How To Launch an Agentic AI Pilot in 10 Steps

Key Takeaways

• Identify the right use case to ensure AI aligns with business needs.
• Start with simple tasks to build confidence and competence.
• Maintain healthy data for effective AI performance.
• Collaborate with a Salesforce partner to leverage expertise.
• Adapt and iterate based on feedback to foster successful AI initiatives.

Now we are going to talk about finding the best way to use technology in our organizations. It's all about pinpointing a specific need before jumping into the tech pool headfirst.

1. Nailing the Use Case

Travis Gibson, the chief technology officer at Big Brothers Big Sisters of America (BBBSA), has a nugget of wisdom that might just save companies a lot of headaches: “Start with the problem,” he suggests, “not the flashy gadgets.”

Imagine being a part of an organization that's been matching adults (“Bigs”) with kids (“Littles”) for over a century. That’s BBBSA for you, and they’ve become quite the matchmaking pros. They currently have around 135,000 kiddos paired up, but there’s a catch—30,000 Littles are still waiting for their Big. Talk about making you feel like a kid in a candy store with no cash!

In the past, matching specialists spent six to eight weeks sifting through tons of information on Bigs and Littles, trying to find the perfect pairing. It’s like searching for a needle in a haystack—blindfolded! But Gibson and his crew thought, “Why not let an AI agent do the heavy lifting?” So, they got clever and decided to use Agentforce, Salesforce’s platform for whipping up AI agents.

Last November, they rolled up their sleeves and created the AI agent. Fast forward to March, they offered a trial run to 15 different agencies. This tech marvel doesn’t just spit out random suggestions. It analyzes the data and whittles down potential matches to about 8 to 10 choices for every kid, even explaining why each match could be a win. It’s like having your own personal matchmaker on speed dial!

Now, human specialists take the baton from here. They review the AI suggestions before making the final call. The man with the plan, Gibson, insists that this approach not only speeds things up but also allows humans to leverage their unique insights where it matters most. It’s teamwork at its finest!

• Identify a specific problem before thinking tech.
• Collaborate with your team to pinpoint needs.
• Use technology to enhance human decision-making.
• Test your ideas on a small scale before a big rollout.

So, if you ever find yourself head-spinning with technical options, remember Travis’s golden rule—dig into what you need to solve first. Technology should be like a trusty sidekick, not a stunt double trying to steal the spotlight!

Now we are going to talk about how to decide if agentic AI fits your needs. Finding the right tech solution can feel a bit like online shopping—so many options and frankly, who has the time to comb through all of them? Let’s break it down.

2. Choosing the Right AI Solution

Once we've pinpointed our issue, we've got to ask ourselves: what’s the best AI suite for the job? Should we opt for generative AI, predictive AI, a chatbot, or an AI agent? It's like trying to pick the right dessert at a bakery—every option has its merits.

Generative AI is like that friend who always tries to come up with something new, be it quirky Instagram captions, reports, or even the occasional poem. Think of it as your content creator extraordinaire.

Then we have predictive AI, which is the straight-A student of the AI family. It analyzes historical data, does the math, and says, “Hey, here’s what might happen next.” It's fantastic for making forecasts but doesn't quite have the flair for spinning out creative content.

Moving over to chatbots, they’re like the reliable old-school librarians—they can answer set questions and handle straightforward tasks, such as processing a common return. You ask, they respond—simple as pie! But let’s be honest, they’re not much for small talk. “Yeah, they can get stuck in their code,” quipped Irina Gutman from the Agentforce accelerator team. One-track minds aren’t always fun to work with, right?

• If you need personalized interactions and a sprinkle of creativity, generative AI is your go-to.
• If it’s all about predictions and analytics, lean on predictive AI.
• For straightforward queries and tasks, bring in the chatbots—just don’t expect them to win any debates!

But what if you're in need of something that can think on its feet? Enter the AI agent. This smart cookie digs deep into data, employs reasoning, and communicates naturally. Like having a conversation with a mini-you—just a meeker version that doesn’t eat your snacks! According to Gutman, “Agentic technology does require some keeping tabs on, though.” So, it’s a little like babysitting, but with fewer diaper changes!

In conclusion, navigating your AI choices means weighing your specific needs against what each type can offer. It’s truly about finding the sweet spot that aligns with your goals. Remember, whether it’s a hasty email creation or analyzing big data, we’ve got options—so let’s choose wisely!

Now we are going to discuss how to kick things off with manageable tasks that can lead to success.

