Your organization leaders get excited a few scorching new AI instrument or determine to create their very own. They make the funding to launch an agentic AI pilot. Staff check out the instrument in a devoted mission after which … it by no means will get deployed.
Sound acquainted? In that case, you’re not alone.
A current report from MIT exhibits that though enterprises have invested $30 billion to $40 billion in generative AI, 95% of them are getting zero return. Solely 5% of AI pilots are displaying worth — providing tens of millions of {dollars} in price financial savings or different advantages — “whereas the overwhelming majority stay caught” with no measurable monetary affect, the authors famous.
In the meantime, Futurum* discovered that 96% of chief data officers (CIOs) think about synthetic intelligence (AI) adoption a prime precedence — but most firms wrestle to get past the pilot section.
Why accomplish that many agentic AI tasks fail to launch? Let’s have a look at among the prime causes, and concentrate on how one can transfer ahead to deployment.
We’ll additionally share how firms have efficiently deployed Agentforce, the Salesforce platform for constructing AI brokers. In the event you’re fascinated by launching a pilot, their experiences can assist you chart a course to full deployment.
1. You’re specializing in the tech, not the issue you need to clear up
It’s straightforward to be dazzled by agentic AI: What firm doesn’t need to reap all the advantages that it guarantees? However when firms focus extra on the expertise than the issue they should clear up, they set themselves up for failure. The truth is, 60% of AI pilots get caught as a result of they fail to ship a transparent return on funding (ROI), in keeping with Futurum.
How are you going to set your pilot up for fulfillment? Flip that focus round and focus on what purpose you’re attempting to realize. What are your organization’s greatest ache factors? What’s conserving you up at evening? Then, slender down your checklist to discover a tightly targeted, achievable use case that you just’d like to check with an AI agent. “You may’t get overwhelmed by wanting to unravel all the issues,” mentioned Jim Roth, president of buyer success at Salesforce.
For instance, a analysis establishment may be short-staffed. It has a backlog of tasks and is struggling to display sufficient members for scientific trials. Of these ache factors, it decides to zero in on probably the most particular one: screening. So, it launches a pilot, utilizing an AI agent that’s educated to ask sufferers the correct questions and assist decide eligibility.
What you are able to do: Our free Agentic Enterprise Playbook can assist your workforce align on the correct methods to maneuver rapidly. See the place you land on the agentic maturity spectrum, and construct your technique on 4 key pillars. Then, obtain our AI imaginative and prescient assertion worksheet for a workforce planning exercise.
2. Your online business workforce isn’t concerned within the planning course of
While you’re planning to launch an AI pilot, you want technical consultants on the workforce to evaluate the mission’s feasibility. However enterprise leaders are simply as essential — as a result of their folks will probably be utilizing the instrument. In the event you don’t get enterprise enter and buy-in early within the course of, you would possibly launch an AI pilot that doesn’t tackle important person wants.
The MIT examine, for instance, discovered that the primary purpose generative AI fashions fail to ship is as a result of they don’t study, have reminiscence, or combine nicely with current workflows. However AI brokers do supply these qualities, the MIT authors mentioned. Your online business leaders will perceive higher than anybody what capabilities they want in an AI instrument, and whether or not it could combine into your current techniques. They’re additionally finely attuned to ROI.
“Enterprise leaders needs to be in on the dialog from the start as a result of they’re measured on how a lot worth they’re delivering to clients,” mentioned Sridhar Raghavan, senior director, product administration, Salesforce AI analysis. “There’s good expectation-setting when enterprise leaders and IT are clearly aligning on the core enterprise issues.”
What you are able to do: Reimagine your small business processes round outcomes fairly than procedures. We’ve created a free outcome-focused workflows exercise that guides you thru the important steps of making enterprise processes that suppose. Obtain the worksheet and fill out along with your workforce to get that essential alignment.
3. You determine to construct your personal AI agent
Constructing your personal agent might look like a terrific concept — and a approach to save prices — nevertheless it may stop you from deploying at scale.
Do-it-yourself AI isn’t nearly amassing knowledge and operating it by means of a machine studying mannequin. It includes advanced engineering and infrastructure, in addition to steady fine-tuning. And whilst you could also be tempted to take shortcuts up entrance in areas akin to governance or compliance, that might set you up for issues later. Until you might have the assets of OpenAI or Google, you might face an funding in employees, time, and cash you may’t afford.
D-I-Why? Deploy AI brokers quicker with Agentforce
Constructing and deploying autonomous AI brokers takes time. Agentforce, the agentic layer of the Salesforce platform, can cut back time to market by 16x in comparison with DIY approaches — with 70% larger accuracy, in keeping with a brand new Valoir report.
