AcXcel

Kaohsiung, Taiwan · Automation for smaller companies

A 30-minute quote,
done in 5 minutes

We take the work that eats the most hours and the most eyesight, and turn it into an AI-driven process that runs the same way every time.

Where it stalls

Automation usually stalls for reasons that have nothing to do with technology

You explain it again every single time

With a general-purpose AI tool, a vague instruction means another round trip. The time you save goes straight back into explaining.

Without a developer, you never get past the demo

The tools are capable. What's missing is someone with time to tune them to the point where they can actually go live, be handed over, and be debugged.

The slowest work is rarely the hardest work

Checking order numbers line by line, looking up prices, retyping the same figures — this costs eyesight, not judgement.

The problem isn't that AI isn't clever enough. It's that the process was never pinned down. What we deliver isn't a chatbot — it's a workflow that runs the same way every time and leaves a trail you can audit.

Services

What we build

Each one is built around how your company actually works. Nothing off the shelf.

Target · 30 min → 5 min per quote

Quoting, end to end

  • Reads the email body and any Excel or PDF attachment, however loosely formatted
  • Matches your price list and applies customer discounts, currency and lead-time rules
  • Drafts the quote and emails it to the right salesperson to approve before it goes out
  • Every quote records which price list version was used, and who approved it when
One email · A month of reconciliation

Spreadsheet reconciliation

  • Email two spreadsheets to a dedicated address; the results come back by reply
  • Fills a value only when the match is unambiguous — everything else goes on an exceptions report
  • Leaving a blank is always better than filling in the wrong number
  • Built so it can read from your ERP directly later, with no manual export
Round the clock · Customers self-serve

Product support assistant

  • Turns your manuals, FAQs and past support emails into an assistant that answers questions
  • Answers only from your documents, and cites where each answer came from
  • Hands over to a person when it isn't confident, and logs the gap
  • Monthly report on what customers ask most, so your docs can improve

Something else?

If the rules are clear and someone repeats the task every month, it can usually be automated — order entry, monthly reporting, moving data between systems, format conversion. Start with a process review and we'll work out together which one is worth doing first.

How it works

It starts with a review that costs you nothing

Every stage has something concrete to show for it, so you always know what you paid for and what comes next.

00

Process review

We look at how you work today, and tell you which task is worth automating first, roughly how long it would take, and which parts are better left alone. No pitch — if we don't end up working together, the assessment is still yours.

Free
01

Interviews and process mapping

We come in and look at the real files and the awkward edge cases, then write it down: a specification, a field mapping, and a signed scope document. Get this right and nothing has to be redone later.

1–2 weeks
02

Build and prove it

We build a working version and test it against your own historical data for accuracy and figure-for-figure correctness. The acceptance criteria are agreed up front — it isn't finished until it meets them.

6–10 weeks
03

Go live and keep improving

After launch we monitor, back up and tune it. Every time your team sends something back, the system learns another way customers write — accuracy climbs with use rather than freezing on day one.

Monthly service

How we work

Four things we don't compromise on

An audit trail

Three months on, when a customer asks how a price was arrived at, you can answer. Every quote records which price list version was used, what the AI read from the email, whether it was approved or sent back, and who decided when. The customer's original wording is kept verbatim, so you can see at a glance whether it was misread or simply written unclearly.

AI never does the arithmetic

Amounts, order numbers, discounts and tax are calculated by code following fixed rules. The AI only turns messy emails and tables into clean fields — it takes no part in the maths. That's what makes the output reproducible, auditable, and stable across model updates.

A person signs off at the end

Anything that goes to a customer, and anything written back into your systems, passes a human check first. The AI produces a draft; your team decides. Whether and when to relax that is your call, not ours.

Nothing to prompt

Your team keeps working the way they already do: send an email, or click approve in the one they receive. No new software to learn, no new tab to keep open, and no prompts to write — we maintain those and keep them current.

Typical situations

What this looks like in practice

Two common situations, and what changes.

Situation 1

A sales team fielding 50 enquiries a day

Today

Someone reads each email, opens the attachment, matches the items, looks up unit prices and discounts, builds the quote by hand and replies. About 30 minutes each. At 50 a day that's three people doing nothing else — and on a busy day a delivery date gets missed.

After

The system reads the email and attachment, matches the price list, drafts the quote and emails it to the right salesperson. The email shows what was read next to the customer's original wording, with anything uncertain flagged. They check the figures, click approve, and send it on. About 5 minutes each.

Situation 2

Monthly invoice reconciliation

Today

The billing sheet has sales order numbers and amounts, but the purchase order column is empty. Someone fills it in line by line from a separate ERP export. Once a month, hard on the eyes, and one wrong entry turns into a billing dispute later.

After

Accounts emails both spreadsheets in. A few minutes later a reply comes back with two files: the completed sheet, and an exceptions report. Only unambiguous matches with agreeing amounts are filled in; duplicates, misses and mismatches are listed with the reason, for a person to judge.

* Illustrative situations, used to explain the approach. Not figures from a specific client.

FAQ

You might be wondering

We're a small company. Is this for us?
Yes. This approach was designed for companies without spare hands — precisely because there's no slack, the time shouldn't go into repetitive work. We won't ask you to roll out a whole platform; we start with the single task that costs you the most time.
We don't have anyone in IT. Is that a problem?
No. Day to day there are two actions: send an email, and click approve in the one you get back. Monitoring, backups, updates and incidents are ours to handle, and we leave you a runbook plain enough to restart the system by following it.
If we stop working with you, does the system still work?
Yes. It was always in your account. On exit we hand over the full source, configuration files, the runbook and every credential. Everything is open source, so there's no licence to renew — you can maintain it yourself or bring in someone else.
How long until we see results?
From the first interview, usually 8 to 12 weeks to a version you can actually use, validated against your own historical data. In the review we'll give you a firmer estimate for your particular situation.

Start with a process review

30 minutes online, no charge. We look at how you work today, find the task worth automating first, and tell you roughly what it would take. If we don't end up working together, the assessment is still yours.

Email us →