3. Begin with Simple, Repeatable Tasks

Have you ever listened to someone brag about their extraordinary achievements and thought, “That sounds cool, but how do I even start?” We’ve all been there. One expert joked, “People ask me for the most innovative, high-tech solutions I've ever created, and honestly, what they really need is something so simple, it feels like picking low-hanging fruit.”

Take a look at Engine, a business travel management platform. They’re not just about fancy tech; they prioritize strong customer service, offering 24/7 support without the dreaded waiting music. Can you imagine? They handle around 550,000 inquiries a year, with a whopping 60% coming through mobile chat. That's a busy day at the office, to say the least!

As Engine rapidly grew, they faced a critical question: How do we manage all this without adding more employees? Instead of throwing more people at the problem, they decided to identify the mundane tasks that could be automated.

Joshua Stern, the director of GTM systems, shared, “We had to ask ourselves: where are the repetitive tasks? What can we program an agent to do while letting our human staff focus on more complex issues?” Great question, right?

The light bulb moment came when they recognized that one of the most common requests was simply, “Cancel my reservation.” It’s a straightforward task that can be easily automated. This frees up human agents to tackle more sensitive concerns, like a confused traveller wondering why their flight got delayed—let’s face it, we could all use a little help with that!

So, how can we ensure that the agents we create provide immediate value? Here are a few friendly tips:

• Identify the top customer requests.
• Choose tasks that allow for decision-making, like cancellations.
• Start with a pilot program; tweak it as you go!

By focusing on these straightforward tasks, we can quickly turn a cluttered workspace into a streamlined operation. And who knows? Maybe one day, we’ll be the ones sharing tales of our innovative approaches!

Now we are going to talk about the importance of clearly defining the role of your AI helpers. Think of your AI agent as a trusty sidekick in a buddy cop movie—if they don’t know their role, the plot can get messy quick!

4. Express Clear Expectations for Your AI Agent

For any mission to be a hit, be it in the skies or in a digital landscape, we need to give our AI agent some straightforward marching orders.

Take the case of a company named Engine, which wasn’t playing around when it came to its cancellation agent, lovingly dubbed Eva (Engine Virtual Assistant). They had a clear vision for what they wanted Eva to accomplish:

• Verify that the customer on the chat was authorized to make any booking changes or do so on someone’s behalf.
• Assess all travel reservations in play.
• Process the customer’s request to cancel their booking.
• Integrate the cancellation back into Engine’s internal booking system via a snazzy Application Programming Interface (API).

When we chat about roles and expectations, let’s not forget the communication aspect. As one of the team members, Stern, wisely remarked, “We also had to make sure our agent was equipped with the right messaging.”

You wouldn’t want your AI stumbling over its words like it just chugged a gallon of espresso, right? Keeping the user in the loop is crucial—nobody likes waiting on hold, let alone chatting with a robot that’s lost in translation.

Having clear functions not only sets your AI agent up for success, but it also smooths out the customer experience. Imagine being in a situation where your *helpful* AI completely missed the point of the conversation—talk about a trust breaker! It’s like ordering a fancy new gadget only to find an instruction manual written in hieroglyphics.

So, just as we would with a new employee, defining roles for our AI agents is essential for knocking it out of the park. Well-defined roles pave the way for efficiency and leave your customers feeling valued instead of bewildered.

As we embrace this digital age, let’s make sure our AI agents aren’t just robots on autopilot. Clear expectations mean fewer headaches, smoother operations, and let’s be honest—no one wants to replay awkward customer service encounters on a blooper reel!

Now we’re going to chat about data and how to keep it in tip-top shape. You know, how your fridge might look healthy with a sprinkle of broccoli, but behind it lurks a suspicious jar of pickles from who-knows-when? Yeah, data needs a little TLC too.

5. Keep Your Data Healthy

We’ve all heard that familiar wisdom: “Perfection is the enemy of good.” Well, guess what? This little nugget applies to data as well. It might sound like a cliché, but sometimes chasing that elusive “perfect” dataset stops us from using what we already have. Let’s take a moment and think about what Gutman said—it’s not about having an encyclopedia’s worth of data. Instead, having a solid chunk of clean, organized data is where the magic happens. Imagine this scenario: You’re in a fancy restaurant, scanning the menu and trying to decide between the lobster and the steak. But instead, you receive one of those hyper-detailed, 10-page menus. You'd feel overwhelmed! Just like in a restaurant, too much data can leave your team scratching their heads. For a great example, we can look at BBBSA, a nonprofit that decided to rent a cozy little corner of the Salesforce universe for its tech strategy. They had a well-oiled machine for hiring, budgeting, and matching—like a well-tuned engine humming along smoothly. Because of this, their data wasn’t just hanging out in a digital shoebox; it was neatly packed, making it easier for them to get to work. Thanks to Salesforce's data solutions, they had everything in shape. But the cherry on top? It turned out their unstructured data was a hidden gem for making those perfect matches between “Bigs” and “Littles.”

