A.L.E.X.
Area of work · AI assistants

AI assistants for business: where to start so the money is not wasted

An assistant is not "adopting AI". It is one spot where people are asked the same things over and over: your website, a messenger, the phone, or your own staff. We work out where the repetition piles up and start with one spot.

What is in this section
  • →We startwith one spot
  • →The first stepneeds no access to your systems
  • →The exampleruns on this site
The exact scope and timing are fixed after the diagnostic — before that nobody has an honest picture.
Services in this section

What people come with most often

Four spots where an assistant pays for itself first. They differ not by "technology" but by who talks to it and how.

An assistant is not a replacement for people. It takes away repetition — the same question asked for the hundredth time. Anything that needs a decision it hands to a person, and that is written into the script rather than left to the machine’s conscience. If what repeats is not questions but actions — moving data, reports, template emails — that is the neighbouring job: automation. And if the assistant has nothing to stand on yet, start with the website.

Who it fits

When people come to us with this

Four situations in which an assistant is the right place to begin.

The same question ten times a day

"How do I book", "how long is the wait", "are you open at the weekend", "where are you". You already have the answer — it is simply spoken again by a human every time.

Nobody answers at night or at the weekend

The message arrives at 23:40 and the reply goes out at 10:15 the next day. In that gap the person has written to three other companies.

You have tried it before and gave up

You put in a button-driven bot, it did not understand plain speech, customers got annoyed and it was quietly switched off. This is a common story, and it is not about "AI not working".

The knowledge lives in one person’s head

How to file it, where it is kept, who to call in this case — only Lena knows. Lena is on holiday and the work stops.

Outcome

What stays with you afterwards

A list of repeated questions — before any technology

The first result of the diagnostic is not an assistant but a list: what gets asked most, how much time it eats, and what is actually worth handing to a machine. Some of it usually gets solved without one.

An assistant on one spot, not "AI in the company"

One channel, a clear set of questions, a visible result in weeks. We widen it if it works — not the other way round.

Answers from your material, not from thin air

The assistant answers from what you gave it. Where there is no answer it says so and calls a person — and that is configured and tested, not left to luck.

How the work goes

A clear order, no surprises

1 · A conversation

Half an hour about what people ask you most often and who answers it today. No access to your systems is needed at this step.

2 · A list of questions, and the first spot

We collect the questions that keep repeating and pick one spot — the website, a messenger, phone calls or your own team’s questions. The list stays with you either way.

3 · We put it live and watch real conversations

The assistant goes live on one spot, we read the actual conversations and fix the answers against them rather than against guesses. We tell you what changed and why.

The figure is named in a quote for your task, once it is clear which questions the assistant answers, where it takes the answers from and who it hands over to. There is no price list here on purpose: the spread of work is very wide, and an "average price" would mean averaged work. Most of this market keeps prices off the website too — we at least say why.

Evidence

Look at the work, not at the words

Figures about us, not about the market: how many models we keep in service, how many jobs went from request to accepted result, and how long that took on average. The caption and period under each number are placed by the code that builds this page.

—AI models in the park, for different kinds of work
—tasks accepted: result checked and taken
—median task time: from brief to accepted

preparing data…

Where the figures come from
  • AI models in the park, for different kinds of work

    How many different AI models the agency can reach through its connected channels. Uniqueness is counted per “lab + base name”: variants of one model are not counted separately.

  • tasks accepted: result checked and taken

    A task is accepted once the result has been read, checked and taken. There is no second word for this on the site: “accepted” and “completed” mean the same thing.

  • median task time: from brief to accepted

    A typical task: half of the tasks in this window went faster, half took longer. This is NOT an average: one long task cannot pull the number up, so the median describes an ordinary day more honestly.

Straight talk

What we do ourselves and what we do with partners

We do this ourselves

The website assistant runs on us first: the "Ask A.L.E.X." button on this page opens Chrysa. She says hello, states straight away that she is an AI, answers in text and by voice, reads a document you attach, and hands the conversation to a person when she cannot cope. It is a live product, not a screen recording.

Sorting and triaging incoming enquiries we run in production code on our own flow of enquiries — that is, on our own money rather than on a demo stand.

We build this out for you

An assistant inside your own perimeter — on your material, with your access and your rules for handing over to a person — is built for the job. What connects straight away, what goes through an export and what does not connect at all, we say at the diagnostic, before the work starts.

We run this with a partner

Telephony, call recording and accounting systems are usually run by your own contractor. We work on the layer above — answers, analysis, handover to a person — and we do not propose replacing what already works for you.

Where a partner does part of the work — say, your telephony or your accounting system is set up by its own contractor — we stay your single point of contact: we set the task, watch the timing and answer to you for the result. The contract names the contractor honestly — no "we do everything in-house".

Questions

What people usually ask

How is an assistant different from a bot with buttons?
A button bot walks you down a tree drawn in advance: one step aside and it is lost. An assistant understands the question in plain words and answers from your material. And the difference that matters: it has a boundary at which it calls a person instead of pushing the customer around a menu.
Will it make things up?
Making things up is the real risk here and we are not going to talk around it. The assistant answers from your material; for a question your material does not cover, it is required to say so and hand over to a person. That is configured and checked against real conversations rather than promised in words.
Will it replace our staff?
No, and we will not sell you that. It takes away the repetition — the same question for the hundredth time — while the live work stays with people. If somebody promises you a replacement for people, ask them who answers the customer when the machine gets it wrong.
Which spot should we start with?
The one where questions repeat most and where somebody living is counting them. Usually that is the website or a messenger: the conversation is written down and "before and after" is easy to compare. Calls come next — they cost more to work through.
Enquiry

Let us look at your task

Half an hour about what people ask you most often. The list of repeats stays with you even if we do not work together afterwards.

We call back during working hours. Company details are optional — if you give them, we read the open registries first and come to the conversation prepared.