What's the difference between good & great AI business projects? Here are 7 things to consider when doing AI work in your organisation.
What’s the difference between good and great AI business projects? Most advice on building effective AI focuses on the data science and technical aspects of work.
These are of course crucial, but there’s now enough industry experience to start speculating on other characteristics of effective AI work for business.
Here are 7 traits I believe contribute to the business success of great AI work. Hopefully it’s food for thought on your aspirations for your own AI business projects.
Most people working in AI would probably acknowledge that highly effective AI business projects still aren’t the norm. AI work is seen as primarily technical, with much emphasis on algorithms, data science and data engineering.
There are still relatively few examples of corporations talking about quantified business benefits of their AI work. But that’s changing, and measurable business goals are starting to appear more often in commercial AI work.
A good AI business project delivers AI outputs that should or could help a business. Highly effective AI work is driven by the need to achieve and demonstrate business results:
A highly effective AI business project is sharply business-focused, with business outcomes shaping the rest of the project work. These remain front of mind for the whole team, the whole time.
This isn’t yet typical for AI, and is at best aspirational for many businesses investing in AI. But it’s a useful aspiration, shaping tone and culture for all AI work in an organisation.
Even if this remains elusive in your projects, aiming for it is worthwhile. You’ll be less likely to stray into the AI pitfall of being excessively technology led.
Here are 7 “habits” to consider adopting in your AI work. Ideally, they’ll accelerate your progress towards highly effective AI business projects. At a minimum, they’ll help maintain a healthy balance between technology and business priorities in AI work.
AI has a deserved reputation for being technically complex. Building good AI solutions requires deep skills, many technical. However, this doesn’t mean everyone on the team has to understand that complexity.
Rather, it can be helpful to have business team members who understand the principles of how AI does its apparent magic. It can even be an advantage to not be caught up in the underlying maths and statistics.
Instead, it’s more common for business users to be drafted into an AI project with little AI education. They typically pick up their AI knowledge “on the job”. This is often the case when the role of business involvement is to define requirements. In many situations, this is a reasonable approach, and may be the only available option.
However, if you have the opportunity, consider equipping AI team members from the business with more knowledge. This would allow them to to do more than specify business requirements. It will create overheads, and your data scientists and business analysts may not be entirely comfortable. But you’re likely to deliver a better business result.
One of the banes of new technology is unrealistic expectations of what it can do. This is so common, that a phase of Gartner’s Hype Cycle is called the “Peak of Inflated Expectations”. AI is prone to inflated expectations at the moment, partly because there’s so much coverage in the mainstream press.
A challenge for AI projects is businesspeople who expect overly ambitious or unachievable results. A variation is high expectations of timescales for successful AI solutions. This goes hand in hand with an unwillingness to tolerate experimentation or mis-steps.
Another facet of unrealistic expectations is not appreciating the importance of poor or insufficient data. This manifests as impatience with the effort to prepare data for AI use.
There are many ways of validating business expectations, from education to collaborative development of goals. How you approach it depends on your organisation. The habit to acquire is an AI team and affected business with aligned expectations. Bear in mind that alignment may mean just being in the same ballpark.
Good AI business solutions are designed around meaningful business questions. Meaningful includes the availability of enough appropriate data to answer them effectively.
But from a technical perspective the quality of the question matters less. The heart of AI is built on statistics and code, which is indifferent to business meaning. AI technology doesn’t “care” if you design algorithms that calculate something meaningless. They will run just as smoothly as something more useful to business.
A hallmark of highly effective AI business projects is an underlying business problem which ties together the data science and technical development in a quantifiable way. Even a humble chatbot can be driven by a quantifiable business improvement.
So it’s worth checking the link between your expected business outcomes and AI features. If it’s unclear or “fluffy”, you may want to look harder at the underlying business questions.
How well an AI solution meets requirements doesn’t determine the business value it delivers. AI can only deliver business value by incorporating its results into business processes and operations.
The role of business in an AI project is at minimum specifying requirements. It may also extend to helping translate these into AI features. And in some cases, business users might also help add a business perspective to uncovered data insights.
What’s less common is thinking ahead to how business will use the AI outputs in practice. At minimum, this covers the high level goal of delivering business improvements. But ideally, it also should cover operational details. These will include who will use AI outputs, how business procedures will change, and what communication and training to provide.
What’s typical is to start considering this in detail only towards the end of development. In contrast, highly effective AI business projects work on business implementation planning and impact early on, alongside AI solution design.
AI work is by nature experimental in phases, and uses technology that is evolving rapidly. So it’s not surprising that some AI work may go wrong, and some will be wasted. But that’s different to things going wrong because of inexperience or young technology. Effective AI work gets things “wrong” as part of the mathematical process of finding the right answer.
Some of the underpinning statistical and mathematical techniques in AI are essentially based on trial and error. They’re feasible because today’s powerful computers can repeat calculations millions of times until they find the right answer. So you only find the right answer by rejecting many wrong answers. This applies to several aspects of AI work, from optimising algorithms to picking the right data characteristics.
