This week, we got access to the motherlode of AI strategy advice.
14,000 of you signed up for the AI Strategy Summit, where I sat down with 9 leaders to discuss their lessons from deploying AI in the enterprise.
This was not “performative AI” – aka, running around at All Hands bragging about being “AI-first” when you don’t use it yourself.
This was clear, tangible, REAL guidance for deploying AI. It was obvious that every leader I spoke to had been in the trenches – doing the lunch and learns, building the prompt libraries, scoring employee “AI fitness,” and all the other thankless tasks that lead to real ROI.
The big lesson I took away: This is hard, and it’s worth it. Every leader had battle scars AND high adoption rates (some near 100%) and ROI. If you get serious about AI deployment, you can pull ahead of the pack – but it does take work.
I jotted down my favorite insights from the event below. If you want the full playbook, download it here.
1. Start with business outcomes, not cool tech
Lexi Reese from Lanai said too many leaders are falling into the “BAM era” – Blind AI Mandates (I love that). Companies are saying they’ll be “AI-first” without understanding where they are today or what problem they’re solving. Don Bennion from Adobe said something similar: Don’t fall in love with the tech; fall in love with the problem.
Their advice: Start with concrete business outcomes. “We want to improve gross margin by 5%” or “increase customer retention by 2x.” Then ask what employee or customer problems AI can solve to drive those outcomes. Fund exploration around specific use cases, not general AI adoption.
2. Map your workflows before you buy AI (and train your managers to do it)
Shruthi Shetty from SAP said before you sign on the dotted line for an AI tool, take 30-60 days to map your workflows. Start with one high-impact function (e.g., marketing or product) and ask the people doing the work to map the workflow. For example, your procure-to-pay workflow should be mapped by your procurement specialist – they’re the ones who understand the nitty-gritty tasks.
According to Ravin Jesuthasan from Mercer, workflow audits should be an expectation of your managers. Instead of thinking, “I should open a work rec,” they should think, “I should redesign this work with AI.” He says this means deconstructing roles into tasks, then determining which tasks should be outsourced to AI, augmented with AI, or accelerated with AI.
3. Data foundations are 75% of the battle
Amadeus Tunis from Acxiom said “AI isn’t missing imagination – it’s missing integration.” Focus on data foundation and activation before jumping to modeling. Two questions matter to AI enablement: Do you have the data to support it? And (more importantly) do you have the data infrastructure to enable it? Most leaders say, “Oh, we have a ton of data,” but that data is trapped in silos and hard to activate.
Tony Gentilcore from Glean agreed: “It’s about context, not models.” The models have become commoditized and the leaderboard changes every week. What matters now is organizing and governing your data so that you can expose it to a variety of AI tools.
4. Focus on frontline managers, not just senior leadership
Don Bennion from Adobe had one of my favorite insights of the day – win the hearts and minds of your frontline managers. They’re the ones who understand how work gets done, and they’ll give candid, critical feedback about what’s not working. Senior leaders can advocate, but frontline managers drive real adoption.
5. … But also force your execs to get hands-on
Olya Taran from Manulife said it’s critical to get execs using AI themselves. Rather than trying to find extra time on their calendars (impossible), she joined existing leadership forums and sessions and forced executives to tackle real-world leadership challenges with AI.
Once they saw the power of AI, they became advocates and the next level down wanted to participate too. Olya ended up scaling “prompathons” to the whole company – they have a waiting list.
6. Validation outweighs explanation
Zachi Attia deploys AI at Mayo Clinic, so he’s working in a highly regulated (and often reluctant) industry. When Mayo moved from explainable machine learning models to deep neural networks, physicians constantly asked, “Yes, but what is it looking at? How can you use something when you don’t know what it’s doing?”
His solution: Create a process of validation that builds trust, rather than endlessly explaining how AI works. His team did retrospective studies, prospective studies, diversity testing, multi-site validation and rigorous quality checks – to prove that their heart failure detection algorithm worked reliably and safely across diverse use cases.
7. Go organization-wide with deployment
This is a controversial one – some AI leaders will tell you to start small, and others will tell you to go org-wide.
Brice Challamel from Moderna has driven 80%+ adoption of AI, and his point of view is to launch to everyone. He says: “Utilities are effective at scale. When everyone has them, knows them, and sees their neighbor using them, they get best practices from the next person and get more and more proficient.”
He compares AI pilots to piloting the internet with five people – it would be hard to see the value at so small a scale.
The bottom line
AI deployment isn't just a technology challenge – it's a leadership challenge.
The companies pulling ahead aren't necessarily the ones with the best AI tools. They're the ones with leadership willing to do the hard work of understanding their workflows, changing their own behavior first, and making fast decisions based on real evidence.
AI is “truth serum for organizations.” It will expose poorly run companies while allowing well-run companies to pull away from weaker competitors.
Ready to start? Pick one high-value workflow, map it in 30 days with the people actually doing the work, deploy an AI solution, and measure results fast. The companies that master this cycle will own the next decade.
And join our next event like this one – The AI:ROI Conference, September 25.
