Hello — Okada from Zenken’s AI division here.

Following our press release on April 24, 2025 — “Zenken cuts external outsourcing costs by ¥50M/year through company-wide ChatGPT Enterprise rollout” — this is the second case study, focused on the accounting team. The numbers should be useful as benchmarks for your own goal-setting.

Concrete back-office case studies for ChatGPT Enterprise are still scarce, so if this is useful, please share it.

Case 1: Faster, more accurate accounting decisions

Before: Many situations required judgment calls, and looking things up in references or running them past managers consumed a lot of time. Accounting standards and tax law are written in dense, idiosyncratic language that takes time to parse.

After: The team set up a “Ask the CPA” project in ChatGPT, and now consults the model first when judgment is needed. Throughput on new cases more than doubled, and the time spent escalating questions to managers dropped sharply. ChatGPT translates accounting standards into plain language and serves as a sparring partner, so by the time a question reaches a manager, it has been thought through. That lightens the manager’s load too. Accounting documents can be summarized and rephrased in plain text, making day-to-day work materially more efficient. Final decisions still get verified, but the model’s answers rarely contain serious errors. The benefit isn’t just time savings for managers — the accountants themselves are picking up accounting knowledge along the way.

Case 2: A new way to look things up

Before: Looking up domain terms or operational knowledge meant Google searches that often didn’t return an exact-match result, costing time.

After: Information-lookup behavior shifted significantly from Google to ChatGPT. Users can ask for usage examples and disambiguation against similar terms in one go, getting answers faster and in a form more likely to stick. Searching and analyzing legal and industry rules became much easier, sharply reducing time spent on regulatory research. Faster access to optimal information directly raised work efficiency. For accounting and legal research, ChatGPT’s natural-language understanding helps grasp the gist quickly, materially improving research productivity. It also helps users frame how to look something up, making information gathering more efficient overall.

Case 3: Turning manuals and policies into queryable knowledge

Before: Reading manuals when adopting a new system or tool was a chore. Looking up internal policies or extension numbers also took time.

After: The team converted manual content into queryable knowledge by building dedicated GPTs, dramatically improving access to internal information. A GPT was created for the Azabudai Hills office covering its internal rules and extension numbers, giving instant access to information that previously took time to locate. Asking questions in chat form is more efficient than navigating manuals, lowering the learning cost for new IT tools and improving day-to-day usability. This sped up work overall, lowered the barrier to adopting new systems, and accelerated technology adoption inside the company.

Case 4: Creative thinking and problem-solving support

Before: Brainstorming and problem-solving from a blank page took time, and ideas were limited. Finding fresh perspectives on operational challenges was hard.

After: Brainstorming and problem-solving became materially better. When ideas were needed, users described the situation to ChatGPT and asked it to widen the field of options. ChatGPT became a brainstorming partner for everyday operational challenges, surfacing new angles and solutions. Starting from one of ChatGPT’s suggested directions is much easier than starting from zero, leading to more creative and varied ideas. Even under tight time pressure, ideas come fast, sharply reducing time-to-solution. Brainstorming sessions that used to drag on now produce results in less time, leaving more room for creative work and a higher-quality output overall.

Case 5: Faster Excel work

Before: Excel formulas were researched via web searches and blog posts. When no exact match existed, users were stuck with no clear next step, and time evaporated.

After: Excel work became materially more efficient. Asking specific questions like “How do I do this calculation in cell A2?” or “How can I aggregate the numbers in column B this way?” now produces ready-to-paste formulas, sharply cutting both research and authoring time. ChatGPT is widely used for formula proposals and calculation requests. The repetitive cycle of running multiple web searches to assemble a complex formula is gone. With VBA, users can hand the model a rough version and get back suggestions for cleanup and speedups, producing more efficient scripts. Data processing and aggregation are faster, working hours are noticeably down, and Excel skills are improving along the way. Tasks that used to require specialist skills — complex formulas and VBA — are now within reach for any team member, lifting the data-processing capability of the whole department.

Case 6: Better written communication

Before: Choosing the right words and structuring sentences was hard, and replying to emails or chat could take a while. In situations that required care, finding the right phrasing was a frequent stumbling block.