• Focus on cleaning your data before you focus on gathering more.
• Your existing data can often do more than you realize.
• Use tools that help you manage and integrate information seamlessly.
• Don’t let the search for perfection stop you from moving forward.

So, the next time you’re caught up trying to craft the “perfect” dataset, remember: even the best superheroes leave some of their capes hanging up in the closet. It’s all about being ready to use what you have and sprinkle in a little creativity. Your data cleanup can lead you to surprising insights—so let's make that clean data project happen!

Now we are going to talk about the importance of having the right boundaries set in place when using AI tools. It’s like having a toddler in a candy store—delightfully chaotic, but definitely needs some rules!

6. Set Clear Boundaries

Just as a seasoned sailor knows when to avoid rocky shores, any agent needs to grasp both the do's and the don'ts. Thankfully, if we’re chatting about Agentforce, some of those boundaries are already waiting in the wings. Salesforce has its Einstein Trust Layer that cuts out bias and other unwelcome guest appearances. Think of it as a bouncer at your local pub—no intoxicated nonsense allowed!

However, every organization must roll up its sleeves and create its own set of guardrails. “When we were gearing up for our pilot,” shared Gibson, “we spent a hefty amount of time reviewing our standards of practices.” And what are those, you ask?

• Criteria for matching, such as gender
• Lived experiences that resonate with both parties
• Common career aspirations and hobbies

Basically, BBBSA was cooking to ensure that their "Bigs" and "Littles" got along like peanut butter and jelly!

To further secure the matchmaking process, BBBSA laid down several boundaries. For instance, they established that matches need to be of the same gender. Some agencies piloted for additional filters like age or religious backgrounds, and they wisely added those into the mix. But the biggie? The agent could only toss out recommendations. When it comes to pairing a Big with a Little, a human being always gets the final say. You can’t put a price on that personal touch, can you?

We could all use a bit of that caution in today’s tech-driven society. In a time when misinformation spreads faster than gossip at a family reunion, it’s crucial to lay down clear rules. Whether in AI or any venture, a good safeguard saves us from the chocolate river overflow we didn’t see coming!

As we move forward, let’s keep in mind that implementing some well-thought-out boundaries opens doors to better outcomes. And isn’t that what we all want? A smooth sailing experience, minus the chocolate rivers!

Now we are going to talk about how collaborating with a Salesforce partner can simplify your journey into AI implementation. It might sound a bit like hiring a contractor when you want to put up a new fence—sometimes you need extra hands on deck.

7. Looking for Assistance? Team Up with a Salesforce Partner

Your tech wizards might think they have everything sorted to launch Agentforce, especially after recent research reveals that companies can whip up an AI agent 16 times faster using Agentforce than crafting one from scratch. But let’s be honest, sometimes an extra pair of hands can help when juggling those flaming torches of innovation. That’s where your trusty Salesforce partner comes in. These heroes are part of the Salesforce family and often know the ropes better than anyone. For example, take the adventures of Engine and BBBSA—they both hitched rides on the partner bus to success!

• Engine teamed up with Astound Digital last October to give birth to “Eva” in under two weeks. Talk about a fast delivery!
• On the flip side, BBBSA brought Coastal Cloud into the mix. Their process took about two months, but as Gibson puts it, “That was the beauty of the solution.”
• By bypassing the whole “let’s hire AI experts” route, BBBSA leaned on Coastal Cloud’s Salesforce solution architect and savvy administrators. This not only saved time but also a chunk of change!

To find your very own Salesforce partner, don’t hesitate to reach out to Salesforce’s professional services team. They can guide you on this thrilling expedition or you can follow this handy guide.

So, whether you need a quick fix or a steady hand to guide your project, remember there’s always a partner ready to lend a hand. And hey, sometimes asking for help isn’t a weakness; it’s just smart teamwork!

Now we are going to talk about how we can adapt and refine our expectations for an AI program as it grows. Much like learning to bake a cake, it’s all about mixing ingredients, adjusting the oven temperature, and occasionally licking the spoon—which, by the way, is the best part. Here’s how we can get comfortable with making adjustments along the way.

8. Embrace the art of adaptation

No one ever said running an AI pilot program would come with a manual—like assembling furniture from that infamous Swedish store. It’s crucial to keep tweaking things and making adjustments. For instance, when Engine partnered with Astound Digital to develop Eva, they quickly realized that the initial week was just a warm-up lap.

