Weekly pageviews, my own portfolio
626K → 1.7M
2.7x from the lowest June week (Jun 22) to the week of Oct 5.
Source: site analytics · weeks of Feb 23–Oct 5 2026 (to Oct 11)
Our philosophy
Build a Business That Gets Better Every Day.
Two ideas drive everything we build. AI should do every job it can do better, faster and cheaper than a person. And every result should make tomorrow's work a little better than today's. Put those together, with measurement and human limits, and you get a business that compounds.
Free, 30 minutes, phone or Google Meet. No pitch deck.
Where this starts
I stopped asking AI to help me run my business. I started asking AI to run the business.
For a long time I used AI the way most owners still do. A faster draft of an email. A caption. A summary of a long document. It was useful, the way spell-check is useful, and it barely moved the business.
In 2025 I changed the question. Instead of asking "what can AI help me with?", I went through my businesses one job at a time and asked, "can AI do this job better, faster and cheaper than a person?" Research. Writing. Images. Social posts. Ad launches. Bids. Email. Reporting. Website fixes. Questions about the books. When the answer was yes, AI got the job.
At first it felt reckless. Plenty broke. I fixed what broke, kept what worked and scaled it. And something happened that I didn't fully expect. The businesses didn't just get cheaper to run. They started getting better on their own schedule, every week, because every job AI did produced a result, and every result went back into the rules for the next job.
That second part is what this page is about. Handing work to AI is the first idea. Making the work improve every day, safely and measurably, is the second. The old name for that second idea is Kaizen.
Kaizen
Small, measured improvements. Every day. For as long as the business runs.
Kaizen is a Japanese word that roughly means "change for the better". As a business practice, it grew up in Japanese manufacturing after the Second World War, most famously at Toyota, and Masaaki Imai's book Kaizen introduced it to Western managers in the 1980s.
The idea is simple, and that's its strength. Instead of waiting for a big transformation once every few years, you make small improvements constantly. The people closest to the work notice what's slow, wasteful or broken, try a fix, measure whether it helped, and keep it if it did. Then they do it again tomorrow. No single change is dramatic. The sum of thousands of them is.
Factories that practiced it well became famously hard to compete with. It wasn't because they had one brilliant idea. It was because they got slightly better every day and their competitors didn't.
Here's the problem for an ordinary business. Kaizen needs two things most owners don't have: time and measurement. Somebody has to look at the results every day, notice what changed, propose a fix and check whether it worked. In a 12-person plumbing company or a 3-person online store, nobody has those hours. So improvement happens in bursts: a new website every five years, a new agency every two, a strategy meeting once a quarter that runs on memory.
In an AI-operated business, the people closest to the work are AI systems that never stop measuring. That changes everything about Kaizen. The noticing, the measuring and the proposing can happen every day, on every campaign, page, post, call and lead, without anyone losing an evening to it. The owner's job becomes the part that always mattered most: deciding which changes to make and setting the limits.
The one rule
"AI should do everything possible because it does it better, faster and cheaper."
That's the rule I run my businesses on, and every job has to pass all three tests.
Betterequal or higher quality than a person doing the same job.
Fasterdone in minutes, around the clock.
Cheapera lower cost per result.
If AI can't do a job as well as your people, a person keeps it, and we check again as the models improve. We never add AI for the sake of AI, and you never trade quality for a buzzword.
Why bolting AI onto the old process fails
There's a common mistake I see in almost every business that tries AI. They take the existing process, exactly as it is, and stick AI on one step of it.
The marketing manager still writes a monthly plan, still waits for the agency's report, still checks the ad accounts on Fridays, but now she uses a chatbot to draft the campaign copy. The office still returns calls when it gets a minute, but now there's an AI-written script. The owner still gets a profit-and-loss three weeks after month-end, but now it has an AI-generated summary on top.
Every one of those is a slightly faster version of a slow process. The bottleneck was never the copy, the script or the summary. It was that the campaigns were looked at once a week, the calls were returned when someone had a minute, and the numbers arrived after the decisions were made.
