You've successfully subscribed to Ulysse Bottello — ML Product Manager
Great! Next, complete checkout for full access to Ulysse Bottello — ML Product Manager
Welcome back! You've successfully signed in.
Success! Your account is fully activated, you now have access to all content.
Why guap exist?

Why guap exist?

Long story short: because the luxury of doing an AI/ML project for the sake of it no longer exists.

Ulysse Bottello
Ulysse Bottello

Yes, it all comes down to this missed opportunity in which I've got a profound trust, applying AI productively.

From the 2000's ML analytics phase, where value comes from incorporating simple model insights into decisions to 2010's ML hype phase where values come from marketing promises and FOMO.

It's 2020, more than $50 billion is expected to be invested in AI systems globally in 2020, according to IDC, and 2% of this budget will generate value.

You may experience it like me. I’ve read it everywhere: "every AI/ML/data project starts by linking research with business".

But in practice, usually, the story goes as follows. The director's committee decides that their company should lead the AI innovation ecosystem in their industry. To this, they hire a team with great skills, spend a vast amount of money on resources, and after a few months of ‘PUM’ they find themselves at the starting point with less money and no trust at all for AI.

I can’t evaluate the business impact of what my AI’s doing. It’s been 4 years. There's no way I'm alone.

That being said, what guap is aiming at?

Lack of collaboration

Guap's first mission: Make collaboration healthier and clearer between tech and non-tech people.

Listen to data scientists, they may rant about being isolated in the organization, doing the grunt work. Then, listen to managers, they are struggling to understand what the data team is doing, why is it taking time, and what they can expect.

The dialog between the data team and managers isn't fluent. Resulting in a frustrating experience for everybody, and ultimately in a failed project.

We need a translator, a shared language that helps both parties. A zero-friction language to learn, a language-centered around something the broader audience in a company can understand, the dollar-sign.

Ok but the data team talks $, now what?

Because a failed ML initiative is often caused by a poorly scoped project. The adoption of a non-technical metric comes with an effort for non-tech people. It requires that we do collaboratively an extended project setup. No requirements, no magic.

How to collaborate with guap :

  • Make sure your project is set up before any data work
  • Communicate guap scores along with technical metrics

Better decisions

Second mission: Inspire the next generation of AI teams to make better decisions while applying AI.

One overlook skill a data scientist must have is its ability to get aligned with its organization's vision and goals. Doing research is awesome, we're convinced, but it's one-third of the equation. Finding how to productively leverage it in your product, as François Chollet says, is highly rewarding considering the challenge of making something viable and desirable at the same time.

Being product-first is the antidote to getting lost on training models, trying new techniques, and increase accuracy.

Also, if the value comes from models optimized to improve the business' product or service, it means that you will aim for reproducibility, scalability, and maintainability over complexity.

But don't get me wrong, profit alone shouldn't be the only success criteria, and everyone onboard on the AI journey should be educated on every other metric. At the end of the day, it's all about trade-offs, but tradeoff decisions with an exhaustive and comprehensive view are game-changing.

Make well-informed decisions is one thing, making them at the very beginning provides a lot of value too. Nobody wants to invest too much time and effort in an unviable idea the board had in a meeting I guess, not even them.

How to make a better decision with guap :

  • Start using simple models to validate the feasibility, continue or kill early
  • Leverage guap early in the lifecycle, even when you are training the first epochs
  • Check often the profit curve, to anticipate a possible plateau, or diminishing returns

This is guap raison d'être.

Solving data and model concerns is hot rn. But it misses the point, feasibility is one way to prioritize, but you still can have an impact...

Now we know why AI fails at generating value for people and businesses, Guap should be the kickstart of this shift, by becoming the standard for evaluating investments at every stage of the AI/ML lifecycle.

Our mission is to align all stakeholders with measurable business outcomes and the trustworthiness of results by including the three core teams — business, data science, and IT — throughout the life cycle of the AI model. With a product, a data scientist AND an executive will love to use it.

Embark with us on this journey and let's see if profit is all we need!

Algorithms outputs to business outcomes. The magical ML evaluation metric everyone can agree on 🎩 - guap-ml/guap