate. Artificial intelligence in fact creates a gap in the most important element of a manager’s job: investing in their teams and providing personalized coaching to every individual.
Another promise of artificial intelligence in the sales enablement space is to recommend content for sellers to use in a meeting or other interaction based on the prospect’s industry or previous experiences with similar prospects. Again, these recommendations come with minimal context; the AI is simply following the rules, which may be based on keywords for demographics (industry, company size, company location, etc.). What is missing is the intent of how the content can be best used as designed by marketing and enablement professionals based on feedback from sales teams and customers.
With the technology where it is currently, sales teams that want to develop both their reps and their personal relationships with buyers cannot remove the critical human element of business.
Where do we stand when it comes to artificial intelligence?
Showpad, like other SaaS providers, was interested in incorporating AI into our solution. We piloted with an AI vendor but quickly discovered that the technology was not where it needed to be to empower good decisions for our customers, and have ultimately decided against any strategic investment at this time.
This isn’t to say we don’t want to do it. We just want to do it right, which will take time. We’ve developed a roadmap for creating AI-driven solutions, which starts with administrators creating rules that define human tasks, from which computers can learn and automatically apply in the future. The goal is for this to move to rule-driven processes: creating configurable rules to enable statistical learning and making predictions; then to AI-driven processes, where computers and systems create explanatory models for the real world. Realistically, achieving this is still years away.
Kick the tires on AI solutions
As you research tools for sales enablement, engagement and operations, you’ll likely see AI being touted as a feature. It’s important to ask: What are you really getting in terms of accuracy and efficiency? Does the company simply have an AI feature they have added? How many data scientists are defining feature sets and models, and how frequently is that model updated? What are their false positive and false negative rates? Are any of their claims backed by a third party?
Artificial intelligence is a cool and powerful tool with plenty of potential. But it is still evolving, as we see some of the world’s largest technologies struggle for strong AI use cases. The biggest question: Can AI replace unique experience of experienced sales managers when it comes to enabling and coaching sales teams?