Imagine spending days building a perfect model and then realizing that it crumbled under pressure. Stern and the team dedicated an extra week to back-end testing, trying to break their creation, much like a kid testing their sibling's Lego fort. They scrutinized every nook, investigated why things failed, and fixed prompts that went haywire.

It’s a learning curve, folks! And once they went from “oops” to “ah-ha,” they kept a watchful eye on how their AI was performing. They noticed the agent-to-human transfer was a tad clunky—like trying to pass the baton in a three-legged race. Human agents needed to quickly catch up and not spin in circles asking customers the same questions over and over. It’s like that awkward moment at a party when you realize you’ve already asked someone about their pet goldfish twice.

To make the experience smoother, here are some key takeaways that can help us improve our AI programs:

• Test rigorously. Break things intentionally to understand where improvements are needed.
• Monitor the performance. Regularly check how your AI is interacting with users. It’s like checking your tire pressure—pretty boring, but necessary.
• Refine the handoff. Ensure human agents are well-prepped and ready to engage, so customers feel like they’re in good hands.
• Collect feedback. Whether it’s from customers or your team, like a well-placed suggestion box, keep the channels open.

In the end, adapting to new information and feedback is half the battle. It’s like trying to dunk a basketball—practice makes perfect. With proactive tweaking, the team could deliver stellar customer experiences and fine-tune the AI like a well-tuned saxophone. It's a harmonized effort, really! So, let’s roll with the changes and keep those improvements coming.

Now we are going to discuss how measuring success in an AI pilot can be both interesting and revealing. After all, it’s not just about seeing pretty graphs; it’s about what those numbers say about us! We’ve all been there, right? Seeing a statistic can make you feel like a math whiz… or send you spiraling into existential dread. Let’s break this down a bit.

9. Evaluating Your AI Pilot Success

To figure out if your AI trial is on the right track, you’ve got to pay attention to the digits. Sometimes, we need to dig a little deeper, as Gutman reminds us, to interpret those shiny figures beyond just surface value. It’s like opening a box of chocolates—looks great on the outside, but what's lurking inside?

Take Engine, for example. They were quick to roll out their new assistant, Eva, in a mere two weeks! Now, Eva is sauntering through cancellations like it’s a walk in the park—dealing with around 30 to 40 cancellations a week! That’s a big sigh of relief for human agents who can now focus on more complex tasks, rather than drown in a sea of cancellations.

Then there’s BBBSA, which is wading through its six-month pilot and gathering whatever data it can get. They’re like detectives, trying to crack the case of effective matching. They’re especially curious about whether their AI can perform fewer steps to pin down potential matches. And yes, they’re on the lookout for the quality of those matches, because just matching up two parties quickly doesn’t cut it. It’s like setting two people up on a blind date and hoping for the best—without checking if they even like the same pizza toppings!

Gibson chimed in, hopeful that they’d cut match times down to half, but he also pointed out a crucial factor: “We’ll take our time with this solution to see whether match retention is on the rise.” So, they’re not just racing to speed things up; they're making sure those matches stick! It’s about building relationships just as much as it is about quick fixes.

The reality is, when examining AI’s effectiveness, we need to consider:

• The volume of tasks handled
• Time saved in operations
• Quality of outcomes and satisfaction rates
• Changes in engagement and adoption

At the end of the day, measuring the success of an AI pilot isn’t just about crunching numbers; it’s weeding through the details to figure out how it suits our needs better than a well-tailored suit. Balancing speed and quality is a dance, much like trying to assemble IKEA furniture without the instruction manual. But hey, isn’t that what makes it all so engaging? Cheers to maintaining a keen eye on those metrics!

Now we are going to talk about scaling your success, especially after you’ve had a promising pilot run. It’s like baking a cake: one layer at a time, friends!

10. Ready to Expand?

So, imagine you’re at a party and your homemade cake is the hit of the night. What do you do next? You might want to bake more, right? That’s exactly what groups like BBBSA are doing after their successful pilot. 

They’ve cooked up a plan to onboard about 10 to 20 agencies each month over a six-month stretch. Talk about a sweet rollout!

Let’s chat about Engine, which is taking things nice and slow, like savoring that first bite of cake. They recently managed to make Eva their cancellation agent, but wait, there’s more! In April, they decided that Eva could also tackle FAQs, cutting down the need for a human to answer those simple questions. Who knew a cake could be so versatile?

