01 / SUMMARY
The same working week. One difference.
We set out to measure what Styrke does for a Swedish business, in time and money, under controlled simulation conditions. We created three businesses, then made three identical versions of each. The only difference was the tool: Styrke, a standard AI assistant, or no AI. All nine versions received the same emails, customers and events, at the same minute, over eight working days.
With Styrke, 85 percent of tasks were completed on time, compared with 68 percent with standard AI and 61 percent without AI. Staff spent 38 percent less time on the same work. Annualizing these results under the stated assumptions gives a business like Nordvik Klimat, with 40 employees, an estimated net value of around $83,579.98 after the assumed Styrke cost.
02 / WHY A SIMULATION
No two weeks are the same. Here, they are.
In a real business, it is difficult to see what would have happened without a particular tool. A simulation lets us run the same weeks three times, changing one thing. It follows the basic principle of a controlled study: hold the conditions constant, change one variable, then compare the outcomes.
03 / THE SIMULATED WORLD
Three businesses, six customers and a typical working week.
The simulated world includes email, calendars, customer records and accounting. Everything runs inside that world. No email left it, and no real person was contacted.
Customers make their own decisions. The six customer businesses, including Vasastadens Kontorsfastigheter, Skeppsbrohotellen and Vätterstrandens Livsmedel, have simulated AI buyers. They decide whether to buy, negotiate, complain or pay late, applying the same rules to all three versions.
Staff have a finite working day. Each business has a CEO, a salesperson, a finance lead, a customer service role and a project manager. Each has limited time. When demand exceeds that capacity, the remaining tasks queue up.
The businesses are already established. The Styrke versions represent businesses that have used Styrke for six months. Styrke builds its knowledge from their history. In the other versions, staff have worked there for years and know their customers.
04 / THREE WAYS OF WORKING
The same business. Different tools.
Styrke's agents carry out the work using the application code. The simulated CEO reviews and approves each email before it is sent. Styrke's work is measured in the run rather than replaced by a fixed time assumption.
Staff use a standard AI assistant to draft text. We assume it reduces time on each writing task by 40 percent, based on a study published in Science. Staff handle everything else themselves.
Staff handle every task manually. Task durations come from a conservative time table, with a rationale for each estimate.
05 / A DAY IN THE SIMULATION
From 6 am to 6 pm, for eight days.
06 / HOW WE MEASURED IT
A tougher test for Styrke. By design.
Each task Styrke performed was checked against a reference fixed before the run: the correct customer, price, quantity and date, and completion before the customer's deadline. 82 percent of Styrke's work passed these checks.
When Styrke could not complete a task, staff took over and the full assumed human task time was counted. If incorrect output was sent, rework time was counted as a cost. Staff and standard-AI output were assumed to be correct. These assumptions favor the alternatives; the reported outcome still favored Styrke in this simulation.
07 / RESULTS
Faster replies, more tasks on time, and less staff time.
| The same 117 tasks | With Styrke | With AI | Without AI |
|---|---|---|---|
| Completed on time | 85 % | 68 % | 61 % |
| Customer-meeting commitments kept | 6 of 9 | 2 of 8 | 2 of 8 |
| Staff time | 31 hours | 43 hours | 53 hours |
| Time reduction, including errors | 38 % | 19 % | 0 % |
08 / FROM EIGHT DAYS TO AN ANNUAL ESTIMATE
From a simulated working week to an annual estimate.
The simulation measures the reduction in staff time for the recorded work. Extending it to an entire business and year requires additional inputs. Each has a cited source or is explicitly marked as our estimate. Here is the calculation for Nordvik Klimat:
09 / PROTECTING THE INTEGRITY OF THE RESULTS
Results should stand up to scrutiny.
10 / WHAT THE SIMULATION CANNOT SHOW
The limitations, set out in full.
Eight days is a short period from which to estimate a year. Customers are simulated, even though they make their own decisions. Human task times and the share of administration covered by Styrke are assumptions. Styrke also does not outperform on every task, including meeting scheduling. We show those cases, their recorded causes and subsequent changes.
Other work is excluded entirely: evenings and weekends, recruitment, contract review and knowledge lost when someone leaves. Those effects are not included in the estimated value.
See where Styrke does not outperform yet11 / EVERY ASSUMPTION
Each input and its source.
| Assumption | Value | Source |
|---|---|---|
| Cost per working hour | $36.43 | Statistics Sweden, wages 2025; Swedish Tax Agency, employer contributions 2026 |
| Share of the day spent on email | 28 % | McKinsey Global Institute, 2012 |
| Share of the day spent finding information | 19 % | McKinsey Global Institute, 2012 |
| Administrative roles: construction, manufacturing, transport | 27 %, 47 %, 38 % | Statistics Sweden, occupational register, 2024 |
| Share of administration covered by Styrke | 90 % | Styrke's estimate |
| Time reduction from standard AI on a writing task | 40 % | Noy and Zhang, Science, 2023 |
| Working days per year | 220 | 52 weeks, less annual leave and public holidays |
| Writing a quote manually | 90 min | Conservative assumption |
| Replying to a customer email manually | 10 min | Conservative assumption |