So we redesign the work before we hand it over. We ask what the job would look like if production were nearly free and attention never ran out.
If ads can be checked every morning, there's no reason to judge them monthly.
The job becomesJudge every campaign daily, by rule.
If a page costs cents to produce, there's no reason to publish two a week.
The job becomesPublish what customers search for, check every page, promote it the moment it's live.
If a reply can go out in a minute at 9 pm, there's no reason to wait for the office to open.
The job becomesAnswer every lead within minutes, qualify it and book it.
If yesterday's numbers can be reconciled overnight, there's no reason to wait for month-end.
The job becomesA checked profit-and-loss every morning.
The redesign is where most of the gain lives. AI is the reason the redesign is now possible.
Delegation, job by job
Go through the business one job at a time. Hand over what passes the test.
"Use AI in the business" is too vague to act on. "Can AI judge our Google Ads campaigns every morning better than our agency does monthly?" is a question you can answer. That's why we delegate job by job, never department by department in one leap.
In the Week 1 audit we list the jobs in your business: every recurring task in marketing, sales follow-up, scheduling, reporting and admin. Then, for each one, we ask five questions.
- What does it produce? If we can't say what a job produces and measure it, we can't tell whether AI is doing it better. Measurement comes first.
- Does AI pass all three tests today? Better, faster and cheaper, honestly assessed. Some jobs pass easily. Some don't yet.
- What does it touch? A job that spends money, contacts customers or changes the business gets limits and an approval step before it's handed over.
- What's it worth? We rank jobs by the money and hours they'd free up, so you start where the payback is largest.
- What does it depend on? Ad judging depends on tracking. Follow-up depends on knowing where the lead came from. Dependencies decide the order.
That's why most businesses start with tracking, ad buying, lead follow-up and the morning report. They pay back fastest, and everything else depends on them.
Some jobs stay human, and we're specific about which ones: setting the goals and limits, approving big decisions and new strategies, pricing, hiring, relationships with customers, partners and staff, judgment calls on brand, ethics and risk, and any job AI can't yet do as well as a person. The list of jobs AI can't do well gets shorter every year, so we check it again as the models improve.
The loop
Every job measures itself. Every result changes the next job.
A feedback loop is the engineering behind Kaizen. It's what turns "AI does the work" into "AI does the work a little better every week". Every system we build runs the same six steps.
- AI performs the work. It launches the campaign, publishes the page, answers the lead, sends the reminder.
- The system measures the outcome against its source. Spend is matched to the ad account. Leads are matched to the page and the ad. Revenue is matched to the books. Nothing runs on "seems fine".
- AI analyzes what worked and what didn't. Which campaigns paid. Which pages earned attention. Which follow-up message got replies.
- The system identifies improvements. A rule that misfired. An offer that should get more placements. A format that should be retired.
- Approved or pre-authorized changes are implemented inside set limits. Small, routine adjustments you've pre-authorized happen automatically. Bigger ones wait for your approval.
- The process repeats, and the lesson is written down. The rule change, the reason and the result are recorded, so nothing is learned twice and nothing walks out the door with an employee.
The loop runs at different speeds for different jobs. Ad campaigns are judged every morning. Social formats are re-ranked weekly. Strategy proposals come once a week. The rules that govern all of them are themselves tested against results.
That last point is the one most systems miss. A loop that only adjusts the work, and never questions its own rules, will happily repeat a bad rule forever. Ours puts the rules on trial too.
Kaizen you can see: in a command center, every experiment, result and rule change is logged in the improvement history. See an improvement history in the live demo →
Experiments, not opinions
Every idea gets a hypothesis, a check-in date and a verdict.
Most businesses have more ideas than they test, and test more than they measure. Something gets "tried" for a while, nobody writes down what success would look like, and six months later nobody remembers whether it worked. The idea quietly becomes either a habit or a ghost.
In an AI-run business, every change of any size is an experiment with four parts.
01A hypothesis.
What we think will happen, and why. "Readers will stay longer on pages with a stronger visual break between sections."
02A check-in date.
When we'll read the result. Not "eventually". A date, written down at launch.
