From System of Record to System of Action: Using Automation and AI to Unlock CRM
At a large technology company, 1,500 spam and abuse cases arrive every day. Before automation was in place, 10 support agents could manage approximately 300 to 400 cases each day. Each case required three to four minutes to complete. For every case, agents had to read the complaints, look up the IP address blocks, suspend user accounts, and send closure emails. This was a tedious process. It took time to complete a small part of the work. The volume was huge. As a result, many cases remained in the queue well into the night.
The real issue was not that the tasks were difficult. The tasks must be completed each day without fail as the tasks cannot be missed. Every CRM system ends up with the kind of work cases that keep coming back over and over again, even though they do not require much skill. Even simple tasks require someone to make a decision. That is what makes them take so long.
The above processes are added to hundreds of instances every day for rotating staff. This results in delays, missed steps, and a CRM that only records what has occurred. It never speeds up. This is the problem with systems of record: they write down everything but make no decisions.
A recent implementation on a CRM platform demonstrates the power of closing that gap using both artificial intelligence and robotic process automation, and not just one of them.
- AI handles judgment tasks: A model built into the case management system now analyzes every case on arrival, detects whether it is spam, reads the sentiment, summarizes the case, and updates the CRM record with this information in seconds.
- RPA tackles rule-driven processes: A bot accesses the network security console in the same way as its human counterpart to analyze the IP address mentioned in the ticket, and, if necessary, blocks it, also on a schedule.
The second phase of automation was for the account suspension process. This is particularly noteworthy in that one back-end account management system is accessed directly via API in parallel, while another is manipulated via an RPA bot that has been appointed administrator:
With the back-end account management system, the account suspension process is invoked directly through API.This system is accessed via an RPA bot that acts as an administrator for the account suspension process.
The decision regarding which is selected depends on the ownership of the customer's account. After resolution, the case is automatically closed, transferred to a different automaton for audit purposes, and the customer is sent an email notification, which is also selected through automation. Every status change generates an updated ticket that is visible in the interface and is used to produce automated analytics dashboards.
Thus, the entire process works in a continuous loop 24 hours a day, with AI determining the necessary actions and API/RPA automation performing them.
The business case for these changes can be found in several places, many of which can be quantified for your own instance post-launch, pre-publication:
- Resolution time: The process time has been reduced from three to four minutes of agent work per case to an automated end-to-end solution that finishes within a few minutes.
- Coverage: About 1,500 cases come in every day. These cases are split into two groups of 750 each. One group arrives in the morning. The other arrives in the evening. Automation takes care of all cases from beginning to end. Every case is handled automatically. There is no need for an agent to check anything. There is no work involved at any point.
- Consistency: Every case follows the AI, API, RPA, closing, and logging steps, which reduces differences and missed actions.
- Time saved: Handling 1,500 cases used to take 75-100 hours of manual agent work, and now that time can be moved to higher-value or more complex tasks.
None of this requires abandoning CRM, as it is leveraged here as the coordination layer for the next steps, wired into the systems carrying them out.
Under the Hood, in Plain Language
Stage one is AI: a prompt based on the case subject and description, which is interpreted by an AI model and placed back on the record as a structured verdict of spam/not spam, sentiment, etc. This is the only stage where anything approximating thought happens; the rest are deterministic processes.
Stage two is RPA: a software robot, on a schedule, which logs into the network security console using its own credentials, reviews each flagged IP address, and blocks any that are not already blocked; in other words, performing the same click tasks a human would do, but automatically and without supervision.
Stage three, account remediation, is where things get interesting in terms of orchestration and where AI passes off to RPA and API. Based on the back-end system where the user account is located, the case gets routed, using either an API call or an RPA process, to either User Directory Platform 1 or User Directory Platform 2. The former has an API; the latter does not; the RPA bot must log in to the console of User Directory Platform 2 and perform tasks as if it were a human. No one working on the case needs to know on which directory the user is; the CRM does.
Stage four is closure and communication: after steps three and other required steps have been completed, a scheduled process closes the case, routes it to a queue for auditing, and sends an email to the user, which email, based on the steps taken, is selected from a series of templates — all native processes in the CRM, requiring no AI or RPA.
Stage five, logging:which is really where everything takes place, is where the status updates from both the AI and RPA steps get logged on the case record, providing visibility and fuel for dashboards.
What Changes, Concretely
Before: A case waited in a queue until an agent became free. The agent would look at the case and check if it was spam. The agent would open tools, such as the security console and account management systems, to decide what to do. The agent would write a closing email by hand. All of this work depended on whether an agent was free. With 10 agents each spending three to four minutes on each case, the team could finish 300 to 400 of the 1,500 cases that came in each day. This left a pile of cases that remained from day to day.
After: Cases come in two groups each day, approximately 750 in the morning and 750 in the evening. When a case opens, AI sorts it away. Robotic processes and API calls run on a set schedule, so they do not wait for anyone. The customer receives an automatic message when the set rules are satisfied. The entire trail of actions can be checked later through a case history. No agent needs to touch a case at any point. Because of this, the time needed to finish a case dropped from three or four minutes of work to a fully automated routine. The backlog that used to spill over business hours disappeared. The 75 to 100 hours per day that agents used to spend on triage is now free for more important tasks.
Where This Kind of Automation Goes Next
The pattern is simple and clear: AI handles judgment, RPA and API automation handles action, and the CRM is in the middle, ensuring that the correct tool is used at the correct time. This division of labor is essential, as a CRM alone cannot perform actions in another system, such as logging in to a legacy administrative console, and AI alone cannot perform these actions. The two work together to have a much broader scope within a business. Possibilities include adding in RPA or API coverage for other back-end systems as the business grows, shortening the feedback loop between case creation and resolution, or using similar AI triage for other high-volume queues that require human judgment, such as general complaint escalations, security incidents, or risk of renewal issues.
A more interesting discussion is about philosophy; a case management system that can read a case, interpret it, and make decisions about what actions to take to resolve it is much closer to a copilot for the team, allowing them to focus on the cases that need human involvement. A CRM that merely records cases is merely an archive; however, one that can truly read and understand them is something closer to a valuable employee.
Charudatta Thute is a CRM and enterprise technology leader and founder/CEO of Vegacompass Technologies. Pradip Pawale is a CRM developer and consultant with Vegacompass.