I’ve spent the last decade doing a mix of management consulting and software development – sometimes selling engagements in specifically one or the other, and sometimes doing both in parallel as part of a single project.
A question I’m profoundly interested in is “How much of the skillset that we pay management consultants for could be automated, and for what cost?”
The answer to this question is consequential. Consultants represent the heavy hitters of white-collar knowledge work – the people that companies hire to make big changes happen. They typically charge two-to-three times the median salary paid to employees performing the same tasks, and they bill themselves as having superior skills and a larger breadth of industry knowledge than what you’ll find in the job market.
So for automation, entering the management consulting market represents a high performance bar to clear. You’d be substituting a narrow AI for human beings who’ve been selected for their high analytical intelligence and communication ability. It also represents a lucrative opportunity – essentially bringing the marginal cost to produce a luxury good (management consulting) down to zero. Because of the nature of bits, automation software introduced to compete with consultants, once established, would naturally extend to all the white-collar job roles that management consultants tend to displace: jobs with titles like analyst, project manager, strategy advisor, and researcher.
A good starting place for exploring automation is to map out the typical interactions that a business leader (a client) has with a management consultancy. Off the top of my head, I’d list: scoping out a work engagement and negotiating fees and terms, preparing and sending data files to the consultant, reviewing findings on your own company data and how it relates to market trends, choosing from among a list of proposed alternatives for moving forward, collaborating with the consultant to communicate the new strategy within your organization, and measuring and implementing the new strategy.
I believe the middle chunk of these interactions will lend itself best to automation, but we can begin to make progress on the entire set.
A framework I use to evaluate any potential new product is to ask myself under what circumstances of quality and cost would my purchase be a definite “yes”, even with a company with which I had no prior experience. In order for me to give an emphatic “yes” to any software purporting itself to be “automated management consulting”, I would expect the following:
- I would expect to get at least 80% of the way to the deliverable produced by a flesh-and-blood management consultant
- I wouldn’t be willing to pay more than 2-3% of the cost of the management consulting equivalent
- I would want the ability to see and edit any particular piece of the completed analysis (just as I might ask for the backup excel model or editable powerpoint slide deck from a consultant). I would be more skeptical of the quality of the analysis than I would of the analysis purchased from an established consultancy
- And lastly, I would expect the technology to produce results instantly
Are those conditions achievable?
Over the last two years, I’ve been coding out an AI platform that meets these requirements for a sub-section of the management consulting skillset: specifically, three components of management consulting engagements that lend themselves well to automation:
- Extracting, consolidating, and interpreting client data
- Running trends analyses on client data and making observation and recommendations
- Writing reports and providing guidelines to institute new business practices
The remainder of this post is an exploration of some specific products and algorithms I’ve developed. They are live and free to explore at https://go-contessa.com.
Gender pay equity
An analysis of gender and pay is a good test case to see how well automation can hold up. It’s a sensitive topic, subject to nuance, and it’s an area where junior analysts who aren’t familiar with statistical significance and p-hacking are likely to lead themselves astray. That can lead to unproductive conversations and lot of internal division and angst.
The automation algorithm’s responsibility is to take one or more data files from the user, to interpret them based on their contents, and to generate robust and useful insights in the form of an editable report with backup Excel workbook.
Below is the output. You can see both a simple UI for submitting data files and tracking the algorithm’s progress, and the resulting .pptx file as opened in Powerpoint. Just drag and drop a workbook, csv file, or json file from your HR system and wait 30 seconds for the numbers to and the files to be created. See below.
The algorithm that interprets incoming data files has been trained on sample data that I’ve gathered from clients over the years. As it scans the data files, it comes to conclusions such as “this data field probably represents a horizontal grouping of employees, such as department or cost center; it will make a good dimension to combine with job level or salary range as a means to break out employee cohorts for relative comparisons”. Once you identify the data fields in these ways, you can get pretty detailed in the analysis.
The algorithm writes a .pptx report file that includes both general info that is included in every report (“gender pay equity is a sensitive topic; let’s be precise in our language and be both open to different perspectives and forgiving others”), as well as particular callouts (“It looks like these particular 3 job roles in the organization where individuals show a statistically significant correlations between gender and base pay. Here’s some next steps to explore further”).
Sales pipeline analysis
Over the years, I’ve done consulting engagements to help around 20 smaller (<100 employee) Oil and Gas organizations transition into using CRM software for their sales teams. One key component to getting the most out of a CRM is to do regular sales pipeline coaching with a manager to identify areas of improvement and trends in the market that could lead to a more targeted and effective sales cycle.
I used a similar process in coding out this algorithm. The purpose of this automated report is to give salespeople and their managers a quick sense for the state of the pipeline, and some tools to assess gap to goal.
The analysis includes a snapshot of the sales funnel, top open opportunities, trends in products, opportunities past estimate close date, and a handful of other metrics. The specific analyses are based on what data is provided and what the algorithm is able to interpret.
One of the areas I’m most excited about is using automation to identify trends and make intelligent recommendations. This is where you take a mix of pure narrow AI data analysis that doesn’t know anything about the domain of the data, (“it looks like items tagged with this value in this field are larger in this other numeric field”), and some actual heuristics learned from years of consulting (“15% – 30% of customer accounts often represent more than 80% of sales. Are we appropriately segmenting customers and opportunities and pursuing share-of-wallet where we know it exists?”).
There are dozens of hand-coded placeholder observations in the algorithm that may or may not trigger based on what the data analysis finds. See the below call out of high-margin and low-margin product segments in the pipeline.
Going beyond automated reports
The sales and gender pay equity reports represent a similar pattern: take information already tracked in your systems and produce meaningful analysis instantly.
I’m interested in exploring how automation can be used in other ways as well.
One example is the creation of web apps. Let’s say you have a report or piece of information you’d like the team to monitor. The time cost of having a software developer code the logic of the analysis in javascript or in a data product like Tableau or PowerBI is one measured in tens or even hundreds of high-value hours.
For this leaderboard app, you drag and drop a data file, and get access to a hosted web app with a viewer password and an admin password. Share the link and viewer password with anyone who needs to monitor it, and use the drag and drop interface to dump a fresh data file whenever the analysis updates.
Another area with huge potential payouts is the extraction and consolidation of information from many files. In many situations, a business has MS Office files created manually or received from many sources, and there is a desire to consolidate them into a database. It’s difficult to find a business analyst with the hours needed to make it happen. What you need is an effective algorithm that can do the task instantly.
So far, I’ve coded out intelligent data extraction for .docx and .pptx files. The algorithm looks for patterns across files (the text contained in a red smart-art shape in a similar position on the page across many files, or the text contained immediately after a large-font bold header of “Next Steps”) and then extracts the text into a single file. In the example shown below, I dumped thirty MS Word files that contained job description information into the UI, and the algorithm categorized the various data elements (“Duties”, “Required Competencies”, etc.) and put them into tabular so we could upload them into Workday.
A new frontier
I think we have the potential to automate a meaningful fraction of management consulting within five to ten years. The software developer and management consultant time-cost of developing any of the automation algorithms depicted here is in the tens of thousands of dollars, and so if you were to price any of these products at $5 to $20 per analysis, you’d have a break-even point on the order of 1,000 to 10,000 units sold. This is not trivial. Not every analysis would be worth automating. It’s also not prohibitively expensive. How much of the workday of an employee with a title like “business analyst” could be automated in this way? My current hypothesis is somewhere between 30% and 50%.
I created a company to make these algorithms available to the public and to solicit feedback on other analyses and workflows to automate. You can try out any of the products depicted here at https://go-contessa.com. Please take a look and let me know what you think.