Productize the service: turn bespoke AI work into one thing you can sell twice

The Ledger · Selling

Every bespoke sale starts at zero. A scope call, a quote you build from a blank page, a job you architect from scratch, a margin you will not know until it is over, and then next month you do all of it again for the next client. The tenth project costs you the same effort as the first, which means the only way to earn more is to work more hours, and there are only so many of those. This is the ceiling every AI freelancer hits, and AI is quietly making it lower.

A productized service breaks the loop. It is one named deliverable, at one public price, delivered the same way every time. No discovery call, no custom quote, no reinventing the job. You built the pricing and learned to scope it and made it recur; this is the next move, repackaging what you sell so a buyer can look at one line on a menu, see one price, and buy without a meeting. This piece is how to do it, and it is less a pricing trick than an operations one.

The short version #

If you are packaging the offer this month, the method in six moves: Pick the service whose demand is already uniform. A niche where every client needs the same thing can be templated; one where every client needs something different cannot, at any price.Cut down to one bounded outcome at one public number.Designjoysells a single $4,995-a-month tier, run by one person, and rode it to roughly $1.7M ARR; the believable solo ceiling for the AI version sits around$8,000 a month, self-reported.Lock the scope so the price survives. A written feature list where extra requests move the timeline, and for nondeterministic AI work afixed budget with controlled scopeinstead of a promised feature list.Decide who pays the token bill before you publish the price. Buffer it at two to five times the estimate, add a passthrough clause, or put the client’s own API keys in.Productize the judgment, never the raw build. The deliverable is commoditizing; what holds its price is knowing which automation moves the P&L, and the call you make the day the model gets it wrong.Write the SOP so it repeats. A structured intake form, an async board, and a delivery checklist, validated on one discounted client before you scale it.

Everything below is the evidence, the inflated agency numbers included.

The hours ceiling, and why AI makes it worse #

Hourly billing has a paradox built into it: the better you get, the less you earn. A logo that took twenty hours in your first year takes five by your fifth, and billing by time that is a roughly 75 percent pay cut for the crime of getting good at your job. Expertise makes the work faster, and hourly pricing punishes exactly that, as Jonathan Stark has argued for years.

AI turns the screw harder. A build that took ten weeks now takes six, so your cost base to deliver it drops 30 to 40 percent. If you bill for your time, you just handed that entire gain to the client and cut your own invoice. If you priced the outcome, you keep it. AI is the biggest efficiency jump most of this work has ever seen, and time-billing converts every bit of it into a discount you give away. The escape is to stop selling hours at all.

What a productized service actually is #

Set the definition precisely, because most people hear “productized” and think “raise my prices.” That is not it.

A productized service is one bounded outcome, one public price, one repeatable delivery pipeline, sold identically to every buyer. The textbook case is Designjoy: one flat tier at $4,995 a month, one design request at a time, roughly 48-hour turnaround, or cancel anytime. Its own site says it plainly: run entirely by one person, no other designers, no outsourcing. Brett Williams took that single standardized offer past a million in ARR as an agency of one, and by 2024 was reportedly around $1.7 million, still solo, though the exact MRR wanders by source so read that as an as-of-2024 range rather than a live number.

The mechanic is the point. Here are the operators who prove it repeats:

Every one is a founder’s own disclosure on Indie Hackers or a roundup, unaudited, so treat them as claims. But they rhyme, and they rhyme on the same note: the same offer, sold again and again, with the delivery cost falling each time.

Brian Casel put the diagnosis in one line. Billing hourly, he said, means freelancers reinvent what they do every single time. Productizing removes the reinvention.

And that is the real lens on this, which is where my own background comes in. I have not run a productized freelancing shop, so I will not pretend to. What I bring is years as an operations analyst and program manager, shaped by lean thinking, and from that seat a productized service is standard work: the best-known way to do a job, written down so it runs the same every time, so nobody has to rebuild it from scratch for each instance.

A bespoke service is a warehouse where every order is picked a different way. A productized service is one where the pick path is documented and identical. The founder numbers above are just what happens when you stop paying the waste tax of doing it fresh each time.

Which of your services to productize #

You do not productize your whole business. You pick one thing.

Pick the one whose needs are already standardized, so you can template instead of reinvent. Casel chose restaurant websites for exactly this reason: every restaurant needs to showcase a menu and show its hours and location. The demand is uniform, so the delivery can be. The opposite of that, a niche where every client genuinely needs something different, cannot be productized no matter how you price it.

Then eliminate, do not add. Tyler Gillespie cut whole service lines, funnels, websites, design, down to just the most profitable and lowest-headache work before he scaled what was left. The instinct to offer more is the instinct that keeps you bespoke. ManyRequests offers three questions to pressure-test a candidate offer: can you state it in one sentence, are people already paying for it, and can you name a place where those buyers gather. If any answer is no, it is not ready to productize yet. And there is a margin prize for getting the niche right: specialist agencies report gross margins in the 40 to 75 percent range against 15 to 20 for generalists, a founder-surveyed figure so directional, but the gap shows up across sources.

