The quickest way to overstate your AI ROI is to watch one person do something impressive with a chatbot and assume the whole company just got that much better. It rarely works that way. New research from Salesforce this week put a number on the gap: most of the value people get from AI never leaves the person who found it. If you are trying to work out what AI is actually returning to your business, that gap is the thing to measure.
Salesforce surveyed 250 U.S. executives with decision-making authority over AI spending, in mid-September, and asked where the value goes. The findings are about where the value lands after one person creates it:
Salesforce calls this "single-player AI": people getting faster on their own, in private chats, while the company around them learns nothing from it. Rob Seaman, who runs Slack, put the fix as a shift in era: "The single-player era of AI made people faster. The multiplayer era is what makes a whole company smarter, together."
Whatever you think of the framing, the underlying point holds. A prompt that cuts a review in half is worth very little if one person knows it and nobody else ever sees it.
ROI is a ratio: the return over what you spent. The spend is easy to see. You are paying for seats, for tokens, and for the hours people lose wrangling the tools. Executives in the same survey estimate that people lose about 3.6 hours a week moving between their AI tools and everything else, roughly 22 business days a year. That cost sits on the ledger whether or not anyone measures it.
The return is where teams fool themselves. A per-person speedup is real for that person. It is not a business return until the work it produces reaches other people and repeats without them. One analyst who got faster is a nice anecdote. A process ten people now run the faster way is a number you can take to the bank.
So when the value stays trapped in one person's chat window, you keep paying the cost side of the ratio and collect almost none of the return. The bottleneck is distribution. What one person learns never travels to anyone else, and that is why a lot of AI budgets look busy and return nothing you can point to.
Pilots make this worse, because a pilot is single-player by design. You hand the tool to a few motivated people, they produce a big percentage saved on one task, and that number goes in the deck. It is a true number. It is also the best case, from the people most likely to make it work, on the one task they chose.
The jump from that number to a company-wide return is the hard part, and it is the part pilots skip. Before you trust a pilot figure, ask what happens when the tenth person, who did not design the pilot, has to use the same thing on a Tuesday. If the answer is that they will not, or that they will do it their own way, the pilot measured enthusiasm, not return.
Getting AI ROI you can actually bank comes down to a few decisions.
Our fractional COO work is built around exactly this. We follow the work from first inquiry to final invoice, find where it stalls, rebuild and document the process, then build the AI agents and automations that keep it running. The documenting is the part that matters here: the win does not live in one person's head or one person's chat, so the team can run it without the person who found it.
We run BYLD itself on Byld Ops, so its AI assistant works across the CRM, proposals, invoicing and projects, rather than as a separate chat window off to the side. That is the practical version of getting out of single-player mode: the tool is where the work is, and anything it drafts is visible to the people who need it.
When the return is measurable, you can check it honestly. We built the AWS infrastructure for an AI learning platform, Budly, and brought its daily cloud cost down about 30% while running three environments instead of one. That is a number you can set next to what it cost to get there, which is the whole point.
Measure the return where it shows up for the business, not for one user: time on a full process, cost per job, or work shipped per period, against everything you spend on the tools and the hours spent using them. A per-person speedup counts only once it becomes a process other people run.
Often because the value is trapped in single-player use. The Salesforce research found most AI insights go unshared and work stalls inside one person's tool. You pay for the tools either way, so if what people learn never spreads, the cost lands but the return does not.
Salesforce's term for people using AI individually in private chats, where prompts, insights and results stay with the person who made them. The opposite is AI working in the shared systems a team already uses, where one person's work is visible and reusable.
If you want to figure out where your own AI spend is returning something and where it is just busy, book a call below and I will go through it with you.