contact

(Questions)

The questions worth asking before you hire us

Portillo Technologies is an AI consulting, organization-optimization, and custom software practice in Orange County, California. We work out how a company should run, then build the software that runs it. A team of creative and systems engineers, led by principal Jason Portillo, maps the workflow and writes the code. Builds cost 10 to 20 percent of the annual value they create. Work is remote-friendly across the US.

Twenty-one of them, answered straight. Prices included, because a price is the fastest way for both of us to find out whether there is anything here.

(By the numbers)

How often does enterprise AI actually reach production?

Not often, which is the backdrop for every answer below.

95%of enterprise AI pilots deliver no measurable P&L return (MIT NANDA, 2025)
88%of enterprise AI pilots never reach production (Institute of Project Management)
85%of AI project failures trace back to data quality (Gartner)
67%ship rate when the work is bought from a specialized vendor, about twice the internal build rate (MIT NANDA)

Those four numbers explain the shape of this practice. Diagnose first, write down what working means, then build once.

(Who and where)

Who is Portillo Technologies, and where does it work?

What is Portillo Technologies?

Portillo Technologies is an AI consulting, organization-optimization, and custom software practice in Orange County, California. We work out how a company should run, then build the software that runs it. In practice that is two halves. The first is diagnosis: which workflow is bleeding the most money, and what stopping the bleed is worth in dollars per year. The second is the build. Both halves are done by our team of creative and systems engineers. There is no sales layer in front of the work and no separate delivery organization it gets thrown to. The engineers who map the workflow are the ones who write the code.

Where is Portillo Technologies based?

Portillo Technologies is based in Orange County, California. There is no storefront and no published street address, because this is a service-area business rather than a place you visit. Irvine, Newport Beach, Costa Mesa, Santa Ana, Anaheim, Tustin and Huntington Beach are all a short drive, so on-site days are easy to arrange. Los Angeles works too. Everywhere else in the US is remote-friendly and has been from day one. The only contact address is jason@portillotech.com. There is no phone number, which is deliberate.

Who is Jason Portillo?

Jason Portillo is the principal of Portillo Technologies. He sits in the discovery interviews and stays on the account from the first conversation through the build. The work is done by our team of creative and systems engineers, and the ones who map your workflow are the ones who write the code against the acceptance test. The practice works across industries rather than picking one vertical, because operational waste tends to rhyme from company to company even when the jargon changes. He lives in Orange County and the team travels for on-site work when a project earns it. Email the address on this site and it reaches the people doing the work. There is no account manager to route you through.

What does Portillo Technologies do?

Portillo Technologies does two things that fit together. The first is working out how a company should run: which process is costing the most, what a fix is worth per year, and what finished would have to look like in writing. That is the AI consulting side. The second is building the software that does it, which is the custom software side. A build might be a line-of-business web app, an LLM integration, a retrieval system over your own documents, or an ongoing automation retainer. Same team on both halves, which is the point.

Do you work with companies outside Orange County?

Yes. Portillo Technologies works remote by default and always has. Discovery interviews happen over video and the deliverables are written documents. The build is ordinary software collaboration. Orange County and greater Los Angeles get on-site days when watching the actual work beats hearing a description of it, which during discovery is often. Anywhere else in the US, an on-site visit is rare and priced separately. US time zones are all workable. Outside the US, ask first, because contracting and data residency get complicated enough to matter.

(What this costs)

What does this cost?

How much does an AI consultant cost?

Portillo Technologies does not bill hourly, so the answer is a price list. Discovery is $2,500. AI consulting retainers start at $2,500 a month. An automation retainer runs $2,000 to $15,000 a month. A first LLM strategy or pilot is $5,000 to $25,000. For comparison, Groovyweb puts a loaded in-house AI team at $1 million to $1.8 million in the first year, with a three to six month ramp before anything ships. Paying a consultant to find out whether the problem is worth solving is the cheap version of that question.

What does a discovery cost and what does it produce?

Portillo Technologies charges $2,500 for a discovery, about two weeks. Discovery produces two documents. One is the quantified annual dollar value of your best workflow, built from counted hours and error rates rather than a guess. The other is the written acceptance test for that workflow, meaning the specific conditions software would have to meet for you to call it done. The guarantee is attached to it: If discovery doesn't surface at least 10x the fee in quantified annual value, you don't pay. Either way the documents are yours, and you can take them to any other developer.

