Many organizations have begun to recognize the benefit of using analytics to support decision-making. At least in the abstract. But how to get the capability is not so clear. Do you hire a data scientist? Do you build your own analytics team? Or do you just bring in the high-powered consultants for a quick tactical mission?
Before we get too far, let’s make sure we answer the first question: Do you really need your own analytics capability? Not every organization does.
Consider these questions:
Is your environment highly competitive? Or are there natural or structural boundaries between you and your competitors.
Can you envision more than three projects where analytics could help? Or do you just have a single question you need answered.
Is your problem one where people will continually adapt and respond to your actions (for example, fraud reduction)? Or is it a static problem where the model will stay the same.
Alright, you’re still here, so you must think you need the capability. There are a few different options, but before we explore them, let’s consider some additional questions:
Is your data ready to go? Your data is almost certainly worse off than you think it is -- unless you’re regularly using it. If you’ve never looked at your operational, customer, or transaction data, then don’t expect quick results. And you shouldn’t hire a data scientist to clean it up.
Do you have IP to protect? Are you in a highly competitive environment? Do you have customer or operational insights your competitors don’t? If so, then you want to keep your data close and your analytics closer.
Does the expected value exceed the expected cost? Analytics capability is expensive -- sometimes prohibitively so. What are you willing to spend to get better decisions? How long till it pays for itself?
How soon do you need answers? Are you willing to wait a year for results to start coming? Or do you need to justify your investment with tangible results in the current quarter?
Let’s look at a fraud reduction group in a benefits provider as an example. How about Acme Benefits. We might expect the following answers to the questions:
The data is a mess. The systems were set up to pay claims and report to customers. Analytics was never considered.
Developing analytics to reduce fraud would be a huge competitive advantage if it were successful. Accurate algorithms shouldn’t be shared through crowd-sourcing.
Fraud analytics is one of the most valuable applications of machine learning. Tens or even hundreds of millions of dollars are on the table and the cost is a pittance compared to this.
Acme is skeptical of analytics and wants to see some actual results in the first quarter.
Looking at Acme’s results, they’re going to need to look outside for help. The time to clean up their data and to demonstrate value precludes them building from scratch. But outsourcing is problematic as well. There will be ongoing work for years to come and paying that all to a consultancy doesn’t make a lot of sense.
Thankfully, the choice isn’t simply to rent or to buy. We’ll look at four in this article: insourcing, outsourcing, co-sourcing and crowdsourcing.
Note that there isn’t a category for automating. There is no software that will magically give you analytics capability. If a software vendor is telling you this, then show them the door.
Many organizations build analytics capability the traditional way. They have lots of data with obvious insights hidden within. They can justify the investment on value, but there’s not huge time pressure to get it done. So they hire themselves a team. Let’s call this insourcing.
Ideally, they start by recruiting a data science cowboy who can do a little bit of everything. The cowboy then hires his own team of data engineers, information designers, and analysts.
In the most successful organizations, the analytics group has the five roles, their leader has a seat at the executive table, and the team has a mandate to cross functions. They can respond to requests from other departments or they can pursue their own ideas. They have access to the entire organization’s data and they have influence at the highest levels of decision making.
This model is often called an Analytics Center of Excellence and it operates as an internal consulting group advising the organization over the long term. Team members develop company-specific domain expertise and gain the trust of the organization. Airlines, cities, and sports teams have embraced this concept with stunning results. In the best case, they help change the culture of the entire organization to be evidence-based.
So what’s the downside? First, there’s no guarantee that your team will be competent. Anyone can put “data scientist” on a resume and there is a huge difference between academic competence and real-world competence. Most organizations are incapable of differentiating between a skilled analyst and a charlatan.
Furthermore, it’s going to take time for this group to start showing results. They need to learn about your business, clean up your data, and build some credibility.
Finally, once your group is up and running, it’s like any other department or division. It doesn’t have to compete to stay in business, so it may get complacent and bureaucratic.
The next obvious option is outsourcing. The traditional consultancies all have analytics capabilities -- or at least claim to. (Competent data scientists are difficult to keep around and often don’t play well in the traditional consulting model). There are also a number of new firms that specialize in analytics consulting.
Some firms are functional specialists with specific expertise in marketing analytics or HR analytics. Some are built around a specific industry like financial services or pipelines. Still others are built around a specific capability like machine learning or data visualization. As with any vendor, due diligence is a necessity.