Another area where “wrong” answers are part of finding the right answer is when exploring available data. There’s an early phase in most AI projects called Exploratory Data Analysis or similar. This is all about experimenting with available data, seeking ways of extracting relevant business insight.
An inevitable risk of AI work is spending more time than “necessary” on exploratory work. But you’ll only know the minimum possible time after finding the right answer you’re after. If you try too hard to reduce this, a worse risk emerges. You risk developing an ineffective or inefficient AI solution to your business problem.
Highly effective AI business projects strike a balance between responsible and profligate experimentation phases. In finding this balance, there’s a strong argument that it’s better to err on the side of better long term business results. The alternative is erring on the side of lower short term project costs and timescales.
AI isn’t cheap, but it doesn’t have to be expensive. If you understand the process of building AI, it’s possible to do meaningful work with reasonable budgets and timescales. If you don’t, then it’s also possible to incur significant and/or avoidable cost.
With more traditional IT, there are benchmarks, rules of thumb and experience to use when setting expectations for effort, duration and budget. There’s also a relatively mature marketplace, so if you request quotes for a piece of work, most will be comparable.
This doesn’t yet exist for AI, so estimating AI budgets and project durations is tricky. It’s exacerbated by the experimental nature of some phases of AI work. This means getting AI teams to commit to budgets and dates can be an uncomfortable exercise. Sometimes estimates seem high because the work is genuinely difficult. But other times, it may be a reflection of inexperience or over-enthusiasm.
Highly effective AI projects provide enough investment and time to give teams a chance to succeed. But they don’t provide so much as to take away healthy pressure. The right balance prevents any tendency to lose focus on the outcome and its business value.
The right budget is probably more than sponsors are comfortable with, but less than AI teams feel they ideally need.
It’s not uncommon for the gap between those two to be too large to bridge. In that case, the mismatch is probably pointing to an issue in one or both sets of expectations.
AI is still relatively young, with intrinsically experimental phases and variable industry experience. The plethora of AI “snake oil” salespeople doesn’t help. It’s going to be a while before there is reliable, industry-wide data on how best to do AI work. Until then, it will remain tricky to judge how much to spend, how long AI should take, and how much business benefit to expect.
As long as this challenge remains so large, managing and delivering AI work will continue to be a series of balancing acts.
Careful thinking and planning is critical, armed with enough knowledge to not be misled or worse. But developing AI is a hands-on activity, and it’s easy to over-think it early on. You can achieve more in a few days of exploratory data analysis than weeks of strategising and planning. So the balance between thinking and doing is one aspect.
There’s also an in-built tension between business, technology and data science. An effective AI project team reflects and respects all three. It balances the tension constructively — most of the time, at least.
Finally, AI started in labs, and is still the realm of scientists and mathematicians. But the audience, beneficiaries and sponsors are business people, motivated differently. Perhaps the biggest balancing act between these two mindsets is finding the right point between “best” and “good enough”. This is particularly apparent during experimental phases to optimise models and algorithms.
Highly effective AI business project teams understand the need to balance “thinking”, “managing” & “doing”. They find a way of maintaining this most of the time, despite the different mindsets, pressures and priorities involved.
Building AI solutions is a difficult enough technical exercise. Whether a business will use them effectively can seem an added complication. But highly effective AI business teams are conscious that business profits pay for their work.
Effective technical work should be the minimum expectation a business has of AI teams. Effective data science work shifts the expectations towards the meaning of what the technical work produces. At the heart of good AI teams is acceptance of this or even enthusiasm, rather than tolerance or endurance.
To step up from good to great AI work in business, there’s a further shift from business meaning towards business value. But this is invariably beyond the direct control of AI teams, and requires building bridges into the wider business.
A highly effective AI business project has the wider business in mind from the outset. It adjusts its activities, team composition and mindset accordingly. The 7 habits in this article are characteristics of such teams and the work they do, but are aspirational in most organisations.
Nevertheless, if you’re leading or contributing to an AI project, you’ll have opportunities to try putting at least some of them into practice some of the time.
And that’s a good first step towards making all of them a habit all of the time.
Keeping up in the new silicon-based survival of the fittest
Innovations in Artificial Intelligence - Various sectors in which AI is remarkably used & has brought changes in humanity - Education, Healthcare,automobile
Explore to understand how AI artificial intelligence has advanced and presently serves as a roadmap to augment your business in 2020.
Lemonade is one of this year’s hottest IPOs and a key reason for this is the company’s heavy investments in AI (Artificial Intelligence). The company has used this technology to develop bots to handle the purchase of policies and the managing of claims. In this post, you'll see How to Create an Artificial Intelligence (AI) Model
Every week we bring to you the best AI research papers, articles and videos that we have found interesting, cool or simply weird that week. Have fun!