How are builds priced?

Portillo Technologies prices builds at 10 to 20 percent of the annual value discovery measured, never by the hour. Hourly billing pays a builder to be slow and ties your price to our typing speed instead of your return. The bands, so you can budget: a custom web app or line-of-business system runs $15,000 to $75,000 for an MVP and $250,000 and up when it is genuinely complex. AI or LLM integration is $5,000 to $25,000 for a strategy or pilot and $25,000 to $100,000 mid-size. A RAG or knowledge copilot is $8,000 to $45,000 to build, plus $650 to $1,750 a month to run. That is custom software development in Orange County.

Who owns the code you write?

You do, outright. Portillo Technologies never resells the code, never licenses it, and never rents it to the next company in your industry. You get the repository, the deployment configuration and the documentation, and there is no platform of ours you have to keep paying to run what you bought. If you want to hand it to another developer next year, that is a normal Tuesday rather than a negotiation. The only thing we keep is generic knowledge, the kind any developer carries from one job to the next. Lock-in is a business model, and it is not this one.

What is the guarantee?

Portillo Technologies stands behind two. On discovery: If discovery doesn't surface at least 10x the fee in quantified annual value, you don't pay. On builds: we work it until it passes your acceptance test. Both only work because the acceptance test gets written down before any code exists, in your language, at the end of discovery. A guarantee against a vague ambition is theater. A guarantee against a document you signed off on is an actual obligation, which is why the acceptance test matters more than anything in the contract.

(Why trust it)

Why believe the work will land?

Why do most AI pilots fail?

Most AI pilots fail because nobody wrote down what done means. That is the read Portillo Technologies starts from. MIT's NANDA report found that 95% of enterprise AI pilots deliver no measurable P&L return, and the Institute of Project Management puts the share that never reach production at 88%. Gartner attributes 85% of AI project failures to data quality. IDC found 65% of teams name data readiness as the blocker. None of that is a model problem. It is scoping and data. The longer version, plus the two changes that usually unstick a stalled pilot, is in the AI Pilot Autopsy.

What is an acceptance test and why does it matter?

A written description of what the software must do for you to call it finished, agreed before any code exists. Not something vague like improve intake efficiency. Closer to: a new intake is complete and in the system within four minutes, with no manual re-keying, on 95 of 100 consecutive real cases. It matters for a dull reason. Pertama Partners found that 61% of AI projects approved on projected ROI were never measured after launch. If nobody writes the target down, nobody can check it, and the project quietly becomes impossible to fail.

How long does a project take?

Discovery is about two weeks. A build depends on the band. An MVP in the $15,000 to $75,000 range is usually weeks rather than months. A $250,000-plus system is months, and anyone promising otherwise is selling. Retainers are ongoing by definition. The real variable is how fast your side answers questions and grants access to the systems involved, because a build stalls on a missing API credential far more often than on hard engineering. That is not a complaint. It is a scheduling fact worth knowing before the calendar gets set.

How many projects do you take at once?

A deliberately small number. The team takes on few enough engagements that senior engineers stay on each one from discovery through delivery, rather than selling capacity we then have to go and staff. Discovery engagements and advisory retainers run alongside builds because they use different hours. That restraint is why the build guarantee is affordable to offer, and why nothing gets quietly handed to a junior you never met. It also makes the calendar a real constraint. If the book is full when you call, you will get the actual start date instead of a comfortable one.

(What gets built)

What gets built, and on what?

What technology do you build on?

Cloudflare Workers and D1 for most applications, because the running cost stays low and there is no server to babysit. Plain web front ends, no framework religion. Postgres when the data is genuinely relational. For AI work, whichever model family fits the task. An evaluation harness sits in front of it, so swapping models is a configuration change rather than a rewrite. Retrieval runs over your documents, not somebody else's index. Stack choices follow the acceptance test, and if your team has to maintain the thing after we leave, that outranks our preferences.

Do you do websites?

Portillo builds the application, not the brochure. If what you need is an informational marketing site, that work belongs to a sibling company and you will get pointed there rather than upsold here. What gets built here is the software behind the interface: the client portal, the quoting engine, the scheduling system, the internal dashboard nobody sells off the shelf because it only makes sense inside your company. Those have web front ends, so they look like websites from the outside. The value sits in the logic and the data model.

(The alternatives)

What are the alternatives to hiring a specialized builder?

Four ways companies get AI and internal software built, with what each one actually costs and what the published research says happens next.