You’ll want to outsource when you need results quickly but intermittently -- perhaps a four-month project every year. A competent consulting group will set a high bar for the analytics, but will also understand education, adoption, and the difficulty in transitioning to an evidence-based culture. The vendor will propose a team, scope out the project, and then deliver to a specific set of outcomes.
For these types of projects to be successful, you want experienced consultants with deep domain or industry expertise. You can’t afford to pay for them to learn on the fly. When it works well, you get high quality results quickly and are happy to pay the fees because the value is so much more.
The downside of outsourcing is the same as with any consulting agreement: misaligned incentives, scope creep, inexperienced team members, and high costs. You have to trust that they’ve considered the details and that they’re not going to sell their newfound skills to one of your competitors. It’s worthwhile to consider incremental or iterative approaches and to explore alternative contractual arrangements (fixed price, shared benefits, bonuses, etc.).
Finally, your organization likely won’t know how to fish once the consultants leave.
Co-sourcing is a hybrid approach that tries to have its cake and eat it too. It ends up being an insource model, but it gets there more quickly by leveraging the expertise of a consulting partner.
The idea is to start out with an outsourced consulting team, but in a longer term relationship. They don’t simply deliver a single project result. Instead, they start by setting a strategy, prioritizing a set of projects, and beginning to educate the organization. As the engagement progresses, the consulting team helps the organization hire replacements for itself. Over time, the team size remains roughly the same, but fewer and fewer of them are consultants until eventually, it’s an internal team.
Most consultancies have flirted with this approach, but in many cases it’s misaligned with their strategy. If your business model is to sell fish, then you shoot yourself in the foot when you teach others how to do it. Sometimes an engagement with a university can work: professors and grad students act like the consultants and over time, some of them form your core team. But be careful. Academic goals are not aligned with your organization’s.
The benefit of co-sourcing is its speed. You get quick results to analytical problems and you get a ready made team at the end of the engagement. The consultants take on much of the difficult change management and infrastructure work while still delivering results. The whole process may take a year or more to complete, but in the end, the client has a team, data infrastructure, some data products, and an engaged organization that has proven the value of analytics.
The downside of this approach is that it’s expensive. You’re paying someone else to do analytical work along with HR (recruiting and training), change management, and data infrastructure upgrades.
Let’s take a quick detour to Acme Benefits here. It appears that co-sourcing will best meet their needs. It will bring them quick results in a particularly complex environment. But it will also build the ongoing capability that they require. As an added bonus, Acme operates in an environment where the potential gains from analytics are extremely high. A co-source partner would likely find enough value in the first few months to pay for the entire group for years to come.
Crowdsourcing is a completely different animal. It’s not about building general analytical capacity, but instead, accessing deep capability for specific problems. Typically, crowdsourcing is organized as a contest with a “prize” for whoever gets the best result. Netflix famously offered a million dollar prize to whoever could improve their movie rating algorithm by 10%. Over 40,000 teams entered and spent the next three years testing different approaches.
Goldcorp is another example. With a single property in Canada, Goldcorp was struggling to find enough gold to be profitable. They decided to release their geological data to the public and offer anyone the chance to predict where they should drill. The top five submissions were paid out $500K in prize money, and Goldcorp literally struck gold -- going from a $100 million valuation to around $9 billion.
Crowdsourcing is not fast. You need time to define the problem, to organize the contest, to prepare datasets, and to promote the whole thing to potential participants. Then you need time for the “crowd” to do the analysis. Depending on your business, the price may have to be high. People will participate because of the prize, the novelty of the problem, and the prestige of winning. You need to address all three. If done well, the ROI can be spectacular.
Maintaining analytics capability is often more difficult than building it in the first place. Leaders change, key staff move on, or maybe the tough nuts get cracked and the superstars are no longer needed. Within a couple years, many organizations are right back where they started.
Remember, analytics teams thrive in a culture of continuous improvement. They crave variety and they need to move the dial. If they’re underutilized or spending all their time cleaning your data, then they’re going to jump ship. Demand for these people will vastly outstrip supply for the foreseeable future.
Don’t be afraid to vary your approach over time. Build a small team that’s really good at identifying opportunities. Then engage with some academics to crack a tough problem. Hire the grad student, run a contest, then co-source to build up your capabilities. Then bring in the consultants to solve another specialized or difficult problem. Your capacity continually flexes to keep pace with the demands placed on it.
Finally, don’t forget to track your wins. Before you fix a problem, develop a way to measure the impact of the “fix”. If you can’t point to concrete improvements, your team will soon be defunded. It will also keep your focus in the right place.
Good luck and godspeed.