What the Salesforce layoffs say about Machine Learning

When I started at Salesforce in July of 2017 the stock price was around $87 and headcount was about 25,000. When I left the company in May of 2021 the stock was at $240 and there were more than 50,000 employees.

That’s huge growth, but between 2021 and 2022, headcount grew by a whopping 30,000 even as the stock fell to $130 in January this year after starting 2022 at $250 a share. By now you know what happened and a lot of people were unfortunately impacted, due to the company hiring in advance of demand that never materialized.

I have no special insight since leaving Salesforce in the Spring of 2021 to join Alteryx. However, as I noted in the story about my Salesforce journey, I suspected for a long time that the company’s inability to help customers realize the value of machine learning (ML) was a problem in the making.

You don’t read or hear anyone talk about Einstein much anymore, but it was both a character among characters at Salesforce and the tag given to anything analytics related — machine learning or otherwise. The rapidly escalating number of predictions generated by Einstein in Salesforce Clouds were often the first words spoken by Marc Benioff in earnings calls.

That’s probably no longer the case.

The hypothesis driving the Einstein strategy was that machine learning, packaged as AI for marketing purposes, was a means of driving added value and differentiation within and across Salesforce Clouds.

Data science in 2015 and 2016 was just getting its legs and was very challenging for most companies — especially the huge number of small to medium sized businesses that Salesforce counted among its Trailblazing customers. For some organizations data science is still a mystery.

The added value and differentiation stemmed from packaging the benefits of ML without the usual constraints of expertise and bringing model results into a production system.

For use cases like lead scoring that could operate and learn on data types shared by thousands of customers, this worked very well. For other use cases like recommendation or personalization, results varied based on a customer’s understanding of the use case and capacity to apply it to their business given a whole slew of considerations.

In a nutshell, success was much more likely when the consequences of a use case done poorly were relatively small (less risk) and the customer could rationalize what was happening alongside other analytics work in the business or within its governance model.

Thus…while Einstein utilization escalated every quarter (per Benioff) I came to think that the decisions he was citing didn’t add up to a ton of measurable value for customers.

As proof, all I had to do was look at the customers I managed within the Customer Success Group — all name-brand retailers you know. While some had a decent grasp or willingness to learn, for most, the value of machine learning didn’t register, or the use case was already served by a third party, or a data science team had it handled.

I was experiencing the reality that most organizations fall somewhere in the middle on a spectrum, from relying on a single platform to run their business (like a small private company) all the way to the most complex global enterprises.

The small business owner might very well be served by packaged Einstein functions, whereas a large global enterprise has no use for it due to having the expertise and staffing for analytics you expect. For the massive middle ground where most companies find themselves, they faced a lack of understanding, skills, resources, strategy, and sponsorship. Basically, very analytically immature.

Given this landscape, it was no wonder that adoption of Einstein AI lagged other Cloud functions positioned as features you could easily implement, use, and administer. It would make no difference if not for the fact that it was how customers applied analytics to their Salesforce Clouds that ultimately drove the greatest value.

From baselining business metrics and targeting weakness for improvement, to road mapping what was necessary to implementing and tuning a use case, this was how I believed Salesforce customers could realize the most value from their investments. If Salesforce didn’t provide this, some other company would (and usually did). It happened to also be how Salesforce could hold persuasive sales conversations with customers about investing more.

Salesforce wasn’t in the business of advising customers how to approach machine learning projects, what resources are usually necessary, common risks, and how to realize use cases within a Salesforce Cloud. Few Salesforce partners were either, as they earned money doing straightforward implementation and training work, or the occasional chatbot setup.

This explains why ML adoption never took off among Salesforce customers I observed, but it also reflects customer growth and attrition issues I noticed at the time. I sensed back then and to this day, that many Salesforce customers realize they have a significant opportunity to improve through analytics like machine learning, but remain frustrated by a lack of progress.

It’s water under the bridge, but I would have liked to see Salesforce develop a three-tiered model for analytics, AI and ML.

The lowest tier would have been the mass of small and mid-sized customers happy with low impact/low risk use case support but with aggressive prescriptions for measuring value from a baseline. Any customer might benefit, but these in particular would given their lack of resources and reliance on essentially a single data source.

The next level was the largest cohort of organizations that needed all manner of help, training, partners, and support for their journeys. The topmost tier was those global enterprises that wanted to know precisely how to integrate bespoke machine learning models and development processes into Salesforce Clouds.

That’s a lot of work for an unknown payoff, but as an analytics guy myself, I think the payoff is certain if it’s managed correctly relative to expectations for selling a SaaS application to a non-quantitative worker and their manager.

This leads me to Alteryx, that sees customers of all varieties but often the leaders and teams of analytics professionals supporting daily and strategic decision making, probably working right alongside Salesforce users. For them, ML may be mysterious, but it’s part of their curriculum as professionals.

I think the market for Alteryx demonstrates the specialization required for analytics, but also that the competency is highly valued. Expertise may not live in a Center of Excellence or Data Science group, but in the teams who take up the challenge of supporting decision makers in business functions like supply chain or finance.

For them, applying low or no code tools like those in the Alteryx Analytics Cloud, is a faster and more efficient means to an end that includes time for productive collaboration with decision makers, to understand their urgent priorities, and how to support them with the right data and analytic methods.

Machine learning may be in scope while in other cases more timely access to better data is the priority. Requirements will vary but what won’t change is the need to be ready to help customers at any stage of their analytics maturity journeys.