Why Cherry Place
A business owner with 250 hours can spend them learning AI or spend them on the business. Below is the equation we use, with every number open. Put in your own and see if it applies to you.
$34,500
A build quoted under this costs less than the 230 hours you keep, before counting anything it gives back. Based on 250 hours at $150 an hour, 20 of them yours for review.
$51,125
Your 230 hours, plus the $16,625 more a running build gives back by month 12 than going alone does. A build quoted under this costs less than doing it yourself over twelve months; at exactly this, the two come out the same. Every year after, going alone leaves $20,020 behind, at these numbers.
230 of your hours stay on the business. The build happens beside them.
Illustrative defaults. Every number changes below.
Your numbers
Drag or tap any underlined number. Everything on the page follows. The numbers live in your address bar, so you can copy the link and bring it to a call.
If you spend hours on AI at an hour of your time, about hours a week, doing it yourself costs $37,500, gets you about of the way, and the first useful thing runs around month 3.
A practitioner who gets you to on the same job, running by , with hours of yours for review, costs less than the hours you keep if the build is quoted under $34,500. Counting what each version gives back in the first year, not hiring costs you $51,125: a build quoted under that is cheaper than doing it yourself, and at that exact quote the two come out the same. Every year after, going alone leaves $20,020 behind, at these numbers.
Prefer to keep the hours yourself? months of Claude Concierge is $11,700 and, in our experience, takes an owner to about , with the first pieces running inside month 1. At these numbers it returns $12,110 more than going alone in the first year, so it comes out ahead by $410 at month 12, and runs $14,560 a year higher after. The fluency stays with you and your team; we don’t put a number on that. See Claude Concierge
That first-year math assumes the finished work gives back hours a week across your team once it’s fully running, and that one of those hours costs you today: $3,033 a month at 100%.
We don’t prefill this.
Type any quote and this line compares it with both break-evens.
Every number here is an assumption you can change. The defaults are ours and illustrative, not research. We assume value scales with how far you get, credit nothing before a build is running, and leave recurring costs (subscriptions, AI usage fees, upkeep) out of all three paths equally. The last points of a system are usually worth more than the first, so tying value to how far you get is the conservative reading. If the break-even is below any quote you’d get, the math says do it yourself, or start with coaching.
Press Enter for a slider.
Break-even hours alone $34,500; not hiring costs $51,125 in year one and $20,020 each year after. On your own reaches 40 percent around month 6; coaching 80 percent; a practitioner 95 percent by month 3. Coaching ahead by $410 at month 12.
The unit of comparison, not a forecast.
Not your salary. What an hour of your attention earns or protects. For reference, the BLS median for chief executives works out to about $103 an hour (at 2,080 hours a year); a private-company CEO survey median to about $170 (at 2,500).
Our working assumption.
Applies to the hours the coaching covers.
$1,950 a month, month to month.
Across the whole team. Four people saving five hours a week is 20.
All-in cost of an hour of the work AI takes over (pay plus taxes and benefits). The BLS median pay for general and operations managers is about $51 an hour; most repetitive work costs less.
On your own, we assume nothing useful runs until half your hours are spent. With coaching, after a tenth. Coaching lifts the hours it covers: a three-month engagement over six months of hours lifts about half of them.
The same 250 hours
Same job, same hours, three people doing them. Every column reads from the numbers above.
| On your own | With coaching | Built for you |
|---|---|---|
| Who does the work | ||
| You | You, with coaching | A practitioner; you review |
| Your hours | ||
| 250 | 250 | 20 (230 stay on the business) |
| Your cost | ||
| $37,500 of your time | $37,500 + $11,700 Concierge | $3,000 + the build’s price |
| How far it gets | ||
| 40% | 80% | 95% |
| First useful thing running | ||
| around month 3 | inside month 1 | month 3 |
| Finished | ||
| around month 6 | around month 6 | month 3 |
| Gives back in year one | ||
| $9,310about 265 h | $21,420about 610 h | $25,935about 740 h |
| Each year after | ||
| $14,560 | $29,120 | $34,580 |
Illustrative. Change any input above.
Illustrative. Change any input above.
Claude ConciergeIllustrative. Change any input above.
