INDAGOR — digital engineering studioINDAGOR
Automation

What to automate first: 5 processes that eat up business time

Published: 6 min readINDAGOR
Business automation — INDAGOR

Automation usually starts not with picking a tool, but with a question: where exactly the time goes. Until that question has an answer, any implementation is a tool bought without a task. Below are five processes that most often become the bottleneck, and the signal that it is time to change something.

The criterion is simple: frequency × volume × uniformity. A process is worth automating if it repeats regularly, takes meaningful time and follows clear rules. If every case is unique and requires judgement, automation delivers little and costs like full custom development.

1. Handling inbound enquiries

How it looks today. Enquiries arrive from the website, email, messengers and calls. A manager copies them into a spreadsheet or CRM by hand, some are lost in threads, and the customer waits hours for a reply.

What gets automated:

  • a single intake point — the enquiry is created automatically from any channel;
  • normalisation and enrichment: phone, company, source, duplicates;
  • rule-based first qualification: region, budget, type of task;
  • routing to an owner and task creation with a deadline;
  • automatic confirmation to the customer with an expected response time.

What stays with people: complex and non-standard requests, negotiations, commercial terms.

How to measure the effect: time to first response, share of enquiries handled within 15 minutes, share of lost enquiries.

2. Answering common customer questions

How it looks today. Most enquiries are repeated questions about terms, timelines, payment, availability, delivery. The employee answers from memory or searches an old thread for the wording.

What gets automated:

  • a knowledge base built from your documents and policies;
  • an assistant that answers only from that base and cites the source;
  • escalation to a human when the answer is not in the base or the question is out of scope;
  • gap collection: which questions most often get escalated.

What stays with people: complaints, individual terms, anything that needs empathy and authority.

How to measure the effect: share of conversations closed without escalation, time to first response, share of repeat enquiries on the same question.

3. Reporting and data gathering from several systems

How it looks today. Once a week or month someone exports data from the CRM, spreadsheets, ad accounts and accounting, reconciles it by hand and builds a report. Need another period? Start over.

What gets automated:

  • scheduled API exports into a single store;
  • metric calculation with fixed formulas;
  • a dashboard refreshed on a schedule;
  • deviation alerts: plan vs actual, overdue items, anomalies.

What stays with people: interpretation and decisions. A report shows what happened — it does not decide what to do about it.

How to measure the effect: hours spent preparing the report, data freshness on the dashboard, number of discrepancies between sources.

4. Documents: invoices, delivery notes, contracts

How it looks today. Details and amounts are keyed in by hand, documents get lost between inbox and folders, and errors surface at payment or reconciliation.

What gets automated:

  • scan and PDF recognition, extraction of details, amounts and dates;
  • rule-based validation: tax ID match, amounts, dates, duplicates;
  • an approval route with owners and deadlines;
  • data transfer into the accounting system without re-entry.

What stays with people: disputed documents, non-standard forms, decisions on discrepancies.

How to measure the effect: time to process one document, error rate, share of documents processed without manual intervention.

5. External data collection: prices, vacancies, mentions

How it looks today. Someone opens competitor sites, marketplaces or aggregators once a week and copies numbers into a spreadsheet by hand. The data is out of date by the time it reaches a report.

What gets automated:

  • scheduled collection from public sources;
  • normalisation to a single format: names, units, categories;
  • change history — not just "what it is now" but "how it moved";
  • threshold alerts: price below a set level, a new vacancy, a brand mention.

What stays with people: analysis and conclusions. Collection is mechanical work; interpretation is not.

How to measure the effect: number of sources watched, refresh frequency, time spent preparing the summary.

How to choose the first process

If you are unsure where to start, answer five questions for each candidate:

  1. How often does it repeat — daily, weekly, monthly?
  2. How much time does it take in total, multiplied by the number of operations?
  3. How uniform is the process — are there rules, or is every case unique?
  4. What does an error cost — money, reputation, rework?
  5. Is the data available — API, exports, database, or does it all live in someone's head?

How to calculate the upside before you start

Simple arithmetic that usually brings the conversation down to earth: operations per day × minutes per operation × working days per month. If a manager spends three minutes moving an enquiry from email into a spreadsheet and there are forty such enquiries a day, that is two hours every day — roughly forty hours a month on a single mechanical operation. Compare that with the cost of an employee-hour and the cost of automation: the discussion usually ends there, without slide decks and "digital transformation".

A practical rule: pick one scenario where the effect is visible in two to four weeks. A pilot on a single process is cheaper and faster than a full automation programme, and it produces real numbers instead of assumptions.

What not to do

  • Automate chaos. If the process is undocumented and everyone does it their own way, automation only cements the mess.
  • Start with the hardest case. The first scenario should deliver a quick, obvious result — not test the team's patience.
  • Buy the platform before the task. Scenario and acceptance criteria first, tool second.
  • Measure by the rollout. "We implemented AI" is not a result. Time per operation and error rate are.

What you need to start

Automation does not require "rolling out everything at once", but it does require four things without which a pilot goes nowhere:

  • An owner. One person on the business side who decides on the process instead of approving it for weeks.
  • Data access. API, exports or database access — what the system can read and where it writes results.
  • A description of "what good looks like". Even as a spreadsheet or a one-page policy: what counts as correct handling.
  • An acceptance criterion. The number that tells you it got better: time, error rate, volume processed per period.

If any of these is missing you can still start — but the first stage will be process housekeeping rather than development. That is a normal scenario: half the work in such projects is not code, but agreement on how the process should look.

In short

Automation is not about technology. It is about picking the slice where manual work costs the most. The five processes above cover most of the standard routine: intake, customer replies, reporting, documents and external data collection. Start where rules already exist, data is reachable and the result is measurable within a couple of weeks.