PathWhat it costsWhat the evidence says
Build an in-house AI team $1M to $1.8M loaded in year one Groovyweb puts the ramp at three to six months before the first thing ships
Hire an offshore shop The headline rate, plus 25% to 150% in hidden cost (SmartDev) SmartDev finds QA rework eats 15% to 26% of billed hours
Use a no-code platform Low subscription, low up front BrightByte finds 25% to 30% of no-code builds are rewritten in custom code within two years
Buy from a specialized builder $2,500 discovery, then 10% to 20% of measured annual value MIT's NANDA report found buying from a specialized vendor ships a working AI system about 67% of the time, roughly twice the internal rate

None of those numbers are about Portillo. They are about the category, which is the only honest thing a practice with no published case studies can hand you.

(No products)

Does Portillo Technologies sell its own products?

Does Portillo Technologies sell its own products?

No. Portillo Technologies builds custom technology and AI tools for other companies, upgrades the systems they already run, and consults on how the work should be organized before any of it gets built. Everything it makes is shaped to one client's workflow and owned by that client. There is no Portillo product line to be sold, upsold, or migrated onto.

(Step by step)

How does an engagement run, step by step?

  1. You send one email describing the workflow that costs you the most. Two or three sentences is plenty.

  2. You get the AI Pilot Autopsy back, free, as a PDF attached to the reply. Two pages on why that specific pilot stalled and the two changes that unstick it. No form, no signup, and no follow-up sequence if you never write back.

  3. If the thinking holds up, you book discovery. $2,500, about two weeks, ending in the quantified annual value of your best workflow and the written acceptance test for it.

  4. You decide with both documents in hand. They are yours to take anywhere, including to a different developer. If discovery doesn't surface at least 10x the fee in quantified annual value, you don't pay.

  5. If we build, the price is 10% to 20% of that measured annual value, agreed up front, never hourly.

  6. The build is not finished until it passes the acceptance test you wrote. The team runs few enough builds at once that your start date is a real date rather than a hope.

(Is this a fit)

How do we decide whether this is a fit?

Do you have case studies?

No. None are published. There are no customer logos, no testimonials, and no benchmarks from Portillo's own work, and inventing them would be a strange way to open a relationship built on measurement. Two things stand in. The AI Pilot Autopsy is free and specific to your company, so you can judge the quality of the thinking before any money changes hands. And the acceptance test means the proof gets written before the build instead of after: you define what working means, and the build is not done until it passes. That is the harder promise, not the easier one.

How does this start?

Email Portillo Technologies at jason@portillotech.com, or use the form on the homepage. Describe the workflow that annoys you most in two or three sentences. You will get the AI Pilot Autopsy back as a PDF attached to a reply: two pages on why that specific pilot stalled and the two changes that unstick it. It is free. No form to fill in, nothing to sign up for. If you do not reply, you will not be chased. That last part is a real promise and the reason there is no follow-up sequence behind this page.

(Not a fit)

Who is Portillo Technologies not for?

Companies looking for a body to bill hours against. Anyone who needs a large contract team stood up next month, since the practice stays deliberately small and senior. Anyone whose actual goal is a press release about using AI rather than a number that moves. Teams that cannot get access to their own data, which is the Gartner 85% failure in its natural habitat. Anyone who wants the vendor to also define success, because an acceptance test only works when you own it. Buyers who need a logo wall before the first conversation should hire a firm that has one. The longer list is at Not Us.

(Sources)

Where these numbers come from
  1. MIT NANDA report, 2025, on pilot-to-production rates and vendor versus internal build outcomes. softwareseni.com
  2. Institute of Project Management, on the 88% of enterprise AI pilots that never reach production. institutepm.com
  3. Pertama Partners, AI project failure statistics 2026, covering the Gartner and IDC data-quality figures and post-launch ROI measurement. pertamapartners.com
  4. BrightByte, on how often low-code and no-code builds get rewritten in custom code. thebrightbyte.com
  5. SmartDev, offshore software development cost factors and QA rework. smartdev.com
  6. Groovyweb, in-house versus outsourced AI development costs, 2026. groovyweb.co

(One more question)

Still have a question this page missed?

Send it. Describe the workflow that costs you the most in the same email and the free two-page autopsy comes back with the answer. Nothing to sign up for, and no chasing if you go quiet.

Get the pilot autopsy

Email Jason