Claude CustomPress a path to follow its curve.
| Hours of work | On your own | With coaching | Built for you |
|---|---|---|---|
| 0 | 0% | 0% | 0% |
| 50 | 3% | 15% | 41% |
| 100 | 12% | 46% | 82% |
| 150 | 28% | 70% | 93% |
| 200 | 37% | 78% | 95% |
| 250 | 40% | 80% | 95% |
Built for you
$25,935 · about 740 h back
With coaching
$21,420 · about 610 h back
On your own
$9,310 · about 265 h back
Hours back are the finished work’s hours, not yours. The built path also leaves your 230 hours on the business, which the break-even counts separately. We stop at month 12; each year after, the three run at $14,560, $29,120 and $34,580 at these numbers.
| Path | First useful thing running | Finished | Year one | Hours back | Each year after |
|---|---|---|---|---|---|
| Built for you | month 3 | month 3 | $25,935 | about 740 hours | $34,580 |
| With coaching | inside month 1 | around month 6 | $21,420 | about 610 hours | $29,120 |
| On your own | around month 3 | around month 6 | $9,310 | about 265 hours | $14,560 |
Inside the hours
This is where the how-far-it-gets numbers come from. Learning and false starts are real hours; they just don’t ship anything.
On your own250 hours
| Segment | Hours |
|---|---|
| Learning the tools | 60 |
| False starts and rebuilds | 65 |
| Building | 85 |
| Keeping it running | 40 |
With coaching250 hours
| Segment | Hours |
|---|---|
| Sessions | 12 |
| Setup and first workflow | 25 |
| Building on your own work | 193 |
| Support-line fixes | 20 |
Built for youyour 250 hours
| Segment | Hours |
|---|---|
| Discovery, review and training | 20 |
| Stay on the business | 230 |
Practitioner levels
This is how we describe the ladder. It’s our framework, not a standard. The hours are our estimates, and the percentages are how far each level gets on a job in 250 hours.
Drafts, summaries, one-off answers. Nothing that runs on its own.
Repeatable prompts and templates. A scheduled task that half works.
Working AI workflows and a first agent that owns one job. A basic knowledge base.
Agent fleets on a schedule, knowledge bases the agents answer from, context packs that move between tools, monitoring so it stays running.
Everything at level 4 plus the operating model: who approves what, where people stay on the front lines, what to measure.
Level 1 to 3 is roughly 500 to 1,000 hours of practice; level 3 to 5 is thousands of hours across many builds (our estimates). The percentages are how far each level gets on a job in 250 hours.
| Level | Who they are | What they can build | Time to ship one working workflow | Where the risk shows | How far it gets in 250 hours | Practice to get here |
|---|---|---|---|---|---|---|
| 1 · Curious user | Uses a chat window. Asks questions, pastes answers into their work. | Drafts, summaries, one-off answers. Nothing that runs on its own. | Not yet. The work happens in the chat, one question at a time. | Doesn’t know yet what AI can’t do. | about 15% | The first 20 to 50 hours. |
| 2 · Power user | Saved prompts, projects, custom instructions, a few connected files. | Repeatable prompts and templates. A scheduled task that half works. | Weeks, usually with help; often set aside when the inputs change. | Output varies; nothing is connected to anything. | about 30% | 100 to 250 hours. Most owners at the 250-hour benchmark land between 2 and 3. |
| 3 · Builder | Has shipped working workflows on one platform. Knows its limits from experience. | Working AI workflows and a first agent that owns one job. A basic knowledge base. | A couple of weeks of evenings per workflow, with rebuilds. | Works on the builder’s machine; brittle when the business changes. | about 60% | 500 to 1,000 hours across many builds. |
| 4 · Systems builder | Builds across platforms. Thinks in fleets, knowledge bases, context packs, and what running AI costs against the hours it saves. | Agent fleets on a schedule, knowledge bases the agents answer from, context packs that move between tools, monitoring so it stays running. | Days for an agent inside an existing fleet. | Knows where the tool fails silently and designs around it. | about 85% | 1,500 hours and more, in real businesses. |
| 5 · Transformation lead | Redesigns how work moves through a company, then trains the team to run it. | Everything at level 4 plus the operating model: who approves what, where people stay on the front lines, what to measure. | Same speed as level 4; the extra time goes to the organization, not the code. | Same as level 4, plus knows what to leave with people. | about 95% | Years, across companies. |
At the 250-hour benchmark, an owner gets through the first two levels and a look at the third. Anthropic’s Economic Index (March 2026) found people who had been on Claude six months or more got better results and asked harder questions, and kept improving year over year.
