The Next Wave Of AI In The Enterprise Will Not Be Sexy, But The Impact Will Be Massive
This story originally appeared on Machine Learnings, an excellent newsletter curated by Sam DeBrule, in July 2017
Consider this: How long did it take you to file your last expense report on Concur / Workday? Were there a lot of drop-down menus to for categorizing expenses? Or manual data entry to provide calendar, attendee and other context around the receipt? Did anything on your report trigger a follow up from your expense department?
How much time did you spend filing that report and dealing with follow-up requests? What does X hours of your time cost your company? What if you multiply that cost across your employee base?
What does your company spend on an expense management team that is manually reviewing those reports, and how effective are they?
And yet: All the data you were keying in already existed somewhere. For example, your calendar has all the information on where you have been and who you met with. For sales and customer success employees, any client-related expenses can be contextualized and mapped via other systems of record (Salesforce, etc.). So why are we manually entering data?
And those dozens of people manually reviewing reports for exceptions or abnormal behavior? Today their intervention is triggered by static rules such as dollar thresholds. A calendar and CRM overlay could automatically determine if a $700 meal is out of policy, or a (perfectly acceptable) 10-person dinner with a key prospect.
The size of the prize: Concur was an $8.3B acquisition by SAP, making it the largest Software-as-a-Service exit to date. Concur has no A.I. capability.
However, Concur and other incumbents are sitting on the ultimate machine learning data set — every expense report ever filed within their customer base. They treat that data as inputs to static workflows, where the output is a PDF and some aggregate analytics. They do not treat that data as a dynamic graph, where they could automatically define spend baselines by employee / department / geography / vendor / business context.
The A.I. startup that figures out the spend graph could make manual data entry and manual report review a thing of the past. Detection of fraud of anomalous spend would be automatic. That startup could redefine this $2.5B market (Source: IDC).
The big idea: This piece is not really about expense management. That merely represents an example of the wide range of valuable “back office” enterprise applications that could be reimagined with A.I.
To date, most venture funding for A.I.-enabled enterprise apps have centered on the “front office”. Marketing tech has received $13B in VC investment over the last 5 years (Source: CB Insights). In the customer experience category, leading companies like Gainsight* have raised $156M (Source: Crunchbase) to help B2B companies have a single view of account health and power “next best actions” for churn reduction off of that (previously unavailable) insight.
However, for a prospective founder today, finding white space in the “front office” is daunting. Marketing tech alone has over 5,000 venture backed companies (Source: Chiefmartec.com). Meanwhile, the “back office” is still largely untouched by A.I.
Ambitious founders will change that. Let’s consider the defining characteristics of both Gainsight (an emblematic front office example) and expense management (as a back office example) in the abstract:
- Large, pre-existing end user activity or area of technology spend
- Gainsight — the customer success category did not exist, but the underlying operational pain and activity level was ubiquitous.
- Expense management — we have already covered some of the huge companies built in the space.
2. Disparate data stores that can be stitched together to identify optimal actions
- Gainsight — they stitch together any source of customer data (CRM, billing, support tickets, usage logs, surveys, etc.).
- Expense management — multiple data stores (calendar, email, CRM, historical expense report data) are largely unused today.
3. Incumbents that are BOTH technologically dated AND represent a minority of the data sources for an A.I.-enabled solution, mitigating platform risk.
- Gainsight — incumbents like Salesforce have a critical mass of customer data, but lacks key signals like product usage.
- Expense management — incumbents like Concur have historical data, but do not mine that data today and do not integrate with other signal types.
Said differently, market and customer dynamics support a differentiated solution by an A.I.-enabled player. And a wide range of back office categories have those same characteristics.
- In HR, Applicant Tracking Systems are a major category, but existing players treat resumes as flat files and meta-tags, not a talent graph.
- In security, CISOs have invested millions on multi-layered defense, but suffer from operational overload and alert fatigue. They need automated, intelligent triage.
- Companies like Bench* (SMB bookkeeping) and Signifyd* (e-commerce fraud prevention) are pioneering machine learning in the CFO’s office.
Five to seven years from now, analysts may have a back office tech landscape with thousands of companies in it. That means the time to attack this prize — and get a head start — is now.
Disclosure: Gainsight, Bench and Signifyd are #BackedbyBCV.
Originally published at machinelearnings.co on July 17, 2017.
