One business question costs far more than the time to look it up. Add the hours to pull and match exports, the owner's time to check the answer, the tools that hold the data, and the chance that a wrong answer gets acted on. In our illustrative math, a mid-sized brand spends about $240 per question by hand. The wrong answers cost more than the hours.
Meet Priya. She runs a made-up travel mug brand.
It is Monday, 8:40 a.m. Her spring promo ended on Friday.
Sales were up 31%. The team is cheering in the group chat.
Then her accountant calls. Did the promo make money after discounts and fees?
Priya does not know. She asks her analyst.
The analyst pulls four exports and lines them up. Amazon, the store, Meta, the fee report.
The answer lands on Wednesday afternoon. The promo lost money.
By then, the ad set behind it has spent for two more days.
One business question costs the hours to find the answer, the time to check it, the tools that hold the data, and the risk of acting on a wrong answer.
Most brands only count the first. The last one is usually the biggest.
So let’s put a price on Priya’s question, three ways.
The made-up brand
All numbers from here on are illustrative. Swap in your own.
Priya’s travel mug brand sells about $4 million a year. Most of that is on Amazon. The rest is on its own store. It runs Meta ads to both.
The team asks about 15 real questions a month. Not “what were sales yesterday,” but questions like these.
- Did the spring promo make money after discounts and fees?
- Is the new lid worth its higher cost?
- Are Meta ads paying back on Amazon, or only on the store?
- Why did margin drop in March?
The team also wants a weekly report every Monday.
The four costs in every answer
Every answer has four costs. Some are on an invoice. Some are hidden.
| Cost | What it is | Where it shows up |
|---|---|---|
| Pull and match | Hours spent getting exports and making them agree | Payroll |
| Check | The owner’s time reading the answer and asking “is that right?” | The owner’s calendar |
| Tools | Subscriptions for the places the data lives or is shown | Monthly invoices |
| Wrong answer | What it costs when a wrong answer gets acted on | Lost money, months later |
To put a price on time, we need hourly rates. The analyst rate comes from public pay data. The owner rate is illustrative.
- Analyst time: $54 an hour. The U.S. Bureau of Labor Statistics puts median pay for market research analysts at $37.87 an hour, as of May 2025 (BLS Occupational Outlook Handbook). Pay is not the whole cost, though. In June 2026, wages were 70% of what private employers spent per hour worked (BLS employer costs release). The rest is benefits and taxes. So $37.87 divided by 0.70 is about $54.
- Owner time: $150 an hour. An owner’s hour is worth more, because it is the scarcest hour in the business.
For the wrong-answer cost, we use simple expected value. Expected value means the cost times the chance it happens.
- The cost of acting on a wrong answer: $4,000. Say a wrong read keeps a losing ad set alive for six extra weeks. At $500 a day of spend, losing 20 cents on every dollar, that is 42 days times $500 times 0.20. That comes to $4,200. We round to $4,000.
- Of the wrong answers, 1 in 5 gets acted on. The rest get caught or ignored.
Now the three ways. Here is where it gets expensive.
Way 1: by hand
An analyst pulls exports from Amazon, the store and Meta. They paste them into a sheet. They line up the dates, the product names and the fees.
Per question. Three hours of analyst time is $162. Half an hour of owner time to check it is $75. That is $237 a question. At 15 questions, that is $3,555 a month.
The weekly report. Five hours a week at $54 is $270. There are about 4.33 weeks in a month. So that is about $1,170 a month.
Wrong answers. Hand-matched sheets break in quiet ways. A date range is off by a day. Amazon orders get mixed with Amazon payouts. Say 1 in 10 answers is wrong in a way that matters. Then the expected cost is 0.10 times 0.20 times $4,000. That is $80 a question, or $1,200 a month.
Tools. Call it zero. The data sits in the systems the brand already pays for.
Total by hand: about $5,925 a month.
Way 2: a stack of dashboards
Next, the brand buys three dashboard tools. One for ads, one for the store, one for Amazon. Say each costs $600 a month.
Tools. Three times $600 is $1,800 a month.
Per question. The charts are already built, so pulling is faster. But the three tools count things differently. One counts a sale on the day of the click. Another counts it on the day of the order. Someone still has to make them agree. Say that takes 1.5 hours, or $81. Add the same half hour of owner time, $75. That is $156 a question, or $2,340 a month.
The weekly report. Two hours a week to copy charts into one place and write the notes. That is $108 a week, or about $470 a month.
Wrong answers. Dashboards help, but they add a new trap. Two tools show two numbers for the same thing, and someone picks one. Say 8 in 100 answers are wrong in a way that matters. The expected cost is 0.08 times 0.20 times $4,000. That is $64 a question, or $960 a month.
Total with dashboards: about $5,570 a month.
That is cheaper than by hand, but not by much. The fees eat most of the hours saved.
Way 3: an AI on one checked model
Last, the brand writes its business down once. Every number gets one meaning. Then it lets an AI answer questions from that model, with each number checked against its source.
We use Synthesis here, since its price is public: $999 a month, whole team. The AI app itself, such as Claude or ChatGPT, is a separate bill. Call that $100 a month for this team.
Tools. $999 plus $100 is $1,099 a month.
Per question. No pulling and matching. The owner still reads the answer and checks its sources. Say 15 minutes, or $37.50. At 15 questions, that is about $563 a month.
The weekly report. It arrives on Monday. Nobody builds it. $0 a month in build time.