Knowing where the tool fails silently. In the HBS/BCG field study (2023), consultants using AI on a task outside its strengths were 19 points less likely to be right. RAND, The Root Causes of Failure for AI Projects, 2024 (opens in a new tab) names the same failure causes: the problem was framed wrong, the data wasn’t there, the team chased the technology instead of the problem, nothing was set up to run it, or the task was beyond the tool.
Most owners at the 250-hour benchmark land between level 2 and level 3. A practitioner is someone at level 4 or 5 who has already done the climbing, so their hours all go to the build. In the equation, that’s the built-for-you path.
Our builds are done at levels 4 and 5. Claude Concierge coaches an owner from wherever they start toward level 3, on their own work.
Evidence
None of these measure our assumptions. They measure the direction: learning takes months, knowing where the tool fails is the skill, training and coaching raise regular use.
67%
Vendor-partnered AI deployments succeeded about 67% of the time in MIT NANDA’s July 2025 report; internal builds about one third as often. Enterprise-skewed, not peer-reviewed.
19 points
In a 2023 field study of 758 BCG consultants, AI users finished tasks 25.1% faster with 40% higher quality on tasks inside the tool’s strengths, and were about 19 percentage points less likely to be right on a task outside them.
Dell’Acqua et al., Harvard Business School Working Paper 24-013, Sept 2023 (opens in a new tab)
+34%
Support agents given an AI assistant resolved 14% more issues an hour on average, 34% more for novice and lower-skilled workers, with minimal change for experienced, highly skilled workers. The tool passed along the practices of the most able workers.
Brynjolfsson, Li and Raymond, NBER Working Paper 31161, 2023 (opens in a new tab)
6 months
People who had signed up for Claude six months or more earlier had about a 10% higher (relative) success rate per conversation, roughly 3 to 5 percentage points after controls, and the sophistication of their requests rose almost a school year for every year of use.
79% vs 67%
Regular AI use was 79% among employees with five or more hours of training, against 67% with less, and was higher again with in-person training and coaching. Only a third said they’d been properly trained.
BCG, AI at Work 2025, June 2025 (slideshow exhibit, n=10,635) (opens in a new tab)
76% / 14%
76% of U.S. small businesses say they use AI; 14% say it’s fully embedded in core operations; 73% say they would benefit from more training and implementation support.
Goldman Sachs 10,000 Small Businesses Voices, March 2026 (1,256 owners) (opens in a new tab)
We cite the publisher and the sample. Where a study is enterprise-focused, its card says so. The owner-hour anchors under the sliders are the BLS Occupational Outlook Handbook (opens in a new tab), May 2025 wages (chief executives $213,990; general and operations managers $105,770) and Chief Executive Group’s 2024–25 private-company compensation report (opens in a new tab) (median CEO total cash $425,000). The per-hour arithmetic is ours: divided by 2,080 hours a year for BLS (about $103 and $51 an hour), and by 2,500 for the CEO survey (about $170), since owners work longer weeks.
Where we fit
Both paths are in the equation above. Pick the one whose numbers work, or neither.
Two one-hour sessions a month and a human support line while you build on your own work. $1,950 a month, month to month. The 80% path in the equation.
We build the AI agents, AI workflows, agent fleets, knowledge bases and context packs, then train your team to run them. Your hours stay on the business. The 95% path in the equation. We quote each build; the equation above tells you the number a quote has to beat.
We build in Claude, ChatGPT, Codex, Cursor, Gemini, and Grok Bot.
Questions
It’s the unit of comparison, not a claim about how long anything takes. About five hours a week for a year, or ten for six months. Change it above and every figure follows.
They’re our working assumptions from doing this work, and the page labels them that way. No study measures how far a self-taught owner gets. The evidence section shows what has been measured. Change any percentage and the whole page follows.
Then the build costs more than the hours you’d keep. The second break-even counts what each version gives back in the first year; if your quote is under that, hiring comes out ahead over twelve months. If it’s above both, at those numbers the math says do it yourself, or start with coaching.
That’s what Claude Concierge is for. You do the building on your own work, with two sessions a month and a support line, and the fluency stays with you and your team. It’s $1,950 a month, month to month.
To stay conservative. A running system keeps giving hours back after year one, and, at the default numbers, the gap between 40% and 95% keeps paying. The each-year-after line shows what it gives back each year; we leave the rest uncounted.
They live in your address bar so you can copy the link. We don’t store them, and we see them only if you share the link with us.
Get started
30 minutes on where AI would pay off in your business and whether we’re a fit. Copy the link with your numbers and we’ll start from there.