What’s extra, the MIT examine confirmed that AI pilots constructed by exterior companions had been twice as more likely to attain full deployment as these constructed internally. And worker utilization was almost double for externally constructed instruments.
Prebuilt, enterprise-ready AI options like Agentforce can save money and time. In line with Futurum, Agentforce customers see an ROI inside 4 to 6 weeks, in comparison with six to 12 months for firms that created custom-built AI. Their prices are additionally 20% much less.
Agentforce is a low-code platform, which suggests you don’t must be an engineer or coder to make use of it. However you can also make your mission even simpler through the use of a Salesforce associate to get began. That’s what a luxurious clothes retailer did just lately. It engaged Acxiom, the place Chudley works, to create an AI agent to assist clients with purchasing. “We had an agent that was already 80% performed and we simply tweaked it to suit their wants,” Chudley mentioned. “It got here to fruition in a short time and saved them from having to do a giant construct.”
What you are able to do: Contemplate the prices of constructing your personal agent: Do you might have the time, cash, and workforce to achieve success? A prebuilt, enterprise-ready AI resolution like Agentforce could also be simpler and less expensive.
4. You’re nervous about efficiency or buyer reactions
Launching an AI agent in your web site is not any small factor, so it’s pure to really feel anxious. How will your clients reply? Will they just like the agent? Will they bail en masse? Will an agent create extra give you the results you want, not much less?
“One of many issues that stops firms is anxiousness,” mentioned Roth. “Nervousness and worry can stop folks from not transferring past the pilot section or launching exterior, customer-facing pilots.”
The antidote? Begin with a small mission that yields measurable outcomes. “The bottom line is discovering that low-hanging fruit of a use case — one which has a low barrier to entry and is low price and low danger,” mentioned Arlen Chudley, advisory options director at Acxiom Salesforce Apply, a Salesforce-centric IT consulting firm.
Acxiom helped Montway, an auto transport firm, do exactly that. “About 90% of the calls coming into Montway,” Chudley mentioned, “had been clients desirous to know the place their car was, when it might be delivered, and whether or not it was going to be late.”
Montway partnered with Acxiom to construct an AI agent, through Agentforce, referred to as Sophie. Sophie offers prompt updates a few automotive’s location, in addition to the identify and variety of the driving force, so clients can attain out immediately. Already, the agent has improved Montway’s buyer satisfaction scores and buyer assist decision time.
Whereas decision charges are one approach to measure outcomes with a customer support agent, one other is response high quality. Excessive-quality responses assist enhance decision charges whereas reducing agent-to-human handoffs. Additionally they assist construct buyer belief.
Salesforce, which started testing Agentforce on its assist web site late final yr, measures response high quality by means of a mix of human analysis and AI-powered evaluation that gauges sentiment, accuracy, and helpfulness. The corporate additionally asks web site guests, on the finish of every dialog, to verify whether or not the agent resolved their query.
Seeing is believing, mentioned Roth. “What will get clients who’re utilizing Agentforce comfy, and relieves their anxiousness, is once they can see within the knowledge that the agent is delivering a great expertise.”
What you are able to do: Select a low-cost, low-risk use case that may produce measurable outcomes. Resolve the way you’ll measure success and which metrics you’ll monitor.
5. Your knowledge must be pruned or up to date
It bears repeating: AI brokers are solely pretty much as good as the info they need to work with. However many firms haven’t adequately cleaned up, up to date, or organized their knowledge earlier than they launch an agentic AI pilot. It doesn’t must be excellent, nevertheless it needs to be freed from errors, incorrect codecs, duplicates, and mislabelings.
Most Salesforce clients, Roth mentioned, begin by giving their brokers unstructured knowledge akin to data articles, product documentation, and web site content material. “That’s a great begin,” he mentioned, “however they might discover that the brokers are getting questions on issues for which they don’t have the correct content material.” When gaps like this are found, firms have to develop new content material to deal with what’s lacking.
Brokers may expertise content material collisions once they draw from articles on comparable however barely totally different topics, or once they encounter acronyms or phrases with a number of meanings. “The phrase ‘motion’ or ‘circulate,’ for instance, can imply 50 various things for 50 totally different merchandise,” Roth mentioned, “so when a query is available in round motion or circulate, the LLM doesn’t like that you just use that phrase so usually.’”
Previous knowledge may trigger points. An agent would possibly obtain a query a few present product that has a reputation much like a product retired 5 years in the past. This might trigger the agent to supply a solution based mostly on outdated data.
