What AI can do for one person
From an individual perspective, AI can help you gain productivity. Recurring work that used to need IT help is a clear case. Automating those tasks often meant asking, finding, or employing IT professionals, and that help is usually very expensive. With AI tools, one person can sometimes clear that bottleneck and finish the same recurring work without waiting on a specialist queue.
That pattern shows up in controlled evidence on personal throughput. In a 2023 Science experiment, Noy and Zhang assigned writing tasks to 453 college-educated professionals and randomly gave half access to ChatGPT. Average time taken fell by about 40%, and output quality rose by about 18%. That is individual productivity: faster, better completion of the person’s assigned work. See: Experimental evidence on the productivity effects of generative artificial intelligence.
That gain is real for the person doing the work. It remains personal productivity. It does not automatically become a company outcome.
Completing more tasks is not a company outcome
Completing more tasks does not necessarily translate into one of the four outcomes that matter from a company perspective: revenue gain, cost saving, competitive advantage, and productivity.
Those four labels are a practical company checklist in this essay, not a claim that every consulting firm uses the same taxonomy. Adjacent measurement work makes the same split in different words. APQC’s guidance on AI value says usage and time saved are early signals, and realized value shows up when AI produces sustained improvements in process performance or business outcomes. Its business-value layer includes cost reduction, cost avoidance, revenue, working-capital improvement, customer outcomes, risk reduction, and ROI. See: How can AI value, ROI, and productivity impact be measured?.
A company cares whether the work moved one of those four. Just because an employee finishes tasks quicker does not mean it provides value to the company. Output can rise while the business result stays flat. APQC states the same trap plainly: time saved is potential capacity; it becomes financial value only when the organisation connects it to verified outcomes such as greater throughput, cost reduction, cost avoidance, revenue, or improved service.
Public survey evidence shows the same gap. McKinsey’s October 2026 commentary on AI adoption notes that nearly nine in ten organisations now use AI in at least one business function, while only about six percent of companies attribute at least five percent of earnings before interest and taxes (EBIT) to AI and report value from its use. About thirty-seven percent report some positive effect on earnings. Adoption is common. Clear company payoff is rarer. The same piece notes that studies showing more issues resolved per hour or more coding tasks completed are meaningful for individuals and teams, and still do not prove that adding AI to any process will improve a company’s earnings. See: Why AI’s easiest wins are misleading CEOs.
Why companies treat AI as an investment case
From a company perspective, investing in and implementing AI transformation requires a large amount of investment. The return on investment (ROI) needs to be justified.
In simple terms, ROI is a financial ratio of an investment’s gain or loss relative to its cost: the benefits should outweigh the costs. APQC writes the same formula as (gain from AI investment − cost of AI investment) / cost of AI investment, and says ROI should be the end of a value chain, not the only measure. PwC uses the textbook framing for AI and calls it “hard ROI.” It also notes a practical trap that matches the individual-versus-company split: if you cut the time for a task by about twenty percent, the person may fill that time with other work, so the company does not automatically bank a labour saving. See: Defining and measuring return on investment for AI.
Faster personal throughput feeds that investment case. It is not the case itself.
Token usage is part of the cost side
Token usage is another cost factor. Model usage is not free. If more people complete more drafts, summaries, or automations, token spend rises.
That is not a metaphor. Major model providers bill API usage by the token. OpenAI’s API pricing lists rates per one million tokens, with separate prices for input and output, so more prompts and longer answers raise the bill directly. See: OpenAI API pricing.
McKinsey’s five-layer AI measurement framework puts the same idea into the company ledger. Financial impact covers enterprise outcomes such as revenue uplift and cost-to-serve reduction, and total cost of ownership includes cloud and token spend. At the technical layer it also tracks token cost per interaction. Without a link from that spend to one of the four company outcomes, the organisation can pay more to look busier. See: From promise to impact.
Keep the two ledgers separate
Personal AI productivity answers: did I get through more of my work with less waiting and less specialist cost?
Company AI value answers: after investment and token cost, did we move revenue, cost, competitive position, or true organisational productivity?
Merged ledgers make activity look like progress. Separate ledgers keep the next AI spend decision honest.