Lock the scope so the fixed price survives #

A fixed price only works if the scope is genuinely fixed, and this is where most first attempts leak. The fix is to write down both what is included and what is excluded, and to have a real answer for the outlier. When a restaurant client wanted e-commerce, Casel did not build it into the package for a tiny subset of buyers; he offered a PayPal-button workaround and kept the offer standard. Designpop holds the line with a locked feature list agreed at the start: more requests push the timeline, never the price, so extra scope can never eat the margin.

AI adds a wrinkle here that ordinary services do not have. The output is nondeterministic. A logo is done when it is done, but a RAG system hitting an accuracy target might take one iteration or five, and you cannot know in advance. For those jobs a rigidly fixed scope is dangerous, because a rigid contract has no way to incorporate learning, as Atomic Object puts it.

The answer is a fixed budget with controlled scope rather than a fixed feature list: the client buys a bounded outcome and a bounded spend, and you keep the freedom to reach it the way the work actually demands. Everything about change orders and deposits still applies, and I laid that out in the scoping piece; here the only job is to draw the scope boundary that makes one price possible.

Price it once, in public, and handle the token bill #

The productized move on price is to make it one visible number on a page; assembling a figure per client is the agency habit you are leaving. What to charge is the pricing pieces’ job, and the psychology behind the number is its own piece; the productizing job is only to make it a catalog price. As a market anchor, cross-source rate cards for productized AI work land around $500 to $2,000 for an audit, $2,500 to $15,000 for a fixed build, and $500 to $5,000 a month for a retainer, though those are vendor benchmarks with no disclosed revenue behind them, so hold them loosely.

The lesson worth stealing is Designjoy’s. The same standardized offer went from roughly $449 to $849 a month at its 2017 launch to about $5,000, a rough tenfold rise, and Williams got there by raising the price against a waitlist. A standardized offer with demand behind it lets you move the number without moving your hours, which is the whole prize.

One cost is genuinely new to AI work: the token bill. If your fixed price includes model usage, a heavy client can quietly erase your margin. Three ways to handle it, from the AI-agency price cards: estimate the real token cost and multiply it by two to five as a buffer baked into the price, add an explicit model-cost passthrough or re-pricing clause, or simplest of all, have the client put their own API and hosting keys in and bill it direct so the variability never touches your fee. Pick one before you publish the price, because the first month’s usage will otherwise pick for you.

Productize the judgment #

This is the part that decides whether the offer survives the next two years, so make it the center.

AI is actively commoditizing the raw deliverable. Research synthesis, first drafts, a single-agent chatbot: those are getting cheap fast, and the margin compression is most evident in narrowly defined tools and lightweight workflow utilities. If the thing you productize is the task itself, you are pricing something your client can increasingly do without you. The whole market is repricing around this: per-seat is no longer the atomic unit of software, and outcome pricing is showing up in production, Intercom’s Fin at $0.99 per resolution, Decagon charging per resolution rather than per seat.

So do not productize the build. Productize the judgment the client cannot buy off a shelf: which automation actually moves their P&L, whether the chatbot is genuinely accurate enough to put in front of customers, what to do when the model is confidently wrong. That judgment is the product. The mechanic that carries it is the subscription: one piece of judgment you have systematized, packaged as a productized agent that ten clients each subscribe to at $500 a month, is $60,000 a year that grows with the client list rather than resetting each month, which is the same recurring layer the retainer ladder is built on.

Document it so it repeats #

The last step is what makes it standard work rather than a good month. Systematize the intake: replace the discovery call with a structured form, and replace the meetings and Slack with a dashboard, the way Conversion Factory runs entirely async through a Notion board. Then write the SOP before you hire anyone, including before you hire the future version of yourself who has forgotten how you did it: the outcome, the step-by-step delivery, the first-48-hours onboarding, how a request gets handled, and the quality bar it has to clear. Validate the offer with one discounted client from your niche before you scale it, and only then start delegating, gradually, beginning with the parts that are already written down.

And keep the ceiling in view, because the loudest numbers online are a trap. The people posting $40,000-a-month AI-agency figures are frequently selling a course on how to make $40,000 a month, and that course is the actual income; Nick Saraev’s automation agency did real revenue, but his headline earnings come from teaching, and figures like Liam Ottley’s are not independently verified. Discount them. That is just reading the incentive.

The grounded number for a solo productized AI service is closer to the $8,000 a month an n8n freelancer reported at around 25 hours a week, and even that took getting known: the single biggest thing that decides whether a productized service sells is whether people already trust you enough to buy without the meeting. Which is the same thing the whole Ledger keeps landing on. The offer is the easy part. Fix the delivery so it repeats, put one price on it, and let being good at the job stop being the reason you earn less.

One email, when there's something worth sending

Get the research in your inbox. #

No fixed schedule, no filler. You get an email when we've tested something, run the numbers, or found a tool worth your time.

Free. Double opt-in, unsubscribe in one click.

Priced something with this? Compare notes in the forum ↗

Sources #

indiehackers.com ↗indiehackers.com ↗starterstory.com ↗indiehackers.com ↗indiehackers.com ↗manyrequests.com ↗mixergy.com ↗parakeeto.com ↗consultingquest.com ↗a16z.com ↗bvp.com ↗pickaxe.co ↗spin.atomicobject.com ↗amadeuscapital.com ↗memtime.com ↗jonathanstark.com ↗medium.com ↗Show your work: if a number can't be shown, we don't print it.