Wrong answers. These do not go to zero. A checked model cuts the classic errors, like mixed bases and stale costs. But a fact can still be entered wrong, or a question can be asked loosely. Say 3 in 100 answers are wrong in a way that matters. That rate is a guess for this example, like the other two. It is not a measured accuracy figure for any product. The expected cost is 0.03 times 0.20 times $4,000. That is $24 a question, or $360 a month.
Total with an AI on one checked model: about $2,022 a month.
The three ways side by side
Illustrative numbers, for the made-up mug brand, 15 questions a month plus a weekly report.
| Cost per month | By hand | Dashboards | AI on one checked model |
|---|---|---|---|
| Tools | $0 | $1,800 | $1,099 |
| Questions (pull, match, check) | $3,555 | $2,340 | $563 |
| Weekly report build | $1,170 | $470 | $0 |
| Expected cost of wrong answers | $1,200 | $960 | $360 |
| Total | $5,925 | $5,570 | $2,022 |
| Cost per question, all in | about $395 | about $371 | about $135 |
The last row divides each total by 15. It folds the weekly report and tools into each question.
Look at the wrong-answer row. By hand, it is bigger than the cost of building the weekly report. It is also the cost nobody puts on a budget.
When the math flips
But this math does not win for every brand. Change the inputs and the answer changes.
Say the brand asks only 3 questions a month and does not need a weekly report.
- By hand: 3 times $237 is $711. Add wrong answers, 3 times $80, which is $240. The total is $951.
- AI on one checked model: $1,099 in tools. Add 3 times $37.50, which is about $113. Add 3 times $24, which is $72. The total is about $1,284.
At that size, by hand is cheaper. If you sell on one channel, ask a few questions a month and run no ads, a good spreadsheet may be all you need.
The math also flips if your error rate by hand is already very low. A sharp analyst who knows your data well can beat our 1 in 10. Use your own rate if you know it.
The cost the table leaves out: waiting
There is one more cost we did not price. Time to answer.
By hand, the promo question takes two days. A dashboard might take an afternoon. On a checked model, it takes minutes.
Waiting has a price when money is moving. A losing ad set at $500 a day keeps spending while the question sits in someone’s queue. Two days of waiting is $1,000 of spend at risk. We left this out of the table to keep the math honest and simple. It is real, though.
What makes a wrong answer likely
Wrong answers are not random. They come from the same few inputs every time.
- Mixed bases. Orders placed and cash paid out are added as if they were one number. They are not. Amazon settles seller accounts about every two weeks and holds some money back as a reserve (Amazon’s guide to seller payments). Shopify counts payout time in business days only, so weekend sales land in a later payout (Shopify Help Center).
- Stale costs. A product cost from last spring is used for this month.
- One tool’s guess taken as fact. Meta’s report of Amazon sales is an estimate. Meta cannot see the Amazon checkout.
- Half-finished periods. This week so far is compared with all of last week.
An AI does not fix these by itself. Put an AI on raw exports and it gives the same wrong answers, only faster and with more confidence. The fix is the model underneath. We cover the common traps in why AI gets your ecommerce numbers wrong, and the safety rails in what is an AI harness.
Work out your own number
You can run this math in 20 minutes.
- Count your real questions. Look back over the last month. How many times did someone need a number that took more than ten minutes to find?
- Time one. Pick a recent question. Add up every hour spent on it, by everyone.
- Price the hours. Use a loaded hourly rate for each person. Loaded means pay plus taxes and benefits. If you only know the pay, divide it by 0.70, the BLS wage share above.
- Add the weekly report. Hours per week, times the rate, times 4.33.
- Add your tool fees. Every tool that holds or shows the data.
- Guess your wrong-answer rate. Think of the last ten answers. How many turned out wrong later? Then what did acting on one cost?
- Compare. Run the same math for any option you are weighing.
If the total is small, keep doing it by hand. If it is large, the biggest line is usually the one you never invoiced.
Wrap-up
A question is not free just because nobody bills for it.
It costs the hours to pull and match, the owner’s time to check, the tools and the risk of a wrong answer. For a busy multi-channel brand, the wrong answers often cost more than the hours.
For our made-up mug brand, moving to one checked model cut the monthly cost from about $5,900 to about $2,000. For a brand that asks three questions a month, it would not. Do your own math before you decide. If you want a place to start, our post on what AI can take over for a brand lists the first jobs worth handing over.
Questions people ask
How much does it cost to answer a business question?
Count four things: the hours to pull and match the data, the owner’s time to check it, the tools the data lives in, and the expected cost of a wrong answer. In our illustrative example, that comes to about $240 a question done by hand.
Is it cheaper to hire a data analyst or buy dashboards?
It depends on how many questions you ask. Dashboards cut the time per question but add monthly fees. They can also show different numbers for the same thing, so someone still has to match them up.
How much time does a weekly business report take to build by hand?
For a brand selling on Amazon, its own store and paid ads, a few hours a week is common. Our illustrative example uses five hours. Over a year that is about 260 hours, before anyone reads it.
What is the cost of a wrong business decision?
Multiply how often an answer is wrong, how often a wrong answer gets acted on, and what the action costs. A losing ad set kept alive for six extra weeks can easily cost thousands of dollars.
When is answering questions by hand the cheaper option?
When you ask only a few questions a month, sell on one channel, and do not need a weekly report. In our illustrative math, a brand asking three questions a month with no weekly report is cheaper by hand.
Does using AI make business answers free?
No. You still pay for the tool and for someone to read the answer. And an AI on bad inputs gives confident wrong answers quickly. The saving comes from fewer hours and fewer wrong answers, not from zero cost.