“Machine learning (ML) today is frustrating. There is so much potential and the algorithms are all there, but I just do not know how I can leverage it for my organization,” says the CTO of a major professional services firm. “My CEO wants me to ‘leverage ML to lower OPEX and differentiate our service offerings, but there is nothing out there in the market that would allow me to get this done in a manner that has a high probability of success.” Then of course he asks me what I would do and where I would start, because the guy with “Machine Learning and AI” in his job title must know for sure…
That said, here are my 3 key principles for everyone who wants to get started with ML / AI right now:
The 3 Key Principles of Getting the most out of Machine Learning and Artificial Intelligence
ML / AI Principle 1: Do not constrain your thinking to what others tell you is ‘reasonably possible and what, in their opinion, is science fiction.”
This is absolutely key when thinking about transforming IT and the overall business through ML / AI. Whether you are a line of business guy, a developer, or an IT ops person, the best thing you can do is think about what you, your team and your peers would need out of ML / AI to more efficiently server customers. These tasks may seem too complex to automate, but in reality they may not be. Also, the goal is not necessarily to fully automate these tasks, but much rather to enable faster and more accurate decisions that are optimally aligned with your corporate strategy.
Example: When I was running a team of product managers and marketers, I wouldn’t have wanted my ML / AI assistant to decide for me which features to include in the roadmap and how quickly to develop them. However, what I would have vastly benefited from was an ML / AI assistant that pulls together all decision relevant context so that I can make a fully informed decision quickly and without digging into the raw data myself. Note that these examples are to get you thinking and that I do not want to give the impression that this is a complete case study:
Watch at least part of this video to understand why I believe that today's limiting factor of ML / AI progress is not technology, but artificial (no pun intended) limits in human thinking.
Lost Deals: How many entries were in my CRM that mentioned the absence of the specific feature as the underlying reason to lose a deal? How large were these deals that were lost? How many other features were named? How many previous deals of that size did the salesperson close.
Customer and Partner requests: How many customers and partners escalated the absence of the feature to my own inbox and how strongly were these escalations worded? Did my individual product managers respond with viable solutions or did they recommend awkward workarounds?
Strategic importance: Are these features relevant to win strategically relevant deals? The strategy may not focus on the highest dollar amounts but on conquering a specific market segment or consistently beating a specific competitor.
Specific instructions or commitments: Were there any commitments to customers, my boss, partners, staff, etc. to get this feature implemented by a specific point in time.
Difficulty, Effort & Cost: How much will it cost? How likely is it that we will come up with something that will make a difference? Is there staff available to do it or do we need someone external?
Other strategic considerations: How important is the feature to show market leadership (no direct deal impact)? Will it damage our brand to show gaps compared to the competition? How important is it for getting our staff excited? ‘
ML / AI Principle 2: Do not expect to understand what data is useful for ML / AI to make a decision
Always remember that everything that happens in your organization leaves a trail of logs, metrics, and other data behind. While humans may not be able to deduce causalities between terabytes of operations data, the machine will find ‘something’, if it’s there. Humans will use their own reasoning skills to poke around in the hay stack, while the machine will take apart the entire stack and look at every permutation of straws separately in in correlation with other permutations. Here are a few examples of crazy causalities that humans couldn’t have identified.
These samples show the dramatic effect ML / AI can have on any organization:
Sales guys who wrote shorter follow-up emails with less generic product info were 50% more likely to close the sale within the first 4 weeks.
The product inventory pages of the ERP loading slower and slower is due to a code push of a seemingly unrelated marketing automation system that hammers the ERP’s API to pull large number of full resolution product pictures.
When a specific product manager opens up PowerPoint, Word, and a number of specific other apps and websites, ML AI can reason that it’s time to assemble the quarterly board presentation again. The ML AI can then pull all the relevant metrics in terms of market traction, roadmap progress, P&L and so on from the appropriate sources and log into the EMA library to find some interesting market and competitor data to add color to the presentation.
ML / AI Principle 3: Think about what distracts you from adding value to your organization
Every one of us, whether we are in IT Ops, work for a business Unit, or provide customer service are held back on a daily basis by tasks that “just kill our productivity.” Leveraging ML / AI to automate or support the execution of these tasks will not only make sure staff gets more done, it will also increase staff motivation. Here are my favorite examples:
Filing documents is killing me: I need ML / AI to read my incoming emails and extract all attached documents into my file system. Of course ML / AI also needs to make sure to consolidate different versions of a doc, so that I don’t send the wrong one to production.
Tracking promises and tasks: In my outgoing emails, I often promise customers certain dates for their deliverables. I need my ML / AI to add these deadlines to my Trello board.
Disk full issue: I need my ML / AI to offload large media files from my local disk to my remote drive, as soon as I have added captions to them and published to YouTube.
Social media: My ML / AI needs to read all my publications and come up with automatic tweets that fit my current research schedule so that I don’t have to spend time on generating and scheduling these tweets by hand.
Taking and finding notes: As an analyst, I take excessive notes of everything I find relevant to my current and future research. I store these notes in Microsoft OneNote so that I can do a full text search to find them again when needed. However, the big problem is that OneNote does not let me cluster notes by topic, which would greatly benefit me on a daily basis and here is why:
- Example 1: My current research project on AI / ML included many dedicated customer, partner and vendor conversations I’ve had on the topic. However, a lot of conversations I’ve had on past research, for example on container scheduling, also applies to this topic and could add interesting use case examples. And now, let’s think this a step further: I’ve also had many vendor and customer conversations on problems that could have been resolved through ML / AI, but without ML / AI playing a role in the conversation at the time. I’d love to be able to cluster the topics from all of these conversations so that I can make a fully informed decision of where ML / AI will be the most impactful today.
- Example 2: I would like to be able to automatically pull relevant research figures from the EMA research library, based on, for example, the content of this blog post. These figures should not be around AI / ML itself, but identify problems with high business impact that could potentially be solved through my vision of ML / AI that I have laid out over the previous 3 blog posts.