IBM just unlocked an estimated $4.5 billion in productivity gains. The play wasn’t a customer-facing AI. It was their own HR department.
Read that again. While every Fortune 500 CEO is asking ‘how do we use AI to grow revenue,’ the company that essentially invented enterprise computing did the opposite. They turned the AI inward. They aimed it at the most invisible, least-celebrated workflows in the company – HR questions, promotion paperwork, manager approvals. And it worked at a scale that should reset how every executive thinks about AI ROI.
Here’s the deployment, straight from the data. IBM rolled out agentic AI across its 270,000 employees globally. Two flagship workflows lead the case study. The first is AskHR – an AI agent that answers routine HR questions for any employee, anytime. AskHR now resolves 94% of routine HR inquiries entirely on its own, with no human involvement. The second is a suite of manager-facing AI agents that handle the administrative layer of being a manager – drafting promotion paperwork, processing approvals, surfacing required information. Those tasks are now completed about 75% faster on average.
Add it up across a 270,000-person company, and the productivity gains land at an estimated $4.5 billion.
Why This Actually Worked
There are three principles every business leader should pull from this case.
First – and this is the one that surprised me – the value didn’t come from the AI being smart. It came from the AI being available. Most internal friction at large companies isn’t intellectual; it’s queue-based. Employees waiting for an HR rep. Managers batching approvals to do ‘later.’ Policies being Slacked to three different humans to confirm. When you make a competent AI available 24/7, that queue disappears. The intelligence floor only has to be good enough; the availability is what unlocks the value.
Second, IBM dogfooded internally before going external. They built and proved the agent stack on workflows where they owned the data, controlled the policies, and could measure the gains. That’s the inverse of how most companies approach AI – they want to put it in front of customers immediately. IBM’s order of operations is the smarter one: prove it inside, then ship it out.
Third, they aimed AI at processes, not people. The goal wasn’t to replace HR business partners or middle managers. It was to subtract the repetitive, low-value layer of those jobs so humans could spend time on the high-judgment work that actually requires them. That’s the difference between AI as a tool and AI as a layoff strategy – and in IBM’s case, the redeployed time has been measurably more valuable than the cost it replaced.
I’m Mike Partners, and I started AiExpert.org to bridge the gap between enterprise AI strategy and small business reality. Here’s your action plan.
Apply this week: Write down the five internal questions your team asks each other most often – PTO policies, software access, reimbursement procedures, client protocols, whatever shows up in your DMs every Monday. Put the answers into a single document. Then create a simple AI assistant (a custom GPT, a Claude project, or a Slackbot pointed at that document) that your team can query instead of messaging each other. That is your AskHR – and it takes an afternoon to build.
The SMB Playbook
You don’t have 270,000 employees. You probably don’t have 270. The principle still ports directly. Here’s how to run IBM’s play at your scale.
- Inventory your top 5 most repetitive internal questions. Not customer questions – internal. The ones your team asks each other the same way every single week. PTO policies. Software access. Approval workflows. Reimbursement rules. Whatever it is, you already know what they are because they’re in your DMs every Monday.
- Stand up one tiny internal agent against those questions. You don’t need a custom build. There are now off-the-shelf tools that let you point an AI agent at a folder of policy docs and a Slack channel for under $300/month. The setup is a weekend, not a quarter.
- Measure the time saved, then redeploy. Pick a metric – DM volume on policy questions, time-to-approval, ticket count to ops – and watch it fall. Then take the time you’ve freed up and consciously redeploy it to a high-judgment task. Coaching. Pipeline review. Strategic planning. AI’s value compounds when the recovered hours go to growth work.
The thing that’s easy to miss in IBM’s $4.5B headline is that the play is identical at every scale. Every business has a back-office that’s quietly bleeding hours into invisible queues. AI is purpose-built to plug that hole. The question isn’t ‘should we use AI?’ The question is ‘where is our company already paying for friction we’re not even tracking?’
Frequently Asked Questions
How did IBM’s AskHR AI agent achieve a 94% resolution rate?
AskHR was trained on IBM’s complete HR policy documentation and given access to employee-specific data like PTO balances, benefits information, and approval workflows. Because most HR questions are repetitive and policy-driven, the AI can resolve them instantly without human involvement. The 6% that require human judgment are escalated with full context to an HR business partner.
What is the best way to build an internal AI knowledge base for a small team?
Start by documenting your five most frequently asked internal questions and their answers in a single document. Then use a custom GPT, a Claude project, or a Slack-integrated AI bot to make that knowledge searchable 24/7. Mike Partners recommends starting with the questions that hit your DMs every Monday – those are the ones bleeding the most time across your team.
How much time can AI save on internal HR and operations questions?
IBM’s data shows that AI-powered internal agents can resolve 94% of routine inquiries and speed up administrative tasks by 75%. For a 20-person company where employees spend an average of 15 minutes per day on internal questions and approvals, that translates to roughly 50 hours saved per month – over 600 hours per year redirected to productive work.
Should small businesses deploy AI internally before using it for customers?
Yes. IBM’s “dogfooding” approach – proving AI internally before deploying it externally – is the smarter order of operations. Internal deployments let you control the data, measure the results, and build confidence before putting AI in front of customers where mistakes are more costly. AiExpert.org provides frameworks for running this exact sequence at any company size.
What is the difference between AI replacing employees and AI augmenting employees?
Replacing means eliminating the role entirely. Augmenting means removing the repetitive, low-value tasks from a role so the person can focus on high-judgment work – coaching, strategy, relationship building. IBM’s approach was augmentation: they subtracted administrative busywork from HR and management roles, and the redeployed time proved more valuable than the cost it replaced.
