Productized AI Services: Sell the Same Thing Twice

Productized AI services means selling a fixed scope at a fixed price with a fixed delivery process, instead of quoting each client from scratch. The offer stops being "we can help with AI" and becomes "we build your support chatbot from your existing help docs, three weeks, 6,000." Same work, different shape, and the difference shows up in your margin rather than your rate card.

The trap is productizing too early. An offer built from one client's project is not a product, it is that client's project with a price on the front, and the second buyer will break it in week one.

The test: three deliveries, same first two weeks

Before you write a package page, look back at your last few projects and ask a specific question. Across three separate clients, was the first two weeks of work substantially the same?

Not the outcome. The work. If for three different businesses you spent the opening fortnight collecting source documents, cleaning them, assembling a test set of real questions, and standing up the same retrieval setup, that is a product. If one needed a CRM integration, one needed a data migration and one needed you to write their help centre from nothing, you have three custom projects that happen to involve AI.

Two more checks worth running:

• Can you name the buyer in one sentence? "Ecommerce shops doing 1 to 10 million a year with an existing help centre" is a buyer. "Small businesses" is a hope.

• Can you predict the hours within about 20 percent? If your estimates across those three projects ranged from 30 to 120 hours, the variance is the product. Find out what caused it before fixing a price.

Most people fail the second check and productize anyway. Then they eat the difference.

Pick the narrowest thing that still has a budget

The instinct is to package the biggest service you offer. The better move is to package the one with the tightest variance, even if it is smaller, because a predictable 4,000 offer you can deliver twelve times beats a 15,000 offer that goes over budget on half the deals.

A few shapes that productize cleanly, roughly in order of how contained they are:

The middle column is the useful one. It tells you what to qualify for on the sales call, and it tells you what your assumptions section has to cover. An audit sits at the bottom of the price range and the top of the predictability range, which is why it makes such a good entry offer even though nobody gets rich selling audits.

Turn the variance into qualification, not into scope

Once you know what varies, you stop absorbing it and start screening for it. Every productized offer needs two or three disqualifying questions asked before you quote:

For a support chatbot: does a help centre already exist, roughly how many articles, and when was it last updated? A client with 40 current articles is a three-week delivery. A client with 12 articles written in 2021 is a content project wearing a chatbot costume, and either you scope that separately or you pass.

For document extraction: send us ten representative files. Not a description of the files. The files. Formats that look uniform in conversation turn out to include four templates and a scanned fax.

This is the real work of productizing. You are not simplifying the delivery, you are moving the complexity from the middle of the project to the front of the sale, where it costs you a phone call instead of three unpaid weeks.

Price it once, then leave it alone for ten deals

Set the price from the value to the buyer rather than your hours, the same logic that applies to pricing an AI product. What changes with a productized offer is that you should then stop adjusting it per client. Custom quoting is the habit you are trying to break, and one exception restores it.

Two things make a fixed price survivable:

A tiered structure with mechanical boundaries. Standard covers up to 50 source articles and one channel. Plus covers up to 200 and two channels. The boundary is a number the client can check themselves, not a judgement call you have to defend.

An explicit exclusions list on the offer page itself, not buried in the contract. Half the value of a productized service is that a prospect can self-select out before you ever speak.

Review the price after ten deliveries, not after each awkward conversation. You need the sample size to know whether you are underpriced or just occasionally unlucky.

What productizing costs you

It is worth being honest about the trade, because the usual write-up on this is one-sided.

You lose the ability to say yes. A client with a 30,000 budget and a slightly different need used to be a good month. Now they are a bad fit for your offer, and taking the work quietly un-productizes you. Some people run a productized offer alongside a custom practice for exactly this reason, which works as long as the custom side does not eat the delivery capacity you promised the product side.

You also take on the maintenance of a process. Templates, a delivery checklist, a standard test set format, a handover document. That is a few days of unbilled work up front and a steady trickle after. It pays back around the fourth or fifth delivery, and not before.

And you make yourself easier to compare. A fixed scope at a fixed price is legible to buyers, which is the point, but legible also means comparable to the next agency's fixed price. Your defence is the qualification questions and the delivery record, not the price.

Where it fits in the broader picture

Productized services sit in the middle of the ladder between hourly work and software. They are more scalable than consulting and far less capital-intensive than building a product, which makes them a sensible next step once client work is reliably coming in. The full ladder, with time to first revenue for each rung, is laid out in AI monetization strategies.

If you are earlier than that and still working out where the first buyers come from, productizing is premature. Finding your first client comes first, and the three deliveries that tell you what to productize come out of that work anyway.

FAQ

How many clients do I need before productizing?

Three deliveries of substantially similar work is the practical minimum. Two can be a coincidence. You are not looking for three identical projects, you are looking for three projects with an identical opening fortnight.

Should I publish the price on my site?

Yes, if the offer is genuinely fixed. A published price is the mechanism that does the qualifying while you sleep, and hiding it reintroduces the discovery call you were trying to eliminate. If you are not confident enough to publish it, the scope is probably not tight enough yet.

Can I productize something I have never delivered?

You can package it, but you are guessing at the variance, which is the number that decides whether the price works. If you want to try, sell the first two at a deliberately low price as pilots, on the explicit understanding that you are learning the delivery, and treat the losses as research.

What is the most common reason productized AI offers fail?

Source data quality. Nearly every one of these offers assumes the client has usable content, records or documents, and a large share of small businesses do not. Build the check into your qualification questions and it becomes a filter. Leave it out and it becomes your problem, at your cost, in week two.

Does a retainer work better than a fixed project?

They do different jobs. The project pays for the build, the retainer pays for the tuning, monitoring and content updates afterwards, which is genuinely ongoing work for anything involving a model. Sell the project first and offer the retainer at handover, when the client has just watched the thing work and is at peak willingness to keep it working.