Work has become easier with AI—but what has changed for the company?
Team members say AI has made their work easier. Drafting emails and organizing reports have both become smoother than before. But the answer to “So what has changed for our company?” may not be clear.
It may seem as though everyone is saving time, but we may still not know whether customers are getting responses faster or whether the same team is completing more work. If you are responsible for adopting AI, you may be wondering what to check after saying, ‘We’re using it well.’
In Deloitte’s 2026 survey, 66% of organizations reported productivity and efficiency gains, while 20% reported revenue growth. To see change at the organizational level beyond individual convenience, it helps to consider these two effects separately.
The benefits achieved now and the benefits hoped for later
The first two figures are benefits respondents said they had already achieved; the last is a benefit they expect in the future. They represent the share of surveyed organizations and do not mean productivity rose by 66% or revenue increased by 20%. Responses can overlap, so they should not be added together.
This gap alone does not justify concluding that “the AI investment failed.” Some teams may have started with cost reduction as their goal. But if revenue was the goal, the fact that documents are written faster alone does not show that goal has been achieved.
Organizations are using AI at different depths
Deloitte also separates organizations by how deeply AI has entered their existing businesses and work. The three levels below do not mean every organization must move through them in sequence.
| Level of use | Share of respondents | What changes |
|---|---|---|
| Keeping existing work largely unchanged | 37% | Using AI on top of existing processes |
| Redesigning core work processes | 30% | Changing the flow of work around AI |
| Using AI to transform the business deeply | 34% | New products and services, changes to core processes or the business model |
This summarizes the categories in the official report. Rounded response shares total 101%.
Applied to our own work, this classification shifts the question from “How much are we using the tool?” to “Which part of the work has changed?” The same AI can change work to different degrees.
How far can writing quotes faster take you?
To understand this difference, consider quotation work in a sales team. The following is an editorial application example, not an organizational case from the survey.
The first change is using AI to draft quotation emails faster. If the time a representative spends writing falls, that is already a productivity benefit. But if they still need to find a price list, check inventory, and wait for approval before sending it to the customer, the total elapsed time from inquiry to sending may not decrease much.
The second change is examining the flow from inquiry to sending. For example, make standard prices and conditions immediately available, route only exceptions to an approver, and connect the process through to sending the approved quote. The results to check then extend beyond draft-generation speed to the time from inquiry to quotation delivery and the customer’s next action.
Even then, revenue growth requires separate confirmation. Quotes may be faster, but contracts may not increase if they do not offer what customers want. Whether a quick response leads to a contract must be considered alongside price, product, and customer conditions.
Open your quotation records today
Before introducing a new tool, you can look for what changed in a task where AI is already in use. For quotation work, open your inbox or customer-management records and create a comparison sheet in Excel or Google Sheets.
- Choose the same type of quotation. Start with a small set, such as five recent cases from before and after using AI, but select the same product and customer type. Five cases are a starting point for reviewing records, not a sample size that proves performance.
- Start by copying timestamps from the records. Record the inquiry receipt time and the sending time of the approved quote, then first record ‘sending time − receipt time’ as total elapsed time. To report customer-perceived wait time separately, decide whether to count only business hours, and record interruption periods—such as time waiting for an additional customer response—in a separate column so they can be excluded. Without these records, report only total elapsed time and do not call it customer wait time. Record the time the representative spent drafting and revising in another column, but leave it blank if historical records are unavailable.
- Check why long cases took longer. First see whether total elapsed time decreased, then note whether time in cases that still took long was spent on price lookup, internal approval, or waiting for an additional customer response. Also record differences in conditions, such as weekends, non-business hours, and inquiry complexity.
The first output of this review is not a number saying ‘AI drove X% growth,’ but one part of the workflow to change next. If drafting is faster but sending is unchanged, for example, start by looking at approval wait time. If earlier records do not exist, begin recording timestamps and interruption periods this week, and leave performance figures blank until you can compare them.
Task to review: ______
Change identified and evidence: ______
Area still taking too long: ______
One thing to change next: ______
Owner / date to review again: ______
To assess revenue as well, separately check the share of sent quotes that lead to contracts. Give the quotes being compared the same observation period, and examine changes in price, product, and customer conditions. Do not report a revenue effect solely because draft-writing time or total elapsed time decreased.
By distinguishing time saved by the representative, total elapsed time from inquiry to sending, customer-perceived wait time calculated using defined conditions, and outcomes that led to contracts, you can discuss concretely with the team what to improve after ‘we’re using AI well.’
Survey scope and checking the pilot figures
The survey was conducted in August–September 2025 among 3,235 business and IT leaders across six industries in 24 countries. These are results reported by respondents, not an experiment verifying whether AI caused revenue changes at individual organizations.
Public materials do not show a cross-analysis linking levels of work change with revenue responses. It cannot be assumed that organizations that redesigned work are the same organizations that increased revenue.
We corrected the original phrasing, “75% of pilots fail.” The official result is that 25% of respondents said that 40% or more of their organization’s AI pilots had moved into operation. The remainder cannot be calculated as the failure rate of all pilots. The 54% figure is the share that expected to reach the same threshold within the next three to six months at the time of the survey, and it should be distinguished from results already achieved.