03A verdict.
On that date, the result is read against the hypothesis: won, lost or inconclusive.
04A rule change.
If it won, it becomes the default. If it lost, it's retired and recorded so it isn't tried again by accident. If it was inconclusive, it either gets a longer run or gets dropped.
That's how my own businesses run. Page designs, ad styles and offers are tested this way: each version is shown to a share of visitors or campaigns, the result is read on a set date, the weakest versions are retired and the survivors are ranked again. None of it would happen in a business where someone has to remember to check.
The discipline matters more than any single experiment. Over a year, a business that runs hundreds of small, measured experiments and keeps only the winners ends up somewhere its competitors can't easily follow.
The 24/7 self-improvement layer
A team of analysts and operators working on your marketing and your business, 24 hours a day, 7 days a week.
Imagine hiring a team whose only job is to make your business a little better every day. An analyst who reads every result from every channel overnight. An operator who acts on it first thing in the morning, inside the limits you set. A tester who runs measured experiments on offers, headlines, audiences and timing. And a record-keeper who writes every lesson into the rules so nothing is ever relearned. That team would cost a fortune, and it would still go home at five, take weekends off and forget things.
Behind every system we build sits an automated self-improvement layer that does that work around the clock. It watches what's working and what isn't, across every ad, page, post, email, call and lead, and makes the results better every day. There are no meetings to wait for, no monthly review, and nobody waiting until they have time. It works nights, weekends and holidays. Campaigns are judged at dawn. Leads are answered at 2 am.
The analystReads every result overnight, from every channel. Finds what's winning, what's slipping and why, before you've had your coffee.Every night
The operatorActs on it inside your limits: moves budget to winners, pauses losers, refreshes tired creative, retimes follow-ups.Every morning
The testerRuns measured experiments on offers, headlines, audiences and timing, each with a check-in date and a clear verdict.Every week
The memoryWrites every lesson into the rules, so nothing is relearned and every win becomes the new starting point.For as long as the business runs
This isn't a promise about some future product. It's how my own businesses run today. In one week this year the layer launched 299 new ad campaigns, part of 1,878 launched between Jun 22 and Oct 11 2026. Every live campaign is judged every morning, and every page, post and lead is measured against its source. It's a big part of why my weekly pageviews grew 2.7x from the lowest week in June to the week of Oct 5 2026, instead of drifting flat.
To be precise about what the layer is: it's a set of scheduled AI jobs with rules, limits, measurement and a written memory. It doesn't make decisions you haven't authorized, and it isn't magic. It's simply the kind of attention a business has never been able to afford before.
2.7x since the lowest June week
Jon's own portfolio. Figures are from businesses Jon Dykstra owns and operates on AI. They are not client results. Data to Oct 11 2026 unless noted.
Free, 30 minutes, phone or Google Meet.
Kaizen in practice
Four improvements from my own businesses, and what they taught me.
1 · The ad rules
The ad rule that was killing future winners.
Before AI
- Campaigns judged too soon
- Slow starters paused before they could prove themselves
With AI running it
- The pause rule itself was tested against real results
- The rule was changed to give campaigns a fairer chance before judging
- Profitable campaigns that would have been switched off keep running
- 01Judged too soonstopped future winners
- 02Rule testedagainst real results
- 03Rule changedfairer chance first
- 04Winners surviveand keep paying
The rules are measured as carefully as the work
One rule change, found by measuring the rules themselvesJon's own ad accounts, 2026
My ad system pauses losing campaigns by rule. The rule that decides when to pause an ad was itself tested against real results. The early version judged too soon and stopped campaigns that would have become winners: some campaigns start slowly and turn into winners a few days later. So the rule was changed, and profitable campaigns that would otherwise have been switched off kept running.
No person had to spot it. The loop spotted it because the loop checks its own rules.
2 · The content system
Content that learns what to write next.
The content system reads which topics, titles and formats earned the most attention, and uses that to choose and shape the next batch. Every week of results makes the next week's pages more likely to win. Meanwhile, a second AI checks every page and every image before it goes live, which is how volume grew without quality slipping. My businesses went from 35 pages a week to 145 (4.1x) between Jun 1 and the week of Oct 5 2026. The AI editor alone took the place of a US$6,000-a-month editing role.
