Originally written on Alltius blog: Blueprint of a successful AI pilot for tech leaders
Generative AI is all the rage, with every technological leader dipping toes into it. A survey conducted by Deloitte and Data Foundation revealed that 55% CDOs are using basic or advanced AI already and 99% of them are considering implementing AI in various capacities in their organization next year.
The reason: The promise of AI driving value in their organizations is immense.
But before going to the full blown out phase, it’s imperative to provide a “proof of concept”. These small scale initiatives provide an efficient platform to test out the technology before scaling it across the organization. But, that’s where things get tricky.
According to IDC, almost 50% of AI pilot projects fail. The main reasons being cost of AI solutions, a lack of qualified workers and unrealistic expectations.
All of these could be mitigated with meticulous planning. I’ve overseen multiple AI pilot projects as part of Alltius and as a consultant to Fortune 100 companies, and I’ve created a framework for planning a successful generative AI pilot and some tips to ensure your AI pilot project is successful.
5 steps to a successful AI pilot
As a technology leader, you need to find a systematic way to navigate the chaos around AI. Instead of spreading yourself thin across every other use case, you need to identify, prioritize and drill down on one use case to validate the ideas in practice. Here are the steps I follow. We’ll go deep into every pointer mentioned in the image.
5 steps to a successful AI pilot
Ideation phase
Every tech project should start with business goals. Have a clear understanding of where the organization currently stands and where it aims to go. What are the metrics that align with business goals? Eg. revenue, operational efficiency, risk reduction, conversion rate, churn.
Identifying these goals will be crucial in designing a pilot that’ll help business navigate from current to desired state. Now once you have the goals, try to understand which part of the organization you’d want to implement AI in. Example, the goal could be to increase operational efficiency which can be related to marketing, finance, customer support, sales or product departments. You need to choose which cog you want to focus on.
For this blog, we’ll take an example of customer experience. In case, you’ll have to analyze bottlenecks in both internal and customer-facing processes. Here are some examples:
Internal Bottlenecks:
- Customer Research: How can AI uncover deeper customer insights?
- Customer Support Assistance: Can AI enhance support efficiency?
- Self-Service: Could AI streamline user interfaces for better customer autonomy?
- Admin Tasks: Where can AI take over routine tasks to free up human creativity and problem-solving?
Customer Facing Bottlenecks:
- First Touch: How to make the first customer interaction as impactful as possible?
- Sales Call: How to improve sales conversion rates?
- Purchase: How to streamline the purchasing process?
- Onboard: How to reduce on boarding inefficiencies?
- Engage: How to keep customers engaged?
- Retain: How to reduce churn?
- Advocate: How to get referrals?
Sit with the department stakeholders to identify the bottlenecks and identify the metrics that are needed to measure them. Once done, it’s time to start ideating. You can expand the ideation phase to get inputs from multiple sources.
- Running poll in your company: Circulate a survey across the organization to collect AI use case ideas to collect a diverse pool of ideas that match your business goals & bottlenecks.
- Analyzing other companies: Study how other companies are leveraging AI in their processes. This will help you expand the horizon of possible use cases and avoid reinventing the wheel.
- Consulting AI experts: You can also reach out to AI experts who can help you this entire process starting with ideating use cases that are more feasible and can be solved with current AI capabilities.
Prioritization phase
Assuming you had a successful ideation phase, you can end up with a sea of use cases. It’s imperative to avoid the quagmire of endless possibilities. The next step is to: How to assess which use cases are truly feasible and valuable at this point of time. You’ll need to evaluate each use case with added level of scrutiny.
In order to prioritize use cases, we will map them on a 2x2 matrix with business value and feasibility as the axes.
Business Value: In the first step, we identified the strategic business goals. We will evaluate how our selected use cases fit into the strategic goals & bring value to the organization in terms of improvements of selected metrics.
Feasibility: You need to ensure the pilot AI project is built to scale in the future. You need to ensure that after the pilot is successful, you’ll be able to demand a budget and scale the AI to the next level with available resources. You need to ask yourself the following questions:
- What resources do you need to scale the project?
- What are the setting up & maintenance costs?
- Over what time period will your scaled AI project derive value?
- What kind of risks will be associated with it?
- What kind of additional training will you need for your stakeholders?
Proceeding without a clarity on these is a classic mistake.
Now let’s see how you’ll map them. In order to prioritize use cases, list down the dimensions for all your business and feasibility factors. Now, assign weight according to the business importance and availability of the resources.Make sure all the weights add up to 100%. Take help of other stakeholders to determine the weights. Now, mention your use cases and provide a score from a scale of 0–5 for all the dimensions.
Step to prioritize your gen AI use cases
Now, for the last part, calculate the business and feasibility score using the weighted scores. We’ve created a spreadsheet that automatically calculates it for you.