Now, Engine is rolling out another pilot—cue the confetti—a second AI agent that’s here to analyze all the fabulous work Eva is doing. The new agent not only checks which prompts are the best but also how quickly Eva is responding. It’s like having a baking assistant who not only helps prepare the cake but also gives feedback on your frosting technique! 

Stern from Engine reminds us all, “Keep in mind this is just the beginning!” So true! Imagine all the layers they have yet to bake.

Effective scaling isn’t about rushing into the next phase but ensuring each layer of your strategy is solid and delicious. How many of us have bitten into a cake only to find it's not quite baked through? Yikes! Taking the time to assess and refine each aspect lays the foundation for what’s to come.

• Start with a successful pilot.
• Gradually onboard new teams.
• Evaluate performance continuously.

So, whether you’re baking cakes or rolling out AI agents, let’s take a cue from Engine and BBBSA. Slow and steady, folks! Scaling SMART is the name of the game.

Now we are going to talk about how to successfully run pilot programs for agentic AI. It’s like baking a cake; if you throw everything in a bowl without a plan, you might end up with a gooey mess instead of a delicious dessert. We’ve all been there! So, let’s dish out a more structured approach to piloting AI.

A Plan for Successful AI Pilot Programs: Slow and Steady Wins the Race

When it comes to launching a pilot program, patience really is a virtue. It’s tempting to charge in with all guns blazing, hoping for immediate results. But let’s be honest—most things in life require a little finesse. For example, did you ever try to assemble furniture from that Swedish store? They make it look easy, but I once ended up with a bookshelf that resembled modern art rather than functional furniture.

Running an AI pilot is no different. Here’s a simple roadmap to guide us:

• Define Clear Objectives: What’s the goal? Is it to increase efficiency, reduce cost, or perhaps even amaze your team with nifty tech? Clarity is key!
• Select the Right Team: Having the right folks on board is like choosing your dance partners. A good mix of skills keeps everyone in sync, avoiding any two-left-feet disasters.
• Start Small: Think of this phase as sampling the buffet instead of filling your plate to the brim. Test with smaller datasets and environments first.
• Iterate: Feedback is like gold. Just like their unflattering opinions on that quirky painting you thought would be a hit, take it into account to adjust your approach. 
• Measure Success: Set metrics to evaluate performance. Did it meet your expectations, or did it just collect dust like that fancy kitchen gadget you swore you’d use?

Now, this strategy isn’t carved in stone. We can adapt as we go, much like adjusting our coffee order based on the changing seasons—sometimes we need those pumpkin spice lattes to kick off autumn! Plus, current trends show companies across various industries, from healthcare to finance, leveraging AI. Let's take Walmart’s recent foray into AI-driven inventory management as an example. It’s all about adopting these technologies in a way that fits our unique situations.

And let’s not forget about user training! This can be a bit like getting a cat to take a bath—tricky but necessary. However, once people understand and feel comfortable with AI tools, it’s a game changer for productivity. So let’s embrace these innovations thoughtfully, ensuring we’re not just throwing spaghetti against the wall to see if it sticks.

In summary, approaching AI pilot programs with a well-thought-out plan, a pinch of humor, and the willingness to adapt will help us avoid pitfalls. After all, nobody wants a kitchen nightmare! With the right strategy, we’ll whip up a successful AI initiative that truly leaves a lasting impression.

Conclusion

FAQ

• What is the first step in utilizing technology effectively in organizations?
  Start by identifying a specific problem before considering technology solutions.
• How does Big Brothers Big Sisters of America utilize AI in their matching process?
  They use an AI agent developed on Salesforce's platform, Agentforce, to analyze data and suggest potential matches for kids and adults.
• What are the different types of AI options mentioned in the article?
  Generative AI, predictive AI, chatbots, and AI agents are the main types discussed.
• What should organizations focus on when starting with AI implementation?
  They should begin with simple, repeatable tasks that can be easily automated.
• Why is it important to define roles for AI agents?
  Clear roles and expectations ensure that AI agents can function efficiently and enhance customer experience.
• How can organizations keep their data healthy?
  By focusing on cleaning and organizing existing data rather than constantly seeking perfect datasets.
• What are essential criteria for setting clear boundaries for AI tools?
  Organizations need to create their own standards of practices such as matching criteria and ensuring human oversight.
• How can partnering with Salesforce benefit organizations?
  Salesforce partners can provide expertise and accelerate AI implementation processes.
• What factors should be considered when measuring success in an AI pilot?
  Consider the volume of tasks handled, time saved, quality of outcomes, and changes in user engagement.
• What is the recommended approach for scaling AI implementation?
  Organizations should start with a successful pilot program and gradually onboard new teams while continuously evaluating performance.