From practice · Calculation 8 min read

How much time an AI agent really saves on handling enquiries

The answer “it'll save you hours a day” is worthless until you see the calculation. Here is the whole thing — with numbers you can replace with your own, and an honest admission of what the agent will not take over.

How much time an AI agent really saves on handling enquiries
Author: Maksym Kamenchuk, Nexflow — Banská Bystrica Published Updated

The saving cannot be stated in advance, but it can be calculated. The formula is: number of repeated enquiries × time per enquiry × the share the agent takes over.

A model example (an illustrative calculation, not a client's result): a shop gets 40 repeated enquiries a day, each taking 4 minutes including the interruption to other work. That is roughly 2.7 hours a day. If the agent handles 70% of them, roughly 1.9 hours a day is saved — and the rest, the harder cases, go to a person with the groundwork already done.

The key numbers you have to supply yourself: how many enquiries genuinely repeat, and what share of those have an unambiguous answer. Estimates tend to be overstated — which is why I recommend logging it for three days.

Why answering costs more time than it seems

Ask how long it takes to answer a message and most people say “a minute”. The real cost is higher, because it is made of three parts:

That is why my calculations use 3–5 minutes per enquiry instead of “a minute”. It is not an inflated figure — it is the same figure you will get once you log the interruptions for three days.

The model calculation step by step

The following example is a model scenario for an e-shop — it illustrates the method, not a particular client's results. Below the table you can rerun it on your own numbers.

InputModel value
Repeated enquiries per day40
Average time per enquiry (including interruption)4 min
Daily time on enquiries≈ 2.7 h
Share the agent handles alone (realistically 60–85%)70%
Saved per day≈ 1.9 h
Saved per month (22 working days)≈ 41 h

Enter your own numbers. Nothing is sent anywhere — it is calculated right in your browser.

–daily time on enquiries
–saved per day
–saved per month
–saved per month in €

What to do with that: 41 hours a month at a cost rate of €14/h is roughly €574 a month. An agent at €900 with around €30 a month in running costs would pay back on these inputs within two months. Run it on your own numbers in the payback calculator.

Watch out for the two most common mistakes in this calculation: counting all enquiries instead of only the repeated ones, and putting in 100% instead of a realistic 60–85%. Both produce a number that does not hold up in practice — and disappointment in a project that was actually working.

Model calculation of the savingModel calculation of the saving1 · Enquiries and time• 40 repeated enquiries a day• 4 min per enquiry2 · Daily time• ≈ 2.7 h a day3 · Agent's share• 70% handled by the agent4 · Saving• ≈ 1.9 h a day• ≈ 41 h a monthA model example, not a client's result.
Diagram. Process illustration: the model time-saving calculation step by step.

What the agent takes over and what it does not

The split I use when designing. The percentages are illustrative — yours will differ, and you will find them in your real messages:

Takes over reliably

Questions with an unambiguous answer from data — availability, order status, terms, opening hours, the returns procedure.

Takes over with groundwork

Unstructured enquiries that have to be understood and categorised — the agent sorts them, adds context and hands them over prepared.

Does not take over

Individual prices and discounts, expert advice, complaint decisions, grievances and sensitive situations.

The third category is why I never plan for 100%. And that is fine — those are exactly the cases where having a person present has the greatest value.

What the agent takes overWhat the agent takes overCategoryExamplesTakes over reliablyAvailability, order status, termsTakes over with groundworkUnstructured enquiries: sorts, adds contextDoes not take overIndividual prices, complaints, grievances
Diagram. Three categories of enquiries by what the agent can handle.

How such an agent is built

  1. Collect real enquiries. The last few hundred messages. Without them, any deployment is guesswork.
  2. Sort them by topic. It usually turns out a small handful of topics accounts for most of the volume — and those are the candidates for automation.
  3. Define the scope. Which topics the agent handles, what it escalates, where a person confirms. This is the most important part of the project.
  4. Connect it to data. So the agent answers from the current state, not from a description that has since changed.
  5. Test on real messages including the awkward ones, and tune the boundaries.
  6. Deploy with supervision and expand the scope only in line with what operation reveals.

Companies most often skip the first step and want to go straight to deployment. But without real messages you cannot even say whether an agent pays off — and that is the information that costs the least and decides the most.

How the agent is builtHow the agent is built1 · Collect enquiries• The last few hundred messages2 · Sort by topic3 · Define the scope• What it handles, what it escalates4 · Connect to data5 · Test and deploy• Deploy with supervision
Diagram. Process illustration: five steps of building an enquiry agent.

When it is not worth running an enquiry agent

When there are few enquiries. Below roughly 10 repeated enquiries a day, deployment and tuning do not come back.

When they do not repeat. If every enquiry is different and needs expert judgement, the agent will manage sorting and groundwork at most.

When there is no data. The agent has nothing to answer from if the information is written down nowhere or is out of date.

When there is nobody to take an escalation. The agent has to have somewhere to hand what it cannot manage. Without that it only postpones the problem.

Frequently asked questions

How much time does an AI agent really save?

It depends on the number of repeated enquiries and on what share of them have an unambiguous answer. The formula is: number of repeated enquiries × time per enquiry × the share the agent takes over. The realistic share is 60–85%, not 100% — there will always be cases for a person.

Why do you count 4 minutes per enquiry?

Because writing the answer is only part of it. Add establishing the context (opening the order, checking stock) and above all the interruption — getting back to a half-finished task takes several minutes. With dozens of interruptions a day, that is the biggest loss.

Are those numbers from a real client project?

No. This is a model calculation showing the method, not measured results from a particular deployment. I state that deliberately, because numbers you have no way to verify are worth nothing for your decision. You put your own numbers into the payback calculator.

Will the agent handle 100% of enquiries?

No, and it should not. Individual prices, expert advice, complaints and sensitive situations belong to a person. The realistic share of automated enquiries is 60–85%, and you should plan on the rest always needing a person.

From how many enquiries does an agent pay off?

Roughly from about 10 repeated enquiries a day. Below that, deployment and tuning usually do not come back. What decides is the number of repeated enquiries, not the total volume of communication.

What do you need from me to give a price?

Ideally an export of the last few hundred messages (without personal data — the text of the questions alone is enough). From those it is possible to determine how many enquiries repeat, how many have an unambiguous answer, and whether an agent pays off at all.

How fast does the agent answer?

Practically immediately, because it reacts to an event and does not wait for working hours. That is one of the main advantages — enquiries arriving in the evening or at the weekend get an answer straight away rather than the next day.

Related services

ServiceAI agent for business ServiceAI chatbot for business

Related articles

PaybackHow to calculate the payback of automation DecidingAI agent vs chatbot vs automation SectorAI agent for an estate agency

Related work

Showcase buildAI sales call analysis Showcase buildAI analysis of YouTube comments

Want to know what it would save you?

Send me a sample of your enquiries — the text of the questions alone is enough. Free of charge I will tell you what share of them have an unambiguous answer and whether an agent pays off at your volume.

Get a free consultation →