Suppose a sales team uses AI to prepare 1,000 documents a month. Preparation falls from 20 minutes per document to eight. The project report records 200 hours saved each month.
Finance opens the accounts and finds the same payroll, the same overtime bill and no additional sales. There is also a new AI subscription to pay.
Both teams may be reporting accurately. Part of the work is faster. Profit changes when the time released changes actual costs or sales outcomes. The next question is who will use that time, for what purpose, and how the result differs from what would have happened without AI.
What evidence of faster work actually tells us
Generative AI at Work, published in The Quarterly Journal of Economics in 2025, studied 5,172 customer-support agents at one company. Access to an AI assistant increased issues successfully resolved per hour by an average of 15%. Results varied with experience, with larger benefits for less experienced or lower-skilled agents. This is evidence of better performance in a particular work setting. The 15% figure is not a profit increase available to companies in general. Read the published research.
A similar organizational question has appeared in Thailand. In August 2026, Microsoft described a gap between employees' enthusiasm for AI and changes in the way their organizations work. Its report and company press event provide a useful way to frame the process problem. As a technology vendor's account, they should not be treated as independent evidence of how much profit Thai companies have already made or lost from AI. Microsoft, 14 August 2026.
Business owners can answer a more immediate question from their own records: once work becomes faster, what changes in the accounts and cash flow?
Count through to an accepted result
Return to those 1,000 documents. All the figures that follow are hypothetical illustrations, not results from an Enersys project.
Saving 12 minutes of preparation per document releases 12,000 minutes, or 200 hours. Suppose checking and correcting AI output adds 80 hours a month. Net time released is then 120 hours.
At an average labor cost of THB300 an hour, the released time has an imputed value of THB36,000. This indicates the scale of the additional capacity. It cannot yet be recorded as a cost reduction. If salaries remain unchanged, employees have more time available for other work.
Review time must include corrections that arrive later: an incorrect product code, a discount condition that does not match the agreement, or a document that needs reissuing after reaching a customer. Timing only the AI's first response overstates the benefit received by the people further along the process.
A company may still choose to invest because work becomes less tiring or customer service improves. Those benefits can matter. They should be recorded for what they are, with follow-up on whether and when they affect customer retention, errors or expenditure.
The same time saving can produce different profit outcomes
Assume the project costs THB12,000 a month for AI tools and services, plus THB120,000 in initial implementation costs. For illustration, allocate the implementation cost over 12 months at THB10,000 a month to show the distinction between profit and cash. The actual accounting treatment must be assessed by finance according to the nature of the expenditure.
Compare two cases, both releasing 120 net hours a month.
| Monthly effect |
More available time; existing work and spending unchanged |
Released time assigned to work with a financial effect |
| Actual overtime avoided |
THB0 |
THB18,000 |
| Additional sales less related variable costs |
THB0 |
THB20,000 |
| AI tools and services |
(THB12,000) |
(THB12,000) |
| Implementation cost allocated in this example |
(THB10,000) |
(THB10,000) |
| Effect on pretax profit from these items |
Down THB22,000 |
Up THB16,000 |
In the second case, assume the company uses 60 released hours to avoid overtime it would otherwise have paid at THB300 an hour. Another 40 hours support 20 additional orders, each requiring two hours and contributing THB1,000 after related variable costs. That adds THB20,000 in contribution margin, the amount available to cover fixed costs and profit. The remaining 20 hours provide spare capacity.
The THB20,000 must already deduct the cost of the goods or services, delivery, commissions and other variable costs caused by those orders. Any additional staffing or other new expenditure must also be deducted. Customers must actually buy: producing more quotations alone creates no revenue.
The table does not prescribe reducing overtime and increasing sales at every company. Its purpose is to prevent double counting. The THB36,000 imputed value of the released hours cannot be added again on top of the overtime savings and additional contribution generated by those same 120 hours.
Cash follows a different timetable. In the second case, recurring cash improvement from these items could be THB26,000 a month before tax and working-capital effects, provided the extra sales are collected and overtime spending really falls. The company still pays THB120,000 upfront for implementation. Selling on credit or buying additional stock in advance can delay the cash benefit.
Faster work may still wait at the next step
Suppose sales staff prepare quotations faster but still wait two days for a manager to approve the price. Customers may see little change in total turnaround time. Or sales may accept more orders while the factory has no spare capacity. Outsourcing the extra production could reduce the margin on each order.
Before choosing where to use AI, follow one item from the initial request through delivery and collection. Identify where work waits, who must review it, and what prevents the company from taking on more business. Saving time at a step with no queue may have less financial value than fixing a small approval step that holds up the whole job.
The first useful change may be correcting product data, defining discount authority or setting an approval schedule. Once information and decision rights are clear, the company can test AI on an appropriate part of the process.
Compare against what would have happened anyway
Higher sales after introducing AI do not establish that AI caused the increase. A campaign, a larger order from an existing customer or a new employee may have arrived at the same time.
A useful trial defines in advance which work will be compared, how complex the documents are, what constitutes acceptable quality and how many work cycles will be observed. Where feasible, compare similar work using AI and the existing method during the same period. Where that is impractical, record a baseline and document other changes happening alongside the trial.
Track elapsed time through approval, rework, actual spending changes and additional sales after related costs. For credit sales, follow the result through collection. The operational manager should own quality and the use of released time. Finance should verify which benefits have reached the accounts and which remain capacity that may be used later.
Expansion criteria should describe outcomes: a lower cost per accepted document within a defined error limit, for example, or enough additional contribution margin to cover the project's costs. Requiring employees to use AI every day does not answer those questions.
A starting point for a discussion with Enersys
To assess a process with Enersys, begin with one workflow and enough real documents to show its variation. Bring the current hours, costs, approval responsibilities and the system holding the final outcome. Agree which information needs connecting, how far AI should act, where a person reviews the work and who verifies the results.
The first assessment should produce enough evidence to continue the trial, change the process or stop. ERP and AI integration may form part of the answer, depending on the work and the organization's data. Selecting a tool alone cannot guarantee a profit increase.
Before approving the next budget, take the claim of "200 hours saved" one step further: how many hours remain after checking, what work will use them, and what record will show finance the result?
For the review and approval decisions behind these measures, continue with managing an AI-enabled team. The Enersys AI page also explains how to frame the workflow and data questions for an initial discussion.
