Signs You’re Ready for Automated Intelligence, Part 2
September 1, 2020
Four Things to Look For from Your Leaders and Teams
In our mission to bring modern predictive analytics to organizations of any size and type, we approach analytics as a service, with an emphasis on automation.
What this means is, we seek to build analytics workflows that take care of themselves to the fullest extent possible, weaving decision support into the fabric of your organization. At a high level, we help you plan, build, and implement:
- Predictive models that align with your business goals.
- The infrastructure those models need for automatic delivery and continuous data updates.
- The visualizations your teams need for using the models in decision making.
You can benefit from analytics as a service no matter where you are in your data journey, but there are some signs that you’re truly ready to dive in. Recently we discussed signs of readiness in your data and reports.
Today we’ll focus on signs from your organization itself – specifically from your most valuable asset, your people.
Here are four signs from your leaders and teams that you’re ready to elevate your analytics with automation:
1. An Experimental Mindset
Litmus test time: success with predictive analytics requires that your organization accept some failure. If this statement is a non-starter for you, then your organization may not be ready to automate them.
Many organizations say they have a “culture of learning,” or that they’re committed to building one. The truth is they often have a culture of knowing the answers, not a culture of embracing the learning process.
Predictive analytics is a discipline that includes mistakes by design. Predictive models cannot perfectly illustrate the future (unless, maybe, Christopher Nolan starts building them). But they get progressively better at it when an organization carefully studies and corrects model errors. Automated workflows deliver predictive analytics without delay, but not without people – or their oversight.
There are any number of reasons why a predictive model might produce errors beyond an acceptable threshold, including faulty assumptions, faulty data, loss of fidelity somewhere in the process, black swan events (the events of 2020 come to mind), and everyday bugs. We can partner with you to find, address, and learn from these kinds of errors, or build the internal discipline to do so. An organization with a genuine experimental mindset is willing to take this on.
It is also willing to “reframe mistakes as a source of discoveries.” It is ready to put in the work and the cycles, experience setbacks, collect the stories and postmortems, and internalize them. This – not mistake avoidance – is what yields real improvement, which happens over time.
A highly visible example of this is Airbnb, which has described how they conduct controlled experiments and learn in cycles – a process that has contributed to their 43,000 percent growth (yes, really) in five years.
An example in the opposite direction – with mistakes on full display – is Google Flu Trends, a program that went from “the poster child of big data into the poster child of the foibles of big data” when it dramatically miscalculated the 2013 flu season by 140 percent. Learnings mined from this failure could have produced a powerful public good, but sometimes even confident players abandon a predictive effort after an initial disappointment – Google pulled the plug on GFT before subsequent iterations had a chance. (Obviously Google has invested heavily in AI since.)
If your organization’s culture rejects risk rather than taking any, here’s the good news: you can start small with automated predictive analytics, which we discussed in Part 1 of this series. In fact, a small step in this direction can actually be the catalyst for an experimental mindset.
2. Buy-In from the Top
If your organization’s leadership is not on board with automating your analytics, you probably won’t see much success, or even the starting line.
Leadership is where an experimental mindset is championed. Leaders may or may not originate this kind of mindset, but they must create and support the culture of learning where it flourishes. Only leadership can supply and protect the resources required for an organization to learn – like talent, technology, and strong R&D.
And only leadership can drive adoption – which only about 8% of organizations are doing with respect to AI initiatives, thus explaining their meager AI gains.
Regarding automated predictive analytics in particular, leadership must be prepared not only for the setbacks themselves, but to take ownership of them. Teams must know that leadership has their backs in a safe trial-and-error environment. Leaders who are excited by new learnings (and not threatened by them) are ideal for introducing automated analytics to their organizations, and for bringing their teams along the iterative journey of growth that it enables.
If the leaders in your organization are partial to the status quo, here again the start-small nature of automated predictive analytics makes a compelling argument. But if starting anything is a delicate issue, then stopping something might be persuasive – and automation might be the answer to the requests that leaders are hearing from their employees. Which brings us to the next organizational sign that you’re ready…
3. Requests for Help
We’re hardly advocates for complacency, but we also know that there’s a lot of truth to “if it’s not broken, don’t fix it” – and your teams have a way of letting you know if it’s broken. Are they asking for more data, more time, or more access in order to deliver answers? What about less access – is the analytics workflow currently so porous that data too easily loses its integrity? Does data at your organization drive decisions or busy work?
In our work, we hear various types of requests for help over and over again that point to a need for analytics automation:
- One of our mid-size financial clients experienced monthly lags in decision-making because month-end reports required several days of data hunting. They had to collect data both manually from multiple departments and digitally from machines with 24-hour delays built into their turnaround time. And if month-end fell over a weekend, things got even further behind.
- One national industrial equipment manufacturer that serves a dozen different markets through more than 100 local branches was drowning in data ingestion. Its leadership wanted to “move beyond run rates and become proactive” in expanding its KPI set and deploying its sales team accordingly.
- One social media client wanted to standardize and replicate its models that predict user engagement – they were built by a talented internal team, and could be maintained only by that team.
These examples have common themes: companies need to understand and use their data more quickly, more comprehensively, and more collaboratively. Automated analytics is often the answer to these needs.
4. You’ve Rooted Out the Moles
A real culture of learning demands the freedom and willingness to challenge new technologies and their results. But that’s different from actively working against them.
Depending on whom or where any resistance to change is coming from (and it’s probably not a corporate double agent, although we’d watch that movie), it can sabotage your progress with automation, or get it scrapped at the first setback.
At one company we know, a team of ten analysts was manually selecting marketing lists – homogenous reporting that was ideal for automation, as we discussed in Part 1 of this series. Company leadership introduced data science into this group, ramping up its modeling capacity from one model per year to five per month. The needle began to move, but not as far as expected. Some of the new model results suggested that they may have been reconfigured with some of the old model techniques.
And then a couple of other things happened. In one model, a price-related variable heavily dominated likelihood to buy – a mistake that cost the company a few hundred thousand in revenue. The data scientists discovered, reported, and corrected the error quickly, and the company ultimately considered the loss a blip. But the leader of the analyst group – a person with seniority, 15 years of experience in direct mail, and a fear of becoming obsolete – sounded the alarm.
Simultaneously, a totally unrelated macroeconomic event caused the company’s sales to drop across all its products. This was the nail in the coffin that the team leader needed. The company froze all modeling and doubled down on manual list selection.
As Peter Drucker once famously said, “Culture eats strategy for breakfast.” Automation will eventually give someone with seniority and sway the ammunition to torch the whole effort, if they are so inclined and able. So an experimental mindset, buy-in from the top, and requests for help must be shared and authentic, or the naysayers will have more power than they should.
Like we said last time, the field of AI has made a lot of promises, many of which are overblown, and most of which will fail if an organization doesn’t also make and keep some promises of its own. Ultimately your people are critical to the success, and even just the start, of automated analytics. Your leaders and teams will determine the levels of interaction and commitment it receives at your company, which will in turn determine the results it delivers.
We don’t recommend starting any journey toward automation without them.