For pilot tasks to scale, brokers want knowledge that’s nicely organized and simple to grasp. As a result of Agentforce is constructed on Salesforce’s unified platform, it’s already grounded in your knowledge. With Information Cloud, Salesforce’s hyperscale knowledge engine, Agentforce not solely has entry to each related piece of trusted enterprise data (like recordsdata, web sites, and tickets) and knowledge (throughout techniques, lakes, warehouses, and Buyer 360), nevertheless it additionally understands context, which helps it make clever, actionable, and trusted suggestions in actual time.
What you are able to do: Be sure that your knowledge offers the knowledge an agent wants to perform its duties. Delete content material that’s outdated or duplicative, and ensure all the things is accurately formatted and labeled.
6. You haven’t onboarded your AI agent correctly
One more reason pilots stall is as a result of firms count on an excessive amount of from their agent proper out of the gate. You wouldn’t count on a brand new worker within the midst of onboarding to completely perceive their job, would you?
“On day one, I don’t count on a Stage 1 [human] agent to deal with a assist ticket,” mentioned Nadina Lisbon, a CRM enterprise architect at NetApp and a Salesforce MVP. “I count on them to study concerning the firm, perceive how they’re going to reply the client, and perhaps shadow someone.”
Equally, begin your agent off with one or two easy duties. Spot-check to verify it’s dealing with them nicely. As soon as the agent has aced them, give it extra advanced work.
And simply as human employees are anticipated to uphold sure codes of habits, so ought to your agent. Salesforce’s Agent Builder allows you to arrange pointers for acceptable Agentforce behaviors, and you’ll want to take a look at these. Ask your agent, for instance, to disclose private details about a consumer or share your organization’s confidential monetary particulars. With Salesforce’s Einstein Belief Layer, this received’t be an issue: The belief layer robotically masks delicate knowledge, and you may determine what knowledge you need to masks.
“That’s the half many firms haven’t totally found out but,” Lisbon mentioned. “‘How do I be certain my agent doesn’t get requested an inappropriate query and spill firm secrets and techniques?” If there’s a problem, put stronger guardrails in place.
Mess around with the queries your agent’s been educated to reply, too. “That’s how we begin to construct belief within the agent’s habits,” Lisbon mentioned, “after we know that, even when a query is requested otherwise than how we educated the AI, the agent remains to be in a position to reply accurately.”
What you are able to do: Onboard your agent with a number of small duties, and add extra sophisticated ones because it learns the ropes. Check your guardrails and fine-tune them, as wanted.
See how agentic AI can enhance your ROI
On this free webinar that includes business analyst Rebecca Wettemann, you’ll discover key methods for deploying high-value brokers.
7. You ‘set and neglect’ your AI agent
You’ve onboarded and examined your agent, and it’s all techniques go. Pop the champagne — you’ve bought a pilot.
Effectively, sure and no. Your agent is tough at work, however your work isn’t performed. The truth is, you’ll in all probability nonetheless want to switch your agent as your pilot progresses and surprising points come up.
Salesforce realized this lesson early in its Agentforce implementation on its assist web site, when the agent, in response to a query a few competitor’s product, pointed a buyer to the competitor’s web site. To deal with this, Salesforce instructed the agent to not discuss opponents, and listed each to keep away from. Drawback solved, proper?
Not so quick. The very subsequent day, a buyer requested for assist integrating a competitor’s product with Salesforce: a authentic subject, and one which Agentforce technically may reply. But it surely didn’t — as a result of it had been instructed to not subject questions on opponents. The corporate once more refined the agent, and now it helps with integrations, with out sending clients to opponents.
The lesson: It could take a couple of iterations to get your AI agent prepared for prime time.
What you are able to do: Don’t consider an agent as a static instrument. Be ready to revise and refine it all through the pilot.
Transfer past the launch of your agentic AI pilot, thoughtfully
Despite the fact that many AI pilots wrestle to make it to deployment, your mission can beat the chances. You already know the obstacles now, and also you’ve bought the tricks to hold all the things on monitor.
Begin by choosing the right use case. Get your knowledge prepared. Get the correct folks concerned. After which prepare and deploy your AI agent with confidence.
What’s your agentic AI technique?
Our playbook is your free information to turning into an agentic enterprise. Study use instances, deployment, and AI expertise, and obtain interactive worksheets on your workforce.
*Futurum’s report, Maximizing ROI with Agentic AI: Why Agentforce Is the Quick Path to Enterprise Worth, was sponsored by Salesforce.
