Portillo Technologies is the wrong choice when you need a large team, a brand-name signature, or software that already exists off the shelf. It is one builder in Orange County, California. One build at a time, and that cap is the model, not a growth stage. If the problem is bigger than one workflow, hire someone else.
Who should hire a large consultancy instead?
Three situations, and they're real. The first is a multi-year program running across several countries, where most of the difficulty is coordination rather than engineering. Second: a board or a regulator that wants a name it already trusts on the cover of the report. That isn't vanity, it's risk management, and it's worth paying for. The third is a rollout that needs forty people staffed by Monday.
None of those is a one-builder job, and pretending otherwise costs you a year. Call a Big-4 firm or a regional shop with a real bench.
What the brand name doesn't buy is a working system. MIT's NANDA report found that 95% of enterprise AI pilots deliver no measurable P&L return, and those pilots were not mostly run by amateurs. Ask the big firm the question you'd ask us: who writes the acceptance test, and what happens if the thing fails it?
Who should use off-the-shelf software instead?
If a product already does the job, buy the product. A CRM, a decent PSA, a scheduling tool, a document chatbot for a small library of PDFs. Configuring something that exists beats building something that doesn't, and a $99-a-month subscription beats a $40,000 MVP every time it fits.
No-code sits in the same bucket. It puts a working tool in front of real users in about a week. The Bright Byte reports that 25% to 30% of no-code builds get rewritten in custom code within two years, which also means roughly three quarters of them never do. Start there when the process is simple, the volume is low, and nobody's job depends on it running at 2 a.m.
The signal to build custom is the workaround spreadsheet. When one appears next to the tool, the tool stopped fitting.
Who should hire in-house instead?
If AI is going to be a permanent, central function rather than a project, hire for it. Own the roadmap, own the people, own the retraining cycle three years out. Nobody should rent that.
Know the bill first. Groovy Web puts an in-house AI team at $1M to $1.8M loaded in year one, with a three-to-six-month ramp before the first useful output. That's reasonable when AI is your product. It's a lot of money for one jammed workflow in accounts payable.
There's a middle answer people skip. Hire the team, and have someone outside build the first system while they ramp. MIT's NANDA report found that buying from a specialized vendor ships a working AI system about 67% of the time, roughly twice the rate of building it internally. You own the code either way, so the first build becomes the new team's starting point instead of their first six months.
What work does Portillo turn down?
This isn't a preference list. These get declined.
- Cheapest-bid RFPs. If the deciding number is price per hour, someone else wins, and they should.
- Staff augmentation and body-shop arrangements. Renting a developer by the month to work someone else's ticket queue is a different business.
- Projects with no executive owner. If nobody's P&L moves when it works, it won't survive contact with the org chart.
- Work whose scope can't shrink to one workflow. "Modernize the company" is not a first engagement.
- Anything that needs ten people to start Monday. One builder, one build at a time.
- Anything behind a sign-off chain we can't reach. If the person who can approve the acceptance test won't be in the room, the build stalls in month two and both of us waste the budget.
- Marketing-brochure websites with no application behind them. Plenty of shops do that work well. We aren't one of them.
Which option fits which situation?
| Your situation | The right choice | Why |
|---|---|---|
| A product already does this well | Buy the tool | Configuration beats construction. A subscription is cheaper than an MVP and it ships today. |
| Simple process, low volume, nothing critical | No-code | Live in a week. The Bright Byte puts rewrites at 25% to 30% within two years, so most hold. |
| Multi-country program, 40 people, board wants a known name | Large consultancy | Coordination at that scale is its own discipline, and it isn't ours. |
| AI is becoming a permanent, central function | Hire in-house | $1M to $1.8M in year one (Groovy Web) buys an asset you keep and a roadmap you control. |
| Hourly rate is what decides the purchase | Offshore team | It can work. SmartDev puts total cost of ownership 25% to 150% over the headline rate, with QA rework at 15% to 26% of hours. |
| One stalled pilot or one jammed workflow, with a P&L owner who wants it fixed | Us | One workflow, a written acceptance test, and the person who maps the workflow writes the code. |
How do you work out which one you are?
- Name one workflow, not a department. If you can't get down to one, you want a consultancy, or you want to wait.
- Spend an honest hour looking for a product that already does it. If you find one, buy it and stop reading.
- Find the person whose P&L moves when it works. If that person doesn't exist, neither does the project.
- Write down what "working" means, in one sentence specific enough to test. If nobody will commit to that sentence, no vendor on earth can pass it.
- Put a number on the workflow. If a year of it is worth less than $50k, a build priced at 10% to 20% of that value isn't worth either of our time.
So when is it the right fit?
Four conditions. One stalled pilot or one jammed workflow, small enough to say in a sentence. A P&L owner who will sit in the room and name the number. A willingness to shrink the scope until it's one thing. And a willingness to sign the acceptance test, which is the written definition of working that the build has to pass. Builds are priced at 10% to 20% of the measured annual value, never by the hour, and we work it until it passes your acceptance test.
The free way in is the AI Pilot Autopsy: two pages on why your pilot stalled and the two changes that unstick it, emailed back as a PDF. No form, no signup, no sequence. Ignore the reply and nothing chases you.
The paid way in is discovery. $2,500, about two weeks, and it produces the quantified annual dollar value of your best workflow plus the acceptance test for it. If discovery doesn't surface at least 10x the fee in quantified annual value, you don't pay. The documents are yours to take anywhere.
Common questions about picking a firm
Should I hire a boutique consultancy or a Big-4 firm for an AI project?
Size the firm to the program, not to the logo. A multi-country, multi-year rollout needs a bench, a project management layer, and a name your board already trusts. One stalled pilot or one jammed workflow needs a builder who can start this month. Neither firm can promise a result: MIT's NANDA report found 95% of enterprise AI pilots deliver no measurable P&L return, and big budgets are well represented in that number.
Is a one-person firm too risky for a business-critical build?
It is a real risk and you should price it in. The mitigation is that you own the code outright, with no lock-in and nothing resold, so another developer can pick it up. There is also a hard cap: one build at a time. If the timeline can't survive that cap, hire a shop with a bench.
Should I hire an offshore development team instead?
If the headline hourly rate is what decides the purchase, yes, and it can work. Budget past the rate. SmartDev puts the total cost of ownership on offshore projects at 25% to 150% over the headline rate. QA rework eats 15% to 26% of the hours. The gap is usually specification, not skill.
What happens if you turn my project down?
You get told why, in an email, along with the kind of firm that fits it better. You can still ask for the AI Pilot Autopsy. It is free, it is not gated behind a form, and there is no follow-up sequence attached to it. If you don't reply, you don't get chased.
Can I take the discovery documents to another builder?
Yes. Discovery costs $2,500 and runs about two weeks. It produces the quantified annual dollar value of your best workflow and the written acceptance test for it, and both are yours to take anywhere. If discovery doesn't surface at least 10x the fee in quantified annual value, you don't pay.
Sources
- MIT NANDA report, 2025, on enterprise AI pilots delivering no measurable P&L return and on vendor-built systems shipping at roughly twice the internal rate. softwareseni.com
- The Bright Byte on low-code and no-code rewrite rates. thebrightbyte.com
- SmartDev on offshore development cost of ownership and QA rework. smartdev.com
- Groovy Web on the loaded first-year cost and ramp of an in-house AI team. groovyweb.co
Still think it's a fit?
Send the pilot that stalled, or the workflow that keeps jamming. If it belongs somewhere else, you'll be told where.