3 · The daily report
The report that refuses to lie.
My daily profit-and-loss is assembled overnight from the ad platforms, the websites, the email system and the books. One rule is non-negotiable: totals must tie out to their source. If ad spend in the report doesn't match the ad accounts, the report stops and says why, instead of guessing.
That sounds like a small technical detail. It's the foundation of everything else. A Kaizen loop is only as good as its measurements. If the numbers are wrong, the system will confidently "improve" toward the wrong thing. So the system would rather tell me it doesn't know than tell me something false.
4 · The websites
The speed fix, done in an afternoon.
Main images were loading slowly on several of my sites. In one afternoon, AI measured the problem the way a real phone experiences it, separated the real issue from testing noise, fixed image loading and verified the fix live on every affected site. In the same pass it found and removed a heavy third-party script that had been freezing pages on phones. In the old world that would have been a ticket, a developer, a quote and a few weeks.
Jon's own portfolio. From businesses Jon Dykstra owns and operates on AI. Not client results.
A day in an AI-run business
What actually happens, hour by hour.
This is a normal day in my own businesses. The parts that need me are in gold. Everything else happens without me.
Yesterday's numbers are collected from every ad platform, the websites, the email system and the books, then reconciled.
Every live ad campaign is judged on its real return. Losers are paused, winners get more budget, borderline ones are watched.
The morning report arrives: profit by channel, leads by source, what changed and what AI already did about it.
YouNew pages publish, each researched, written, illustrated and checked by a second AI.
The moment a page goes live, its ads are built and launched and its social posts are queued.
Posts go out on every page at the hours each audience is active. Leads are answered within minutes.
Speed and error checks run across the websites. Anything broken is fixed and re-tested.
A few strategy proposals to approve from my phone. AI turns the approved ones into measured experiments.
YouEmail sequences and newsletters send on schedule.
AI reviews every experiment, updates the rules and brings me next week's plan.
My part of the day is reading the morning report and approving a few decisions. That's where "roughly 25x the work in a quarter of the time" comes from. It's my own estimate, not a measured statistic, but it's how the days actually feel.
Humans keep authority
AI does the work. You still run the business.
A system that improves itself needs firm edges. Here are ours. You set the goals, budgets and limits. AI does the work inside them. Anything that spends money, contacts customers or changes the business can require your approval, and everything is logged.
Hard spending limits.
Daily and monthly budget caps and bid limits are set by you. The ad system can move money between campaigns inside those limits. It can't raise them.
Approvals where they matter.
You decide which actions are pre-authorized (pausing a losing campaign, retiming a reminder) and which always wait for a person (a new ad platform, a new offer, a bigger budget, a message on a sensitive topic). Proposals land in an approval queue you can clear from your phone.
A second AI checks the first.
Before a page, image, ad or message goes out, a separate AI reviews it for facts, tone, spelling and anything that breaks your rules. Anything doubtful goes to a person. In my businesses this check runs on every image and every page.
Read-only books.
Accounting connections are read-only by design. AI reports on the books. It doesn't move money.
Numbers that tie out.
Reports reconcile to their sources or stop and say why.
An audit log of everything.
Every launch, pause, budget change, message, fix and rule change is recorded with what was done and why. You can always answer the question "what did AI do yesterday, and why?"
Short alerts, not noise.
When something needs a person, you get a short message on your phone. Alerts that turn out to be noise get tuned out, so the ones that reach you matter.
Rules about what AI must never do.
Some lines are written down and enforced: words it may not use, claims it may not make, topics it must hand to a person.
What stays with you
- The goals, the budget and the limits
- Approval of big decisions and new strategies
- Pricing, hiring and customer relationships
- Judgment calls on brand, ethics and risk
- Any job AI can't yet do as well as a person
What AI takes over
- Research, writing, design and production
- Launching, judging and tuning ads inside your caps
- Answering, following up and booking leads
- Collecting, reconciling and reporting the numbers
- Spotting patterns and proposing the next test
The guardrails
- Hard spending caps and bid limits
- Approval steps on anything sensitive
- Read-only access to your books
- A second AI checks work before it goes out
- A short alert when a person is needed
You own it
Built on your accounts. Your data, your rules, your history.
