The AI That Works Is Beating The AI That Wows

Billy Biggs leads KnowledgeLake’s strategy and growth, bringing over 25 years of SaaS leadership.

gettyThe Department of Veterans Affairs cut average disability claims processing time from 141 days to 81, a 42% reduction, and pushed its backlog below 100,000 for the first time since 2020. There was no keynote for it. It was covered as a staffing story, not a technology one, and that’s the right read. The point was never the AI. It was that case files requiring real human interpretation now move faster through an agency that has lost nearly 2,700 claims examiners since January 2025. But the same coverage flagged a gap: 94% accuracy at the issue level, 83% at the claim level. As the son of a veteran, I know what sits behind that second number: a family waiting on a decision that shapes the next chapter of their lives. Speed is the easy part. The real test is whether you can trace, explain and stand behind every decision at the end of a mission-critical workflow.

Back-office functions—finance, HR, payroll, compliance, data entry—are not a rounding error. The Hackett Group puts median SG&A costs at 14.3% of revenue across the largest US public companies, the highest in five years and only 17.9% of them cut those costs last year. That is an enormous, mostly manual cost base that has sat in plain sight for decades, waiting for someone to act.

Now that AI has been applied, the results are not subtle. Deloitte found that organizations that moved past piloting intelligent automation cut costs by an average of 32%. The VA numbers are not an outlier. They are the pattern repeating itself, in function after function, at companies that never bothered to put out a press release about it.

Compare that to the flashier side of AI. Deloitte’s State of AI in the Enterprise research found employee access to sanctioned AI tools jumped 50% in a year to about 60% of workers, but fewer than 60% of those with access use it daily. Roughly one worker in three is using what their company bought. This AI, embedded in the process rather than handed to an employee as another tool to remember, doesn’t have an adoption problem. It runs whether anyone logs in or not.

AI’s real pitch is never the technology itself. It is how much faster it gets a business to an outcome it was already chasing, measured the way every value-selling framework has measured it for decades: cost savings, time savings, competitive advantage and risk mitigation. Working AI is not a side note to that argument. It is the clearest proof.

Cost savings show up fast, often inside a single budget cycle. IBM pointed AI at its own HR service desk, the least glamorous function in the building. Four years later, that team runs 40% cheaper and 94% of routine employee questions get resolved without a person touching them. That is the proof a CFO now demands before funding anything with “AI” in the name.

Time savings compound. A contract stuck in review delays a deal and keeps the compliance clock ticking. Ardent Partners finds best-in-class accounts payable teams process invoices 79% faster than everyone else, at 79% lower cost, with exception rates 47% lower. Same invoices, same suppliers, same regulatory requirements. The only difference is whether a person retypes the data. That is a bottleneck removed at scale, not a productivity story.

Competitive advantage in flashy AI rarely lasts. In document work, it compounds because the system learns your contracts, your exceptions, your edge cases. Hackett expects the gap between AI leaders and typical companies to widen by as much as 75% as leaders embed AI into core processes. A competitor can buy the same agents and have them running by next quarter. A competitor trying to match five years of your accumulated document intelligence starts from zero.

Risk mitigation used to mean uptime. Now it means defensibility: proving to a regulator how a claim was assessed, with the same standard applied every time. HSBC rebuilt its anti-money-laundering monitoring around machine learning and reported detecting two to four times more confirmed suspicious activity while cutting alert volumes by more than 60%. Fewer false leads, less wasted investigator time and an audit trail that explains itself. A manual process can rarely guarantee that.​

Finding your version of this is not complicated. Rank your processes by how often a person opens a document, reads it and makes a judgment call, then take the one at the top. If nobody can tell you what that process costs today, start there, because a number nobody owns is a number nobody is accountable for.​

Here’s a strange thing to argue in 2026: the least interesting AI deployment a company makes this year might be its most important. I’ve sat in enough board conversations to know the pattern. Ask about AI strategy, and you get a vendor list. Ask which process got faster and what it saved, and the room goes quiet. The P&L doesn’t care what makes a good keynote. It cares about three things: what got faster, what got cheaper and what stopped being a risk. On all three, working AI is winning and most of the market hasn’t noticed.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?