#1 Professional Services Automation, Project Delivery, and Client Onboarding Software. Rocketlane is a purpose-built client-centric PSA tool for implementation teams, consulting firms, and agencies.
In the last 2 weeks, people have asked me a few times about MIT report on the 95% failure rate of AI deployments in enterprise. After all, I've been extremely bullish about AI in every forum I've spoken at, every post I've written. My answer: "I didn't get the time to read it yet, but I am sure there is some nuance to what the 95% or 5% represents, etc. Did you actually read it?" And finally I got time this long weekend to read up on this, and also some articles digging into where the number came from. Turns out this number wasn't what most people interpreted it to be. The success rate of AI deployments were far higher - more like 1/3rd for "build" and 2/3rds of the "buy" deployments of AI made it to production. But when you take some task specific custom AI deployments that were "top down" in nature, which was a smaller subset, the ones that had sustaining ROI and impact were 5%. And why the failure to deliver ROI? Picking the wrong problems, not integrating well, etc. The answer in my head: - Don't try to solve everything internally. Use hackathons, internal initiative to validate, understand, do simple POCs. - Build a business case to see if the problems you are picking are truly worth solving before you go all-in on an initiative - Work with vendors who have solved the problem before or are motivated to "work through the last 20%" with you. Getting POCs to 80% there is easy with internal teams and the 80 to 100 journey is where most often AI is failing today. Fin.
Really interesting perspective! Your point on the 'last 20%' reminded me of TGC Prasad’s The Last Ten Percent-where even well-designed systems fall short if the final stretch isn’t managed well. Do you think AI deployments are hitting the same roadblock-great POCs but struggling with that critical last-mile integration?
For Enterprises AI Adoption to be successful, they need to overcome various challenges - 1. Data Availability and Sanitation - Biggest challenge - internal data is all over place, ERPs, excels, etc. fragmented and unstructured data often is root of all evil. 2. Lack of Business Context - undermining AI accuracy as the business context is not derived accurately 3. Security & governance risks fear of lack of control over AI agents and data both 4. Accuracy & reliability of AI driven decisions - Decision-making is opaque, can generate incorrect actions, leading to errors & business risks Enterprises do not like to put their business on any risk radars. 5. Orchestration between Humans and Agents The final frontier of AI will be how seamless AI & humans are able to function together to deliver business value Difficulty in coordinating tasks/handoffs between humans <> AI agents leads to execution gaps AI itself can be a boon if the foundational structure is prepared well to adopt it and derive results! Else, it starts with a lot of hope and prayers and ends poorly Srikrishnan Ganesan
True Srikrishnan Ganesan 95% failure rate works well for a being clickbaity 🙂 my take - https://www.linkedin.com/posts/lewis-william_are-we-using-it-the-right-way-the-jury-activity-7364542532623515648-fhbB
So much is down to people and change management. With Gen AI, simply handing out licenses doesn't change the way people work, hence why most are simply dabbling and not thinking about how to go from a faster horse...to a car. Big mindset shift required.
Completely agree- AI implementations fail due to a depth of understanding what it takes to drives outcomes on a consistent basis. Over the last five months, I have myself worked with 3 startup’s to solve this exact problem. The patterns behind earlier failures for these companies were eerily similar - 1. Half baked pipelines with basic prompt / context engineering 2. Poor iteration cycles 3. Surface level implementations by primarily Data Science folks. The reality of what it takes to build a quality product is only understood after months of failure and a realisation that an AI implementation at scale requires expertise ( that might not always be present in house)
Kiran Brahma
2w
I think every tech goes through the same adoption curve. The only issue is that AI is talked with crazy stories of changing everything that such stories seem shocking. Too bad we never have any such studies done for older tech adoptions
I believe the common misconception is that only 5% of the deployments make it to production. Rather, this number focuses on the value the business expected to derive from the deployment and falling short on that. And as you called out Srikrishnan Ganesan - "Purchasing AI tools from specialized vendors and building partnerships succeed about 67% of the time, while internal builds succeed only one-third as often". It often comes down to making strategic choices in build v buy, and sometimes working with a motivated expert who will cover the last 20% of ground required to make a solution really practical for businesses and teams.
Professional Services | Delivery Leader | Digital Transformation | AI Advocate
3wThe nuance behind the so-called 95% failure rate is often missed. I agree that the real challenge isn’t the POC but the last 20% of integration, adoption, and ROI. Curious — do you think enterprises should double down on internal capability-building for that 80→100 stretch, or lean more on vendors who’ve already solved it at scale?