The difference between an assistant that answers and a colleague who delivers is everything that happens after the answer.
The real sequence is usually: get the answer, copy it, switch to another system, paste, reformat, save. That step is always there, and nobody books it as cost.
Two ledgers
One: carrying results between systems.
| Input | Example |
|---|---|
| Results moved between systems per day | 20 |
| Minutes each: switch, copy, paste, reformat | 2 |
| Per day | 40 m |
| Per month, 22 working days | ≈ 15 h |
Fifteen hours is close to two working days that produce nothing — data moved from one window to another. With API access, this step goes to approximately zero: the result lands where it belongs.
Two: periodic checks.
| Input | Example |
|---|---|
| Manual checking per day (open systems, compare lists) | 30 m |
| Per month | 11 h |
| Days it actually happens | perhaps half |
The third row is the real finding. The problem with manual checking is not that it is slow; it is that it is the first thing dropped on a busy week — and a busy week is exactly when things slip. A scheduled job does not drop it. It runs on a Saturday, over a holiday, and while you are in a meeting.
In proportion: the carrying step falls close to 100%, and check coverage goes from "when there is time" to every day. The denominator matters — judging, negotiating and approving all stay with people.
The cost of one missed item
This one is not counted in hours:
| Input | Yours |
|---|---|
| Quotes or renewals missed per month | |
| Average value | |
| Win rate |
Multiply the three. For many companies this line is an order of magnitude larger than both ledgers above combined.
What it actually does
- Calls the APIs you already have: create records, update status, write back to tables — the result lands in the right place.
- Runs multi-step workflows, including ones where order matters.
- Scheduled jobs run daily: what is due and untouched, what is unanswered, what does not reconcile.
- Stays quiet when nothing is wrong. A system that sends five reminders a day stops being read within a fortnight.
What it will not do on its own
Committing actions wait for a person — submitting, paying, deleting, sending outward.
Not because the model is unreliable, but because the cost is asymmetric: retrieving the wrong figure costs a second look; submitting the wrong order costs a phone call and a reversal. That gate is also what lets the rest expand — a team that knows nothing is committed on its own will let an agent do far more of the preparation.
Tool access is granted per agent, not per installation. A reader and a writer can be two different agents.
This is a scenario on QSyn
Automation is not a separate workflow engine. It is the tool-use and scheduling capability of the QSyn platform, on the same knowledge, tables and permissions you already connected — nothing is connected twice.
Common questions
Do we have to replace our ERP?
No. Agents sit on top of the systems you already run and operate them through APIs or a suitable integration. Existing processes stay as they are.
Can it write back into our own systems?
Yes. Internal APIs, platform data tables and workflows can all be given to an agent as tools. Which ones each agent may call is configured individually, so the reading agent and the writing agent can be two different agents.
How do we stop it doing the wrong thing?
Three layers: answers come only from your own material with citations, tool access is granted per agent, and committing actions require human confirmation. When something does go wrong, the citation shows within a minute which layer failed.
How should scheduled checks be designed?
Conversation first, schedules second. A check needs a definition of "something is wrong", and that usually emerges from a few months of people actually asking things. Specify the checks on day one and you will specify the imagined ones.
