Navigating the AI Journey – Volume 2: Exploration
Download this perspective
In our previous perspective on Navigating the AI Journey, you'll recall that we dove into the intricacies of the Advisory phase. But having a well-advised AI strategy is just the beginning; the rubber really hits the road in the Exploration phase. Here, people mobilize and move fast, not in a haphazard manner, but through a series of structured, high-impact steps designed to produce value from your AI initiatives.
The Exploration phase is where most companies stumble, tripped up by data issues, unclear priorities, or a poor understanding of how to create and measure value.
To help you successfully navigate this critical phase, we're sharing the precise steps we've honed through our experience delivering AI projects for businesses. These steps are designed not merely to guide, but to act as an accelerator to your AI journey, ensuring you traverse it with fewer hiccups, more purpose, and absolute clarity.
The AI Journey
This volume is dedicated to unraveling the intricacies of the Exploration phase, breaking it down into manageable, actionable steps that yield measurable results. Just like its predecessor, this installment is based on the real-world rigors of 200+ AI projects, and it forms an integral part of our AI Journey framework. So, without further ado, let's jump in.
What is Exploration?
The Exploration phase of AI solution delivery is the first time where abstract ideas become concrete and if approach correctly, where strategy translates into tangible results. The goal here is simple yet highly significant: build AI solutions and deliver value to your business as swiftly and efficiently as possible.
AI Use Case Ideation
The Exploration phase kicks off by convening a small, cross-functional team that combines business acumen with AI expertise. The primary task of this team is to brainstorm potential use-cases for AI within your organization, and then to prioritize them. The idea is to tackle the most impactful projects first, ones that promise to drive financial results and, in some cases, revolutionize how you do business.
Getting Started - Team Composition
The objective is to assemble a small team of leaders who possess an intimate understanding of your business operations. While it can be beneficial to have a multifunctional team present for workshops, especially if the AI program is organization-wide, focusing on one business function at a time is often more effective. It enables a deeper dive into the nuances of each function, maximizes the time available for brainstorming, and ensures that ideation can conclude adequately by the end of the session.
Idea Generation – What can AI do for you?
Ideation always begins by defining the team’s mandate. What specific metrics or KPIs are they responsible for driving up or down? For instance, if the focus is on Customer Success, the team might explore ways to address resolution times, support call volumes, or customer satisfaction.
Initially, the conversation should steer clear of any implementation details or discussions about the type of AI technology that could be applied. The goal is to focus on the business problem first, not the solution. Start by writing down and sharing use cases and solutions that would make the biggest impact to your team’s metrics.
Moving from Ideas to Methodologies
Once the whiteboard is filled with potential use-cases, the team shifts its focus toward evaluating the applicable AI methodologies.
Considerations when evaluating AI Use Cases
Value Creation Potential
Not all AI use cases have the same value creation capacity and the value that a particular use case has for you depends entirely on the circumstances of your business. Identify which lever the use case drives (i.e. increase revenue, decrease cost, or increase valuation multiple) and estimate the impact.
Data Availability
Being overly optimistic about the data you have can derail the project before it begins. Make sure to realistically assess the data that is available for the project and how critical it is to the project’s success.
Technical Feasibility
Based on previous, assign a complexity category to each use-case:
Pre-built AI: Existing or open-source models that can be directly utilized (e.g. OpenAI APIs, downloadable models from Hugging Face, etc.)
Certainly Buildable: Projects where there’s high confidence in the feasibility based on previous experiences and data availability
Unknown Feasibility: These require a proof-of-concept to first establish whether building the AI model is feasible and the performance is satisfactory
Scoring and Prioritization
Finally, each use case should be scored on three dimensions: value creation potential, data availability, and technical feasibility. A simple high, medium, or low scale can be used to plot your use cases and quickly identify which projects offer the best starting point for your AI journey.
Exploration Delivery Principles
When you are ready to move from ideation and use case prioritization to delivery, you need to start putting the teams, tools, and processes in place. The precise tactics that are used execute an Exploration vary based on the nature of the problem, team, use case, and technology. However, across all the Explorations that our team has led there are few common threads that connect the successful ones.
Lean Team Structure - A bloated team will slow down progress, so the team is kept as compact as possible
Iterative Development - Employing agile methodologies, the team builds product iteratively, allowing for flexibility and adaptability.
