Opportunity knocking. For some of us, it’s that moment when you find yourself in a business situation where you can feel wasted time, see inefficiencies everywhere and start to think “there must be a better way of doing things”.

You see tons of data being generated and wonder at the potential richness within. You know there’s some value because looking around, folks are writing scripts all day just to make that data usable.

One day, someone pulls out a sharp-looking dashboard that cuts out noise and managers are all excited … probably because they can now better manage upwards.

A voice whispers “We’ve collected so much data, should we should start looking for trends … and patterns?”.

But who has the time? And energy?

A couple of years later, a company comes knocking on the door with an AI/ML solution tailor-made for your data set.

They promise to find valuable patterns in all that data. Patterns that could lead to better decision making, reducing time wasted and in turn, costs.

But it comes at a price that makes your bosses pause and ask … “Is it really worth it?”

How much value can your AI-driven solution offer?

“Whoever said data is the new oil was right, oil’s worth nothing these days” — Philo Daniel, 27 April 2020

The hardest lesson going around trying to sell AI solutions is finding out most problems you offer to fix are not important enough. And have already been solved in some way or form.

Maybe not efficiently, maybe not effectively, maybe even badly … but good enough to maintain status quo.

Josh Winters covers all of this in his brutally honest piece Hard earned advice for AI which we wish we’d read a couple of years ago when first going to market.

No one builds to fail

If there’s one message that’s been consistently repeated by analysts, it’s that data science/AI/ML/analytics projects have a very high failure rate.

The reasons are numerous — lack of goals, missing expertise, non-data driven cultures and solving a non-problem perhaps.

In summary, the business case simply didn’t work.

But it’s in this environment, where every reason to fail surrounds us, is where we MUST somehow show value … value tangible to every level of customer engaging with our solution.

And we need to be okay knowing that the customer … the business … doesn’t really care about our technology or our machine models.

No excuses

Assuming you’ve gotten through the door and started trials, then you better find a way to quickly to transform the data (no matter how bad), get a working model going and produce near-usable output.

These are table stakes.

If the data is rubbish, find a way to fill the gaps. And it’s risking an early end to say “We can’t give you an output until you tell us what the input data means” given that the customer likely has no idea either!

Excuses won’t matter because there’s often no civil way of answering someone who says “But we thought your AI can figure out what’s missing and compensate”.

Business value

Eventually (hopefully!), we might get to a point where value has been perceived i.e. your AI/ML output finally delivers an outcome for the customer.

This then needs to be translated into a business case which can be sold to those who likely aren’t there in the room.

Vendor — “Look at all these insights … you guys can now replace hardware a month before it fails”

Customer — “Hang on, we pay you $100,000 for this info, but the impact of an unplanned hardware failure is about $5,000”.

Vendor — “Exactly. This information shows you a schedule for 30 pieces of hardware— $150,000 … so you save $50,000”

Customer — “Hmm, how accurate is your AI? Didn’t you say 70%? And what’s this precision thing you were telling me about?

The fundamental question always ends up being, does the AI-powered solution make a significant enough difference … to make it worth buying?

And from this moment, it’s only the business case that the customer cares about.

Detaching from technology

You’re trying to convince someone who will likely need to persuade someone else … when you’re not around. So any business case has to stand on its own.

Inertia, though, is everywhere, and most customers struggle to build their own narratives.

So how did we help?

#1: Using a data visualisation tool such as Tableau or Power BI

Whatever output comes from your AI/ML system whether raw, via a formal UI, or post-built using Python libs, a data viz tool can help make it relatable to your customer.

For us, it gave our sales lead the chance to chop and change visuals on the fly (even during meetings) until she found a mix that resonated most with different members of the customer team.

All without being dependent on a data scientist sitting a thousand miles away.

These weren’t complex or fancy dashboards. They were simple “decision-support” visuals to nudge the sales process in the right direction.

#2: Microsoft Excel

There have been very few projects where we didn’t end up building the foundations of a business case for the customer using Excel.

Almost always, someone from the customer will ask for “numbers” from you (what exactly they might not tell you), and then update it with their own internal projections.

Usually, data from the machine model needs to be stripped and then supplemented with financials, hours and various other metrics that one only figures out are important much later.

Most of our work involved making sure spreadsheets were simple enough for the customer to run with and numbers solid enough for them to abuse!

#3: Microsoft Powerpoint

At some point along the customer journey, your story gets crisper, cleaner, more concise, more targeted to the needs and wants of the moment.

And that almost always needs to be reflected in a one or two page slide deck that summarises the key benefits in the customer’s words.

Again, it’s a case of the customer asking you, you build it and send it on, they chop and change and present it to their stakeholders … without you in the room.

So making sure the quality of the messaging is top notch in its simplicity is number one. And not overloading on data science techno-babble being number two!

Or, if feeling supremely confident, you could risk them building their own slides …

Are you, dear reader, underwhelmed?

Honestly, if you’d asked us a couple of years ago, we’d have admitted we would never expect Microsoft desktop tools to figure in a data science article.

We wish we were bragging about how cleverly we used Bokeh or TensorFlow or how fit-for-purpose our latest ML model was.

But the reality is that we also spend a huge number of hours on visualisations, spreadsheets and powerpoints. All to keep engagements going and to communicate, communicate, communicate…

“We’re not in the coffee business. It’s what we sell as a product but we’re in the people business — hiring hundreds of employees a week, serving sixty million customers a week, it’s all human connection” — Howard Schulz, Starbucks

We asked ourselves how we ended up here, and the only conclusion we came to was that data science tools and concepts, for the most part, are relatively unfamiliar to our customers. Their outputs are too raw … and unforgiving.

And even if understood by their technical staff, the engineers themselves struggle to communicate to the ones who write the cheques.

There’s magic in simplicity

The tools we mentioned were familiar and safe … like a baby’s security blanket.

And familiarity with these tools helped the customer relate to our messaging … on a platform they could understand and engage with.

These tools communicated the outcome the customer would get.

Food for thought

It also helped us understand the customer’s struggles and convey empathy much better.

To us, that was the foundation the relationship was built on.

Of course, this is merely our perspective. And maybe we are smoking something.

So please don’t get us wrong. It takes time … and effort … and more time. The business case has to work. They still have to like you and what you say. And it doesn’t always end with a win.

But the next time you see data science knocking Microsoft Office, there’s a more pointed way to respond.

(Feb 2020) Clouds roll in mid-morning at Reberty, French Alps