LOYour first AI experiment at work
Your company wants you to use AI. You do five or six things every week. Which one should you try it on first — and how would you know whether it actually helped, rather than just felt faster?
This is not open yet. Before building it, we want to know whether anyone wants it — which is why the price is on this page. Leaving your email tells us that, and nothing more.
Price, and what it is not#
USD 9.90, once.
- No subscription and no auto-renewal.
- Joining the list charges nothing. This page is not a checkout: there is no payment page yet, and we will not take money until the service is built and the terms for it are published.
- It is one experiment on one task. A second task would be a second purchase you choose to make — never one we push because the first result was disappointing.
What you get#
- Describe what you actually do each week, in your own words. For free, you get back which task to try first and why — read off this site's own task judgements for your occupation, with the basis of each one shown.
- The paid part is the plan for that one task: what data you may and may not put into a tool, the steps, input and output templates you can copy, what a person must still check, how to take a baseline before you start, and when to stop.
- A place to record the baseline and the trials, and a conclusion drawn from your own records — continue, adjust or stop.
- “It did not help” is a result, not a failure of the product. Knowing that before your company spends a quarter on it is worth something.
A sample plan#
For a made-up person. Mei does not exist; the task judgement her plan rests on is real and is on this site.
Her week: accounts assistant at a made-up 40-person distributor in Singapore. Bank reconciliation (about 4 hours), entering supplier invoices (5), month-end accruals (2), answering sales about payment status (3). Her company allows Microsoft Copilot; client names may not leave the company's own tenant.
Why this task first: bank reconciliation
It happens every week, it is rule-heavy, and she can check every line herself. On this site it reads: Automating ✓ Evidence-backed Reconciliation →
What that judgement does not establish: The unmatched tail does not shrink in proportion to the matched volume. A practice that automates 95% of matching still needs someone who can chase the 5% — and that person needs to have seen the other 95% to recognise what is wrong.
Not chosen: month-end accruals — a judgement she is accountable for, and it happens once a month, too few times to compare in two weeks. Answering sales — the slow part is waiting for replies, not writing them.
The test
- The tool proposes matches between bank lines and open ledger entries and lists what it could not match. She keeps every approval.
- Export the week's bank statement and open items as two tables, with account numbers removed if policy requires it, and paste both into Copilot inside the company tenant with the template below.
- Check every proposed match against the source before posting anything, and log the minutes and any wrong match.
Two tables follow: bank lines, then open ledger entries. Match each bank line to at most one ledger entry, by amount and by date within 3 days. Output a table with: bank line, matched entry, and the reason for the match. List every bank line you could not match separately. Do not guess; if two entries could match, list both and match neither.
How she will know
- Baseline: three weeks done the usual way, then three weeks with the steps above, each logged the same way (minutes spent doing, checking and fixing).
- Continue: the median falls by at least 30 minutes a week and no wrong match reaches the ledger.
- Stop at once: a wrong match is posted.
- Not enough evidence: fewer than three comparable weeks on either side.
What it will not do#
The same refusals as the rest of the site. They do not get relaxed because someone is paying:
- It will not give you a readiness score, a risk percentage, or a number of years. None of those can be supported by the evidence, and a number is where people stop reading.
- It is for you, not about you. It will never tell an employer who to make redundant, and nothing you enter is shared with one.
- It does not promise to save you time, does not do your work for you, and does not touch your company's systems.
- It will not recommend a vendor. Nobody pays to be suggested, and nobody pays to be left out.
What stays free#
Every task judgement, every evidence record and every source on this site stays free, and none of it will ever move behind a payment. What costs money is the work of turning that evidence into a test you can run on your own job. That is a sentence you can hold us to: if a paywall ever appears over a judgement or a record, this page was not telling the truth.
VOLO is one person in Singapore plus a set of tools. Who makes this and who pays for it
Tell me when it opens#
What we keep: your email, your language, and — only if you give them — the occupation page you came from, the one sentence above, and whether you would try an early version. Stored in a separate service from this site. We write once, when it opens, and not again; the confirmation email has a link to leave the list, and you can write to the address on the privacy page to have the row deleted.