Applied AI consulting for business
Every company that is not in the software business hears the same thing every week: that AI is going to change everything. What almost nobody explains is what it can do for their operations next Monday. Meanwhile, real operations carry on: hours of calls nobody can look up, documentation nobody can find, customers waiting for an answer.
This consulting service exists to answer the useful question —which of my problems does this solve and how much does it cost me— with data, not science fiction or «digital transformation».
One use case, with measurable return
Applied AI consulting for business is for companies that are not software companies and want to know, with numbers, what AI can do for their operations today.
The way to find out is not an eighty-page report: it is choosing one concrete use case —one that hurts and whose return can be measured—, getting it running as a pilot with real data and deciding with the results on the table. The goal is not to «add AI»: it is to remove hours, errors or cost from one specific place. If the pilot works, it grows; if not, it is parked having spent the minimum.
What it can do for your operations today
Nothing that follows is theoretical: these are things I have built and are in production.
- Your documentation answers questions. Semantic search and a chat that answers citing its sources over your company's knowledge: manuals, contracts, articles, reports. On the portal de Antonio Escohotado we did it over 21 books and hundreds of articles, videos and audio recordings, with server-side quote verification so the agent makes nothing up.
- Your calls are transcribed and analyzed automatically. Every conversation ends up as text, classified by category and with sentiment analysis, searchable afterwards like any other document. It is what Konect, the platform we built at Irontec, does for a service as traditional as telephone switchboards, with transcription even in Basque.
- Your customers write to you on WhatsApp and an agent answers them with your data. An agent configured from a CMS —its prompt, its tools, its model, its budget— serves users on Telegram, WhatsApp, Discord or Teams and answers with your organization's information. That is how our own platform works; its core is open source.
There are more pieces behind it —RAG over internal knowledge, multi-provider transcription, hybrid search—, but the technology is chosen after the problem, never before.
How I work
- Choose the case. Together we look for a process that consumes hours or money in a measurable way: answering calls, finding documents, replying the same thing to every customer.
- Pilot with real data. A bounded scope and your real information; neither a demo with toy data nor a months-long project.
- Measure the return. The metrics are agreed before starting: hours saved, cost per operation, hits and misses.
- Decide with numbers. Grow, adjust or park it: all three outcomes are a good result.
Cost, controlled from day one
A business is not scared of AI: it is scared of the surprise bill. That is why all model traffic goes through its own LiteLLM gateway that computes the real cost of each run and logs it with its tokens and its latency. Every agent carries its own budget and limits, which cut in before the bill grows, and Langfuse traces every call so we know which prompt cost what.
And because the gateway hides the real model behind stable aliases, switching models is a configuration change, not a development project. The cost per response is a visible figure from the pilot onward, not a surprise at the end of the month.
When AI is not the answer
Part of my job is saying no. If a process can be solved with a fixed rule, we program the rule. If a keyword search engine is enough, we set that up and save ourselves the model. Spotting that early is the cheapest way to apply AI: not applying it where it does not belong.
Who this consulting service is for
For companies that are not software companies —industry, services, media, a publisher— and want AI working in their daily operations, not in a slide deck. And for leadership teams that want numbers before committing budget.
If what you are looking for is for your team to program with agents, that is another service: agentic development consulting, designed for teams that write code.
Related consulting services
Antonio Escohotado's legacy, explorable with AI
Konect: audio transcription and analysis
Agentic development consulting
One deployment per customer or one space per customer
IT governance: who gets in, who can deploy, and how you prove it
From standard to graph to project
Let's talk
If you have a process that eats up hours every week and you suspect AI could take it over, tell me about it. A first conversation is enough to know whether there is a case with a return behind it.