Skip to main navigation Skip to content

Implementing AI: Where do I start?

Getting started - runner at starting line

One of the most common questions we encounter when meeting new firms is how to adopt and start using AI and other analytics technologies.

Often, we find firms come at this from a data-centric perspective: “we have the following A,B, and C bits of data, can we forecast what will happen with X, Y, and Z outcomes?” This data-centric approach is natural because data scientists are generally driving the process.

Data-centric approaches suffer from several fundamental flaws. Some data scientists tend to have an extremely narrow focus and can often be far from the on-the-ground reality and challenges that a firm faces. Additionally, the average data scientist has had that title for 2.3 years. This lack of experience and expertise could mean that:

  1. Some data scientists could lack the understanding of what benefits and insights could be possible given slight changes to data collection systems;
  2. Some could lack that deep command of the underlying math required to master and understand when to use different artificial intelligence approaches; and
  3. Some are not deeply integrated with the senior executive team and therefore may be missing opportunities derived from boardroom strategies or struggle with delivering actionable insights.

In a data-centric approach, one of the early steps is choosing which AI model or methodology to apply.

We often see projects in which many different models are applied to a firm’s data in order to find which fits the data best — this is how many firms drive their model selection process, and it is actually problematic. This is called “curve fitting” and often results in a model that fits the available data (the training data) but at the expense of having real predictive power.

This approach also reduces what could be possible for a firm to analyze, and what would be the most meaningful insight that delivers the most value to their executive team and/or their clients.

Which is why our approach doesn’t do this.

We have combined the best of two worlds — the agency model of professional services with our proprietary technology to create a hybrid approach. We do this by starting with a simple question, which has nothing to do with data.

“What challenges do you face as a business?”

Yes, this is a loaded question. Yes, you’ll likely be able to fill a few whiteboards with answers. And that’s the point. This isn’t a starting point. This encompasses all the places we want AI to take us.

By seeking to find the most valuable leverage points for a client, the intersection of data and business challenges that can lead to unique and actionable insights, we look at the best detailed map we can create of your challenges. This starts with our workshop model, an approach we have used for over a decade with our agency clients. Underlying this model is our decades of experience in applied analytics and AI, and a collection of proprietary technologies that we’ve used to bring success to other firms.

Our workshops begin with a deep exploration of your business and customers. We work together to map out your customer journey from brand awareness through the research and purchase process through onboarding, ongoing service, and support. We identify the pain points to alleviate and the opportunities to capitalize on. We then leverage a series of brainstorming exercises rooted in design thinking methodologies to converge on possible solutions to your business challenges.

This workshop model is the foundation of our solution-centric approach to AI. It helps us partner with clients in an integrated way to ensure that any analytics or AI solutions are all driving towards one objective – generating actionable insights that help to solve business challenges and drive success.

Along with the workshop we apply a discovery phase that determines the following:
Landscape and Market Analysis:

  • What are you doing today?
  • What are your competitors doing today?
  • Where do we think the market is headed?

Customer Research

  • What are the most common Customer Profiles / Personas? How does each interact with your business, and what do they find valuable?
  • What challenges do your current customers face? What would be the best way to address those?

Technology and Data Assessment

  • What technology stack are you currently using? What is the breadth and depth of expertise in your technology organization?
  • What shape is your data in? How “dirty” is it?
  • What data would be possible to collect or analyze? How hard would it be?

Roadmap and Resource Assessment

  • How can we reach Minimum Viable Product?
  • What are the most valuable features or analyses? How can we prioritize them to deliver value quickly?
  • What resources are earmarked for this project and how can we make the most efficient use of them? What would happen if those resources were increased or decreased?

There are many ways to approach an AI project but none can drive success if the right questions aren’t asked. If you have tried to start an AI project without undergoing this kind of process, you likely have not achieved the level of return you desired. And unfortunately, you’ve most likely wasted a significant amount of time and resources.

The common mistakes we have seen are companies:

  • Spending too much time on data engineering for data that is delivering no real ROI;
  • Having difficulty in identifying which business challenges can be addressed by insights from data, and where / how to extract those insights; and
  • Generating general skepticism internally about the efficacy of such analytics techniques, and further division between parts of your business.

What is most important here is to start from a customer- and solution-centric perspective, rather than a data- and technology-centric orientation. That will save you potentially years of struggling with a mediocre AI system giving weak results. And ask yourself, would you wait for your competitors to get great results from their data? Exactly.

AI Thoughts + Insights

Case Study
Legal Document Management and Robotic Process Automation
Urvin was approached by a Fortune 500 company whose in-house legal team was struggling with the volume of legal contracts that they had to review.
Legal Document Management and Robotic Process Automation
Case Study
Pharmaceutical Prediction for FDA Phase 3 Trial
A pharmaceutical company about to begin an FDA Phase 3 trial needed to find a partner who could make predictions on a small, very noisy dataset with confidence.
Pharmaceutical Prediction for FDA Phase 3 Trial
Thought
When Artificial Intelligence Fails, the Perils of Accuracy
Despite its utility, “system accuracy” is often the wrong metric to focus on and can result in very problematic behaviors and results.
When Artificial Intelligence Fails, the Perils of Accuracy