Build for value and nothing else - No effort is spent building components not directly related to the project’s main value proposition. Peripheral features that we call “solvable problems” can wait.
Cost Control - Modulate investment as you build based on progress and what you are learning. Whether it's ramping up investment to accelerate development or cutting it off to mitigate losses, financial flexibility is a key asset.
The Exploration process shares many features with agile methodologies used in software development. However, it's crucial to note that this approach is particularly suited for AI projects, which have often been mishandled when subjected to traditional, waterfall-style project management. In today's fast-paced technological and economic landscapes, such sluggish and inflexible methods simply won't cut it.
By adhering to the Exploration phase's principles and strategies, you not only steer clear of common pitfalls but also turbocharge your path to AI-driven transformation. Through careful planning, agile execution, and a razor-sharp focus on delivering value, you can turn your AI ambitions into tangible business outcomes.
Selecting Your Delivery Team
The composition of your Exploration delivery team is sometimes a better predictor of success than the amount of available data, the chosen use-case, or even the maturity of the supporting technologies. Therefore, thoughtful selection of the team is imperative for the successful execution and realization of your AI strategy. In our experience assembling teams, there are several considerations to keep in mind:
Team Size – Teams should be as lean as possible. When you are solving a novel problem, more minds are not always better than a few sharp ones.
Business Understanding – Someone on the delivery team must have a solid understanding of the business and constantly reassess how the technology being built relates to the business value you're trying to create.
Entrepreneurial Mindset – The team should approach the Exploration as if they were building their own business around it. This requires intense curiosity for the problem, a clear articulation of the value proposition, and the agility to reassess direction and pivot when necessary.
Technical Flexibility – The specific technology needed to solve a problem will not be known at the beginning of an Exploration. The team should be technology/vendor-agnostic and experienced across different technologies to find the best toolset for the job.
Technical Expertise – The team needs deep expertise in AI and machine learning and should have experience delivering real AI-powered products. This speeds up the early stages of the project and helps avoid common pitfalls.
These characteristics apply equally to companies that have machine learning resources internally and those who need an agency to help them with delivery. If you have internal machine learning resources, that's advantageous, but consider that not everyone on a machine learning team is well-suited for Exploration-style projects. On the other hand, most small to medium-sized businesses will not have an internal machine learning team, making it crucial to evaluate potential agencies and vendors (see tips on picking an Exploration Partner on the right).
An Exploration is a mentally intense engagement and for your company to have success you need to ensure that your partner can be flexible in the way that value is delivered, be dedicated to solving problems quicky, and provide a mechanism to wind down costs quickly if needed.
Tips for picking an Exploration vendor
Sign a retainer agreement that allows for flexibility in scope. A statement of work based on a defined scope does is too rigid to be successful.
Ensure your partner is assigning dedicated resources. Explorations are not side-of-desk or 9 to 5 jobs. They require a team’s full attention.
Negotiate reasonable termination terms in your retainer agreement. One of the benefits of an Exploration is that you can cut investment quickly. Do not allow yourselves to be locked into multi-month commitments if the time has come to shut the Exploration down.
Solution Design
When the time comes to design the AI solution, preparation is key. Having the right inputs and involving the right stakeholders can mean the difference between a hit and a miss. As you prepare to dive into solution design, ensure you have these inputs in hand:
Detailed Description of Business Process – Documentation is a good start, but a meeting with the person who oversees or works within the target process will provide your team with a more nuanced understanding of the problem at hand.
Sample Data – Having access to some sample data early in the design process is critical. It will prompt the right questions about the state and format of your data. Whether it's single-sheet CSVs or multi-tab Excel workbooks, or even text-based PDFs versus image-based ones, the type of data you have will significantly impact the design.
Evaluation Criteria – Know how you'll evaluate the success or failure of your Exploration. This will influence the design and make sure everyone is aligned on the objectives.
As the design process kicks off, we have often found it helpful to start by asking the question, "How will we tactically measure this solution’s value?". When you start with that question, you will find yourself considering many possibilities. Sometimes you will need a user-friendly interface to onboard users and showcase what you've built. Other times it is enough to build machine learning models and generate static outputs into Excel spreadsheets to analyze the results. And other times you might release a series of APIs that allow other developers to interact with the AI system you have built and give you feedback.
