$4.5 billion in productivity impact. 3.9 million hours of human work freed. And the most interesting twist in the entire story is this: IBM didn’t build any of it for clients. They built it for themselves. Then they sold it.
The Full Case Study
For decades, IBM has been the company that helps other Fortune 500s deploy enterprise technology. In the last two years, they made a sharp strategic shift: before they sold AI to anyone else, they were going to deploy it inside their own walls – at scale. The result is one of the most under-discussed AI ROI stories in the industry.
IBM didn’t chase a single big use case. They went department by department. They built AskHR – an internal agent that handles policy questions, time-off requests, job transfers, and routine HR transactions for the company’s hundreds of thousands of employees. They built AskIT – an internal IT support agent that now resolves 86% of IT queries autonomously, cut call and chat volume by 74% since launching in 2023, and produced an initial $18 million cost reduction with ongoing annual savings. They then extended the same pattern into finance, sales, services, and consulting.
The aggregate impact in 2024 alone: an estimated $4.5 billion in productivity impact and 3.9 million human hours reclaimed from manual tasks. And here’s the twist – IBM is now packaging those same internal agents and selling them externally as part of their consulting and software business. Internal efficiency became external revenue.
Why This Actually Worked
First, IBM dogfooded. The most reliable way to build a great AI product is to be its first painful customer. Building for clients first invites theory; building for yourself first reveals truth. IBM learned what worked and didn’t work on themselves, then carried only the winners into the market.
Second, they used the volume principle. Internal ops are full of high-frequency, low-variation questions – ‘What’s the PTO policy?’ ‘How do I file this expense?’ ‘How do I reset my VPN?’ Each one is small. The aggregate is massive. AI thrives where volume and pattern coexist, and internal ops is the most volume-rich, pattern-rich part of any company.
Third, they kept the unit of work small. Each agent owns one workflow. AskIT doesn’t try to be AskEverything. That focus is why each agent actually shipped – and why the ROI was easy to measure.
My name is Mike Partners, and as an entrepreneur I’m passionate about helping small businesses compete with the biggest companies in the world – which is why I built AiExpert.org. Here’s how to take this lesson and make it work for your company.
How to Apply This to Your Business
Identify the single most-repeated internal question or request in your business – maybe it’s ‘What’s our return policy?’ or ‘How do I file this expense?’ or ‘Where’s the file for X?’ – whatever it is, that’s your first agent. Then build a thin internal agent on top of your existing docs using tools like ChatGPT custom GPTs, Claude Projects, or Glean, which can be set up with your knowledge base in under an hour and trained on your 20 most-asked questions. Roll it out internally first and measure two things: time saved by you and your senior team, and accuracy of the answers. Iterate weekly. Once it’s working, consider whether the same agent can be productized for your customers, vendors, or members. The single biggest miss in SMB AI adoption right now is people chasing flashy customer-facing AI before fixing the boring internal AI. IBM is proof that ‘boring internal’ is where the multi-billion-dollar lines hide.
Frequently Asked Questions
What are IBM’s AskHR and AskIT systems?
AskHR is an internal AI agent handling policy questions, time-off requests, job transfers, and routine HR transactions for IBM’s hundreds of thousands of employees. AskIT is an internal IT support agent that resolves 86% of IT queries autonomously and cut call and chat volume by 74%, producing $18 million in initial cost savings.
How did IBM turn internal AI tools into a revenue stream?
IBM built AI agents for their own internal operations first, learned what worked through real-world deployment, then packaged those proven agents as products for their consulting and software business. Internal efficiency became external revenue – a strategy any business can replicate at smaller scale.
Can small businesses build internal AI agents like IBM did?
Yes. Tools like ChatGPT custom GPTs, Claude Projects, or Glean can be set up with your knowledge base in under an hour. Mike Partners built AiExpert.org to help small businesses deploy these exact types of internal AI agents, starting with the 20 most-asked questions in your organization.
Why should businesses start with internal AI before customer-facing AI?
Internal AI is lower risk, easier to measure, and builds organizational confidence. IBM proved that internal operations – HR, IT support, finance – contain the highest-volume, most pattern-rich workflows where AI delivers the clearest ROI. Customer-facing AI works better when you’ve already mastered the fundamentals internally.
How do I measure the ROI of an internal AI agent?
Track two metrics: time saved by your team (especially senior staff) on repetitive questions, and accuracy of the AI’s answers compared to human responses. Measure weekly and iterate. Even saving 30 minutes per day across a small team translates to hundreds of hours per year in reclaimed productivity.