How to calculate the business and feasibility score
The image provides a scoring system that quantitatively assesses the impact and feasibility of each use case, aiding in making informed decisions.
With the weights, the next part is to map them on a 2X2 matrix like one shown in the image. In the first pilot, focus on the use cases present in quadrant 1.
Mapping business score vs feasibility score
Pro Tip: It is important to remember the scores are just proxies. They may differ & therefore, work along with other experts to remove biases.
Selecting Team and Technology
Your pilot team and the AI technology can make or break your AI pilot. These are crucial elements of your success. Your pilot AI team should be diverse and comprise of all of the following people:
- Business unit members — These are the individuals whose day-to-day operations will be directly impacted by the AI implementation. By including them, you create a continuous feedback loop, ensuring that the AI solutions are pragmatically aligned with actual business needs and workflows.
- Product experts — These experts (product managers or software engineers) understand the nuances of your product and are crucial in planning the user interface, front-end application, and scalability support. Their expertise ensures that the AI solution is seamlessly integrated into existing products and services.
- AI experts — Internal or external -AI is a specialized and rapidly evolving field. While it may be challenging to have in-house AI experts, especially in non-tech companies, their expertise is indispensable. If internal expertise is lacking, it is advisable to collaborate with external AI platform providers like Alltius. These experts bring in-depth knowledge of AI technologies and can guide the project towards success.
The next decision is: to build or to buy the AI technology. In the image I’ve summarized the different decision parameters. Based on your requirements, you can make a decision. If you’re looking to pilot a use case right now, it’s better to go with buying a platform to test out the waters and see if it works.
Build vs buying AI technology
If you’re exploring platforms, I would encourage you to try out Alltius. With Alltius, you can create skillful AI assistants for sales and support teams that can 2X sales & halve the support costs.
Alltius helps sales agents sell more with an AI-driven assistant that crafts personalized sales pitches, outreach emails, cross-sell pitches in seconds, aligning perfectly with customer needs to improve conversion rates.
Alltius helps customer agents close tickets with the AI-driven assistant that understands customer query, finds root cause & drafts accurate answers to solve customer query within seconds, reducing customer and support agent efforts.
We offer a free trial. In case you’d want to see it in action, book a demo with me.
Design
Now, in the next step, we will start designing the generative AI solution. There are three core elements that form the bedrock of this step: Input, Test, and Measure.
AI is as intelligent as its input. Ensure you make a list of all knowledge sources, skills and channels where you want to deploy your solution. These inputs will lay the groundwork for a capable AI system.
In the next step, plan how to test your AI. Plan extensive test scenarios to ensure better performance. At Alltius, we plan 6000+ test scenarios to find accuracy on multiple tasks before even going live. By deploying extensive test scenarios and using precise evaluation metrics, we push the AI to learn and adapt.
Finally, we ‘Measure’ to see our AI in action. Define metrics to track performance like accuracy, responsiveness, customer satisfaction, or any other metric of choice and see how we are debottlenecking the business. This will determine the AI’s impact on operational efficiency and business outcomes.
Iterate
The last step is to set everything in motion with iterative stops based on stakeholder inputs. Be sure to actively involve stakeholders to fine-tune your AI based on their feedback at regular intervals of 2–3 weeks. This build-test-iterate loop is key to aligning the AI with user needs and assessing its performance against KPIs.
After a certain time period, generally within 2–3 months, you’ll have a clear idea of the performance of the pilot project. And then comes the last decision: to scale or not?
You can stop, if the AI doesn’t meet the use case requirements, document what worked and what didn’t. Then, circle back to ideation for new use cases. You should scale if the AI shows promise, plan for expansion, considering features, architecture, costs, and investments needed for broader implementation.
I’ve combined this entire framework in a compact PDF which is easy to review and share. You can download your copy here.
Parting tips
Here are some pro tips to keep in mind:
- Make sure the data sources are of high quality and reliable.
- Regularly review and update your AI ethics guidelines, and use diverse datasets to remove bias.
- Work with stakeholders and end users to see what changes will improve their experience while defining goals.
- Make your AI pilot team diverse — with stakeholders, engineers, PMs and a business leader.
- Set measurable KPIs to demonstrate impact of your AI project while presenting the case for budget allocations.
- Invest in AI technology that provides you competitive advantage.
- Work with trusted AI experts to maximize value from your AI projects.
Conclusion
The secrets of a successful AI pilot lies in meticulous planning & focused implementation. As technology leaders, your role is to steer this chaotic path with a clear vision. Follow the framework to turn your AI projects into valuable assets for your organizations.
In case you’re looking for assistance for implementing AI at your organization, feel free to book a free demo call with our AI expert team.
Originally published at https://www.alltius.ai.
