AI Consulting & Strategy.

Most companies already know they should be doing something with AI. Few know which thing. The ideas arrive from every direction: a chatbot for the website, an agent for the inbox, a copilot for the sales team, a model trained on company data. Each sounds plausible, and most get started without anyone checking whether the data exists, what accuracy is needed, or what it will cost to run.

AI consulting puts an order on that list. We look at how work actually moves through your business, measure where time and money go, and test each idea against three questions: does it remove real work, is the data there to support it, and will it cost less to run than the work it replaces. The output is a short list of projects worth doing, in the order worth doing them.

We are engineers, not a strategy firm. Every recommendation comes with a working estimate because we would be the people building it, and we will happily recommend that you build nothing yet.

Workflow mapping

We interview the people who do the work and trace three to five core processes end to end: where requests arrive, who touches them, which systems they pass through and where they wait.

Data readiness

We check whether the documents, records and history an AI feature would need actually exist, how clean they are, who owns them and what cannot leave your systems.

Use case scoring

Each idea is scored on value, feasibility, data readiness, risk and running cost, so the ranking is visible and you can disagree with it line by line.

Build options

For each shortlisted use case we compare off-the-shelf tools, automation in n8n or Make, an AI feature inside your existing software, and a custom agent, and say which fits.

Risk and governance

Data protection, human review points, failure modes and how accuracy will be measured, using the NIST AI Risk Management Framework as a checklist rather than a bureaucracy.

A costed roadmap

The first project scoped in enough detail to start, with a fixed price, a timeline and the measure of success agreed in advance.

1. 01KickoffA call with the owner of the budget and the people closest to the work. We agree which processes are in scope and what a good result looks like.
2. 02Discovery interviewsShort sessions with the teams who do the work, plus read-only access to sample data, documents and the tools they use every day.
3. 03Prototype the riskiest ideaWhere a use case depends on accuracy, we test it on a small set of your real examples so the recommendation rests on results, not on a vendor demo.
4. 04Report and roadmapA written report with the ranked use cases, the build options, costs, risks and the first project scoped in detail.
5. 05WalkthroughWe present the findings to your team, answer questions and adjust the ranking where your context changes it. The plan is yours to build with us, in-house or with anyone else.

Most assessments take two to three weeks from kickoff to walkthrough, depending on how many teams are involved and how quickly sample data can be shared.

Our AI readiness assessment typically starts at $2,000 as a fixed fee. That covers the interviews, data review, use case scoring, one small accuracy test and the written roadmap for a single business unit.

Larger organisations with several departments, strict compliance requirements or more prototypes to test cost more, and we quote the fixed fee before starting. If you go on to build the first recommended project with us, the assessment work carries straight into the build, so nothing is paid for twice.

A good AI strategy says no more often than yes. These are the cases where we usually advise against an AI build:

The rule can be written down

If a decision follows clear rules, ordinary code is cheaper, faster and always consistent. In our own product Prozilio, the score each Upwork job receives is calculated deterministically with no AI call, so it costs nothing per job; AI is used only for writing.

The data does not exist

If the knowledge lives only in people’s heads, the first project is capturing it, not training anything.

The error tolerance is zero

Where one wrong answer is unacceptable and no human can review it, today’s models are the wrong tool.

The volume is too low

Automating a task done twenty times a month rarely repays the build and the monitoring it needs.

A plain automation will do

Many requests framed as AI are really integrations between two systems. A workflow in n8n or Make often solves them in days.

Most clients start with one contained project from the top of the list, prove its value in production, then move to the next. Depending on the use case that first project is usually an AI feature inside existing software, an automated workflow, a knowledge assistant over your documents or a custom AI agent.

Because the same engineers scope and build, the estimate in the report is the estimate you get. If you would rather build in-house, the report is written so your own team can pick it up.

We build and run AI ourselves

Our own products, including Prozilio, Quevix and Agenso, ship AI features to real users, so our advice comes from token bills and support tickets, not slide decks.

Recommendations come with prices

Every use case on the roadmap has a cost to build and an estimate of what it costs to run.

No vendor to sell

We are model-agnostic and tool-agnostic. If an off-the-shelf product solves your problem, we will name it.