A guide to launching an AI pilot: pick a concrete problem like matching volunteers to a 30,000-person waitlist, choose the right AI tool, then focus on high-impact repeatable tasks. A cancel-reservation agent for a travel platform handling large inquiry volumes is one example. Use Salesforce partners and clean data to move faster.
Walk with me real quick. More than 80% of AI projects fail. That's not a stat, that's a warning. Why? Most companies rush in without a plan. They try to just AI fi everything. But you need a strategy. TL Dr. Here's your guide to launching a trusted AI pilot. Step one. First things first. You must identify the problem. What are you trying to solve? One of Salesforce's customers, Big Brothers Big Sisters of America, wanted to match their adults to young people from a waiting list of 30,000 littles. Their pilot specifically focused on analyzing data and suggesting potential matches for each kid. Then a human specialist could review and make the final decision. Step 2. Generative AI chatbot or an agent. Which would you choose? Generative AI is great for creating content reports or emails. Chatbots can handle pre programmed tasks without many variables. But they can't behave any other way unless you go in and change the code. Then there's AI agents which are great at making their own decisions. Step 3. Choose a task that will have the biggest impact even if it isn't flashy. The Business Travel Management platform Engine did this with its Agent Force Pilot. Engine was receiving 550,000 customer inquiries a year, 60% of which was coming from MobileChat. They started looking for easy to manage repeatable tasks that could hand off to an agent. So what do they do? Train an AI to manage a common request cancel my reservation. Step 4. Pull in a Salesforce partner if you need help. Businesses can develop an AI agent 16 times faster with AgentForce than by doing it yourself. Engine and BBBSA both got help with their project. Spend more time using agents and less time building them. Step five. You know the saying perfect is the enemy of good. It applies to your data too. Agents definitely need decent data to train on and use for work. Use data cloud. Alright, I have to get back into it. But like for part two and the rest of the steps.
