If you saw the news of Salesforce’s acquisition of Apromore, you might be asking yourself some questions. Specifically, what is process mining, and what does it have to do with Salesforce? Process mining is a decent-sized segment of technology that many who work with Salesforce have never heard of. Fortune Business Insights placed the size of the process mining market at $2.46B in 2024, with projected growth to $42.7B in 2032. And no wonder. This technology has been quietly finding a niche in some of the biggest businesses in the world since the late 1990s.
In this piece, I’ll give a very brief overview of process mining. I’ll also share some personal reflections on why I think process mining matters deeply to Salesforce’s AI strategy. Finally, I’ll lay out some of the challenges Salesforce (and more broadly, Salesforce customers) will face in bringing process mining technology to Salesforce implementations.
What Is Process Mining?
Process mining exists where data science and business process analysis meet. Business process analysis seeks to discover the processes of a business or an organization for the purposes of refining how that business does its work.
It is largely done through observation, interviews, and then the work of refining that down. Data science is, of course, the broad umbrella for any technology that seeks to discover, analyze, and draw new insights from data.
“Data isn’t units of information. Data is a story about human behavior – about real people’s wants, needs, goals, and fears.” Daniel Burstein
But what is the data of business processes? Well, it’s the historical data around that process. Whether that is customer service, a sales process, order fulfillment, or whatever, each of those actions, taken in IT systems, leaves traces of what happened in data. In process mining, this data is referred to as an event log. Using the historical activity data of an event log, process mining typically seeks to accomplish one of three tasks: discovery, monitoring, and process improvement.
This idea, that event logs can be used to discover the real business processes through data, was first proposed by Wil van der Aelst. The discipline of process mining grew its own standards body, the IEEE Task Force on Process Mining. Through exploration, research, OSS projects, and products, it has grown into what it is today. As the IEEE Process Mining Task Force put it: “Discover, monitor, and improve real processes by extracting knowledge from event logs”.
Apart from Apromore, other process mining heavy hitters include Celonis and UI Path. ServiceNow offers process mining. There have even been Salesforce ISVs who have invested in building process mining for Salesforce, such as Hubbl and Processity.
Elements.cloud has also sought to build something similar – configuration mining, as they call it – which seeks the same outcome (understanding business process) with a different data source (metadata).
Why Process Mining Matters to Salesforce
A prominent member of the community asked me what I thought Salesforce’s play was here – where will they get their ROI from this investment? First off, it goes without saying that execution here is key. The process and automation team has been handed a huge weapon to wield in making automations more powerful than ever.
Imagine you have collected and mined data about your high-value customer onboarding process. As an admin, you decide to create a new screen flow for your account reps to use. By describing the requirements and providing the mined process maps, what if Flow were to generate a process based on actual data about how your reps work?
Take that a step further, and imagine once the screen flow is launched, a semi-autonomous agent that observes the process and proactively suggests alternatives to the sales rep – not because of some emergent intuition that the agent imagines, but because it fundamentally understands the process based on what has been mined from your business.
Let’s be clear, this is all prognostication on my part. But if Salesforce can help its customers fundamentally understand their own business processes with a data-driven approach, this would be a huge value add.
A true-to-life process could solve one of the biggest challenges in the current state of AI agents: hallucinations and the high-sounding-but-in-reality-not-so-great 93% accuracy claimed by their own Agentforce implementation on help.salesforce.com. Just two weeks ago, in the True to the Core session at Dreamforce, Chief Product Officer Steve Fisher intimated that consistent agent outputs have been a challenge.
“These AI things, they don’t always exactly work the way that you think they’re gonna work. That’s the lesson that we’ve been learning, and relearning, and relearning especially the last year.”
Is that the candid statement of the head of a huge product organization looking for a solution to agent grounding? Perhaps Apromore and process mining are the hoped-for answer.
Challenges With Process Mining and Salesforce
But it isn’t all straightforward. Some of you may be asking yourselves, “If process mining has been growing since the early 90s, why doesn’t it appear to have ever made headway in Salesforce implementations?” This is the right question. The answer is: incomplete and inaccurate source data.
As in all things data, the source data makes a big difference in the outcome. The cliché “garbage in, garbage out” holds for process mining, too. In many systems involved in process mining, there are very good, complete sources of historical change data. Salesforce is different.
Firstly, field history, the historical record of change, is very limited as far as the number of fields (20). It also does not track old and new values for long text and multi-select picklist fields. Finally, it is inaccurate, tracking precision only to the second. This complicates process mining and can create obstacles to creating the full picture of processes. You can improve the number of fields tracked with Shield Field Audit Trail, but you still can’t track every field, nor can you improve the precision-to-the-second problem.
Apart from data, there is a talent problem. Process mining is its own discipline and uses complex algorithms to associate individual field changes into semantically meaningful activities. Presumably, the expertise of Apromore’s employees and ecosystem will help, but initially (at least), there will be gaps in knowledge for how to transform Salesforce source event log data into meaningful process data.
Neither of these is an insurmountable problem. Salesforce itself has a potential answer to better logging data in the form of Own’s continuous data protection, which could prove a much richer data source. But this uses Change Data Capture, which can cost additional license fees. Will Salesforce bundle CDC to enable customers to have better data?
Processity also solves this problem with Processity Data History. But will Salesforce customers want to pay for an ISV solution in order to get the process mining data they need or want?
Final Thoughts
Process mining is not well-known or understood in the Salesforce ecosystem despite its large appeal with other enterprise software systems. Salesforce’s own limitations as relates to process mining source data appear to be a key cause.
Only time will tell how Salesforce solves this problem in the wake of its acquisition of Apromore. But what’s certain is that without some further investment, Salesforce may encounter problems in bringing process mining to its customers.