After: ChatGPT helps write polished, formal email quickly through edits. By providing the inbound chat plus what you want to say back and the key points, you get an accurate response. Things you struggle to put into words are now easier to organize.

Specific example: By giving ChatGPT the inbound chat, the desired direction of the reply, and the key points, accurate response text comes back. Reply-writing time dropped, and users can now articulate things they previously struggled to express — improving overall written communication skill.

Case 7: Parallelizing work

Before: While doing research, you couldn’t also be working on something else. Tasks had to be processed serially.

After: Looking things up and doing other work can now happen in parallel. Handing research or aggregation to ChatGPT frees up time to make progress on other work. Hopping between multiple systems and tools is also simpler.

Quantitative result: Using the 5 minutes ChatGPT spends organizing data to make progress on something else means more efficient work overall, plus less overtime.

Specific example: Previously, doing research blocked you from doing anything else. Now you can hand research or aggregation to ChatGPT and use that time on other work. For example, while ChatGPT spends 5 minutes organizing data, you make progress on a different task — adding up to a meaningful efficiency gain.

ChatGPT Enterprise impact analysis report

1. Quantitative impact on workload

Reductions across major tasks

Task	Before (hrs/mo)	After (hrs/mo)	Saved	Reduction
Accounting decisions	20	3	17	85%
Regulatory research	40	10	30	75%
Sentence shaping	1	0	1	100%
Training notes summarization	1	0	1	100%
Workflow design	3	0.5	2.5	83%
Drafting chat / message body	1.5	0.25	1.25	83%
Specialized-term lookup	1	0.25	0.75	75%
Excel-based aggregation	1.5	0.5	1	67%
Department-cost allocation	1	0.5	0.5	50%
Reading long / complex chats	1	0.5	0.5	50%
Looking up accounting / formula questions	3	1.6	1.4	47%
Accounting department, all duties combined	180	160	20	11%

Summary

• Average reduction per task: 69%
• Aggregate department reduction: 30%
• Where the savings came from:
  • Substantial time savings for tasks that require specialized knowledge (accounting decisions, regulatory research)
  • Near-100% reduction on writing and document tasks
  • Over 50% reduction on most search and information-gathering tasks
• Highest reduction: sentence shaping, training-notes summarization (100% saved)
• Lowest reduction: accounting department overall (11%)
  • Note: while the department total is 11%, individual tasks frequently exceed 50% reduction.

2. Productivity changes

How users perceive their productivity gain

• “1.5x (about 33% time saved)”: 50%
• “No noticeable change”: 37.5%
• “3x”: 12.5%

Observations

About 62.5% of respondents reported productivity gains, with the strongest effects among those who do a lot of specialized research and document drafting. About 40% don’t feel a difference yet — the impact varies by department and task type. Several respondents in the “no change” group noted that they’re still learning how to use the tool, which suggests training and case-sharing can lift the numbers further.

3. Established usage patterns

Productivity / research support

• Faster lookups and research
  • Specialized-term lookup (ChatGPT instead of Google)
  • Reading, organizing, and understanding accounting standards and legal texts
  • Custom GPTs for internal rules and extension numbers
• Operational support
  • Accounting decisions (used before escalating to managers)
  • Manual GPTs (chat-based question-answering)
  • Document drafting and editing support
  • Data organization and analysis support
  • Excel formula suggestions and on-demand calculations
  • Excel automation
• Communication support
  • Sparring before talking with a manager
  • English translation
  • Drafting chat replies

Creative / ideation support

• Idea generation (describing the situation and asking for variants)
• Brainstorming support and problem-solving

4. Workload changes

Reduced workload

• Time savings
  • Faster accounting decisions, with new-case throughput more than doubled
  • Shorter email-drafting times
  • Faster chat replies
  • Less overtime
• Quality and efficiency improvements
  • Research and other work can run in parallel
  • Fewer escalations to managers — less load on managers
  • Better tool utilization (less time hopping between systems and tools)
  • Higher accuracy and faster turnaround on monthly close and financial-data analysis