Regardless of what form your solution takes, it is equally important to ensure that your Exploration design is solely focused on validating the "hard" parts of the problem. We cannot emphasize this enough: spend your time and resources working on the hard things up front and leave the solved problems for later.
Exploration projects promise to deliver value at the end of every sprint, and this only happens if everyone is marching in the same direction and there are minimal missteps. Involving the entire team in the design process helps enormously. While not everyone needs to contribute to the design equally, each member should have the opportunity add their thoughts. The result is a shared understanding of the project's scope, and it highlights potential design gaps before they impact implementation.
Lastly, keep in mind that a picture really does paint a thousand words. Wherever possible, visualize exactly what you are working towards. If you are building a solution with a UI, produce a low fidelity mock-up. If your solution is several backend services connected to one another, draw a system diagram and highlight how data will get manipulated as it moves through the system. If you are going to produce a written deliverable, create the shell or template at the very beginning. Having visual aids can clarify the Exploration’s vision and align everyone on what needs to be built — and how it will provide value.
Exploration Delivery – Rapid and iterative development for AI
An Exploration delivery methodology for AI borrows many principles from the agile development approach. These principles encourage rapid, iterative development, allowing you to adapt to changing conditions and to continually refine your approach in the face of new information.
Exploration delivery is segmented into sprints that usually last between 1-2 weeks. These sprints incorporate agile rituals like sprint planning, daily scrums, backlog grooming, and sprint retrospectives. These rituals keep the team engaged and allow the project lead to constantly adjust the team’s direction.
The objective of each sprint is to create and demonstrate tangible value. This means that the endpoint of each sprint should not just be a status deck. Instead, you should be able to demo a new feature, showcase a machine learning model making live predictions, or share a crucial conclusion that informs your next steps or challenges your prior assumptions.
At the outset of each sprint, a kickoff meeting is held for the entire team. Here, the project or product lead lays out the specific objectives and expected deliverable for the forthcoming sprint. This meeting provides the delivery team an opportunity to ask questions, clarify their understanding, and align on a delivery approach before diving deep into implementation. But once the sprint begins, calendars are cleared aside from daily scrums so that technical teams can focus solely on building towards the next milestone.
About 2-3 days before a sprint concludes, the team will regroup to determine what can be presented or demoed. The impending presentation often acts as a catalyst, accelerating productivity and inspiring a higher caliber deliverables.
At the end of each sprint, a thorough review takes place. Project stakeholders examine the project's progress in relation to their predefined objectives. They evaluate whether to stick to the planned course or pivot, and whether the Exploration has achieved its intended objectives and delivered the promised value. If not, they can decide to kick off another sprint or end the Exploration altogether. This constant reassessment allows the team to quickly navigate uncertainty and change course as needed.
What Next – Your next step after you have completed an Exploration
Once an Exploration has concluded, the next phase is typically a pilot, putting the developed solution into the hands of a subset of users for further testing and assessment of the value created. However, for various reasons, Explorations may reveal that the solution doesn't offer enough value to justify further investment. It is important that Exploration delivery teams and leaders do not see these situations as failures. Explorations have uncertainty built into them by nature and there is tremendous value in an initiatives that allows an organization evaluate the value of AI in a quick, thorough, and efficient manner.
Navigating the complexities of building AI solutions is a formidable challenge, particularly if your organization lacks in-house expertise in this field. The traditional waterfall-like approach to AI projects does not resonate with organizations anymore. In a world where capital is expensive, leaders cannot afford to invest in year-long research projects just to determine the feasibility of a concept. Even after crossing this hurdle, only a fraction of these projects transition from being "feasible" to delivering tangible value. This creates a risky dynamic where considerable upfront investment is required, with minimal control over escalating costs and limited flexibility to adapt as the project evolves.
Exploration addresses the pain points of traditional AI projects head-on. It places companies firmly in the driver's seat of AI delivery.
Unlike traditional methods, Exploration is inherently agile. It adopts rapid, iterative cycles that not only accelerate the pace of development but also offers the flexibility to course-correct in real-time. This approach minimizes financial risk while maximizing the opportunity to pivot or refine the strategy based on real-world feedback and results.
In essence, Exploration is not just another methodology; it is the leanest, most efficient way to build AI solutions that provide actual value. By embracing the principles and strategies outlined in this piece, you are not just undertaking a technological venture; you are embarking on a strategic journey to make your AI aspirations a value-driven reality.