Every system we build runs on your own accounts: your website, your ad accounts, your email platform, your phones, your field-service and accounting software. The tracking, the dashboards, the rules and the written record of what was learned all live in your business.
That matters for three reasons.
First, control. You can see everything, change any limit and switch any automation off. Nothing is hidden inside an agency's account.
Second, continuity. If we ever part ways, the systems, the data and the history stay with you. The lessons your business paid to learn don't leave with a vendor.
Third, value. A business with documented systems, clean numbers and a written playbook is easier to run, easier to hand to a manager and easier to sell.
If your team wants to run the systems themselves, we teach them. If you'd rather not, we run them for a monthly fee. Either way, you own all of it.
Straight talk
What we don't promise.
The word "AI" attracts a lot of nonsense, so here's what this philosophy is not.
It isn't magic.
The improvement comes from measurement, rules and repetition. Take away the measurement and you have AI producing a lot of work with no idea whether it's helping.
It isn't autopilot without measurement.
No job is handed to AI unless we can measure what it produces. No rule is changed without a result behind it.
It isn't "self-learning AI that improves on its own".
The loop is engineered: scheduled jobs, written rules, measured outcomes, approved changes. It improves because it was built to check its own work, inside limits a person set.
It isn't error-free.
AI makes mistakes. So do people. The system is built to catch mistakes early, with second-AI checks, reconciled numbers, rule tests and alerts, and to write down each one so it isn't repeated. The ad rule story above is a mistake the loop caught.
It isn't a guarantee.
The figures on this site come from my own businesses. Every business is different, and I won't promise you a number. What I can promise is the method.
It isn't AI for the sake of AI.
If a person does a job better, the person keeps it.
Questions
About the philosophy.
What does Kaizen actually mean in an AI-run business?
Small, measured improvements made every day. AI does the work, the system measures the outcome against its source, AI analyzes what won and lost, identifies an improvement, the approved change goes live inside your limits, and the lesson is written down. Then it repeats.
Will AI change my business without asking me?
Only in the ways you've pre-authorized, such as pausing a losing campaign inside your caps. Anything that spends more money, contacts customers in a new way or changes the business can require your approval. Every action is logged.
How do you decide which jobs AI takes?
Job by job. Each one has to pass three tests: better, faster and cheaper than a person. We also check what it produces, what it touches, what it's worth and what it depends on. If AI can't do a job as well as your people, a person keeps it, and we check again as the models improve.
What happens when AI gets something wrong?
A second AI checks work before it goes out, numbers must reconcile before they're reported, and rules are tested against results. When something slips through, it's caught, fixed and written down so it doesn't happen again.
Is this only for tech companies?
No. The best fits are ordinary businesses that win customers online: home services like plumbing, HVAC and roofing, other local businesses, ecommerce, publishers and online education.
Who owns the systems and the data?
You do. Everything is built on your own accounts, and the systems, the data and the history of what was learned stay with your business.
The next step
Most businesses get better in bursts. Yours could get better every day.
Think about how your business improves today. A new website every few years. A new agency when the old one disappoints. A strategy meeting when there's time. Each one is a jump, followed by a long, slow drift.
Now picture the alternative. Every campaign judged every morning. Every page, post and lead measured against its source. Every idea tested with a date and a verdict. Every lesson written down. A small improvement today, another tomorrow, and a business that looks noticeably different by the end of the year, without you working more hours to get there.
That's what I built in my own businesses, and it's what we build in yours. You keep the authority. You own the systems. AI does the work and the loop makes it better.
The free call is 30 minutes. You'll leave knowing exactly what AI should take over first in your business and what it should be worth to you.
Free, 30 minutes, phone or Google Meet. No pitch deck. Only a small number of businesses are taken on at a time.