AI systems for operators who don't have time for hype.

Built and run by a vending operator who uses his own AI-driven software every working day, not by a consultant who has never shipped one. If it doesn't hold up under real operating conditions, it doesn't leave the workshop.

Start a conversation or see what runs in production

Approach

Three engagements. One rule for choosing.

Start with the smallest step that produces a real, usable answer. Not a slide deck, not a roadmap for a roadmap. The right starting point depends on how much is already known about where AI can help.

Assessment

A fixed-scope look at one operation to find where AI genuinely saves time or money, and where it doesn't. You get a clear-eyed answer before committing to anything larger.

You leave with: A written answer: what to automate, what to leave alone, what it would take, and in what order.
Fits when: You suspect there's waste in manual data work, but nobody has mapped where it actually is.
Commitment: Fixed scope and fixed price, quoted after the first call.

Embedded implementation

Ongoing, hands-on work building and maintaining the systems an assessment points to. This is where things get built, tested, and put into daily use by the people who do the job.

You leave with: Working software your team uses, with monitoring, documentation, and a way to undo any change.
Fits when: The problem is clear and you want it solved by someone who will still be around when it breaks at six in the morning.
Commitment: Monthly, ongoing. Stops when the work is done.

Full-scale transformation

A larger, project-based engagement to rebuild a core process end to end. For organizations ready to commit to a structural change, not a patch.

You leave with: A process that runs on a system instead of on one person's memory.
Fits when: Leadership has already decided the process has to change and wants it rebuilt by someone who has done it inside his own business.
Commitment: Project-based, scoped in writing before work starts.

Proof

OVM, the operating system behind On Point Amenities.

Before this was a consulting offer, it was a problem in my own vending route. I built OVM to run it, and it has run the business every day since. Here is what it does, in the words I'd use with another operator.

Inventory: Every unit in and every unit out is recorded as an immutable event. Stock on hand is computed from those events, never edited by hand, so a bad import gets replayed instead of argued about.
Field work: A mobile app for technicians: scan a barcode, record a restock, capture expiration dates with the camera, and keep working when the signal drops.
Suppliers: Distributor invoices and portal exports are parsed into one catalog. An earlier version used a language model to read invoices. A deterministic parser proved more accurate and cheaper, so the model was retired from that job.
Machines: Sales and stock levels pull automatically from the machine vendors' platforms, whether or not those platforms offer a clean API.
Tax: Colorado sales tax depends on the jurisdiction of each machine. OVM looks up the rate for every machine's address, classifies each product's taxability, and keeps the filing calendar.
AI, where it earns its place: Language models normalize inconsistent supplier product names to the canonical catalog. Every AI feature is checked against operator-confirmed test cases in the build pipeline before it ships.
Reliability: Error monitoring, daily health checks, and automated alerts for expiring stock and supply gaps. More than five thousand automated tests run before anything reaches production.

It runs on real equipment under real operating pressure, not in a demo environment. That is the standard client work is held to.

Who's behind it

Justin Krakow.

Veteran, and owner-operator of On Point Amenities, a smart-vending route in the Denver metro. I built OVM because the vending business needed it and nothing off the shelf handled inventory, supplier data, and Colorado tax together. On Point AI Consulting is that same work, done for other operators.

There is no team to introduce. The person you email is the person who does the work.

How I work

Rules I hold my own systems to. Client work gets the same ones.

Reversible before correct before fast. Every change ships with a way to undo it. Being able to roll back matters more than being right the first time.
Hypothesis before action. I write down what I think is true, then test it against your data before building on it.
Plain code where it can be, AI where it must be. If deterministic code does the job, it does the job. Models are for the parts that need judgment.
Nothing ships untested. Behavior is proven against cases you have confirmed are right, and those cases run every time the code changes.
Straight answers. If AI is the wrong tool for your problem, the assessment says so, and you have spent a fixed amount finding out.

Who it's for

A good fit usually looks like

High-volume, repetitive data work done by hand, in spreadsheets, or by re-typing between systems.
Decisions that depend on judgment. The data needs to support a person, not replace them.
Real operational or compliance risk if the numbers are wrong: tax, inventory, invoices, payroll.
Leadership that wants a straight answer before it wants a bigger project.

Probably not a fit

You want a chatbot on the website and nothing behind it.
You need a large team on site next week.
You already have an internal AI team and want a second one.

Questions

The ones that come up on the first call.

Do you guarantee results?

No. I scope the smallest step that produces a usable answer, and I tell you when AI is the wrong tool. An assessment that says "leave this alone" is a successful assessment.

Do we need to be using AI already?

No. Most of the value in an assessment is in the data and the process, not the model. Plenty of good outcomes involve no model at all.

Which AI vendors do you use?

Whatever fits the job. I have run both Claude and Gemini in production, and I replaced a language model with plain code when the plain code was more accurate and cheaper.

Do you build it, or tell us what to build?

Both. An assessment ends in a written answer you can act on with anyone. Embedded implementation is me building it alongside your team.

Do you work remotely?

Yes. In person in the Denver metro, remote everywhere else. Most of the work is in your systems and your data, not in a conference room.

What happens to our data?

It stays in your systems. Nothing is sent to a model without you knowing what is sent and why, and nothing is kept after the engagement unless you ask.

What does it cost?

An assessment is fixed scope and fixed price, quoted after the first call. Implementation is billed monthly and stops when the work is done. Transformation is quoted per project, in writing, before work starts.

Contact

Tell me what's slowing you down.

The first conversation is about understanding the problem, not selling a package. If it's a fit, we'll both know quickly. If it isn't, I'll say so.

What to put in the email

The work. What it is and who does it today.
The pain. Where it goes wrong, takes too long, or costs too much.
The systems. Where the data lives now: spreadsheets, a vendor portal, an app, paper.

What happens next

I read it and reply within two business days.
We talk for thirty minutes about the problem. No pitch, no deck.
If it's a fit, you get a written scope for an assessment. If it isn't, I'll say so and point you somewhere useful.

Email info@opconsulting.ai Denver metro. Remote welcome. Veteran-owned.