How to read the list
Each task is one paragraph: the name, what AI does, the metric and the site page that describes the solution in detail, with the price and term of a pilot.
The metric lets you compare the number before and after the pilot instead of arguing about impressions. If a task cannot be measured, it is too early to give it to AI.
Within each department, tasks go from simple to complex. In its guide “Building effective agents” (December 2024), Anthropic advises starting with the simplest solution and adding complexity only when needed.
The last section covers tasks where AI still makes mistakes. Know them before you start, so you do not spend the budget on a weak spot. If your company has more tasks, the AI Solutions Map shows the rollout order for the whole company.
Sales and leads
1. First reply to a lead: the agent answers the customer within a minute, day, night and at weekends, and asks what they need. Metric: time to first reply. The solution is on the page about AI for lead handling.
2. A deal in the CRM with no manual input: the customer card and the deal appear in the CRM right after the conversation, without the manager. Metric: the share of leads that reach the CRM. This is how an AI sales department works.
3. A consultant in the website chat: the agent answers from your price list and terms and asks for contact details from those ready to buy. Metric: the share of chat visitors who leave contact details. Solution: AI agent for your website.
4. Buyer messages on listing sites: the agent answers on Avito (Russia’s largest classifieds site) and marketplaces at any time and moves a ready buyer into the CRM. Metric: reply time to a message. Page: “AI for Avito and marketplaces”.
5. A draft commercial proposal: the agent builds a proposal from the price list based on the lead text, and the manager checks and sends it. Metric: time from lead to proposal. A calculation on open data is in the review of leads and proposals in manufacturing.
6. Calls outside working hours: a voice agent answers, takes notes and calls back missed callers. Metric: the share of missed calls. Page: “Voice AI agent for inbound calls”.
Support and booking
1. Standard customer questions: where is my order, when is delivery, how do I return it. The agent answers around the clock and passes hard cases to an operator with the chat history. Metric: the share of requests closed without an operator. Solution: AI for customer support.
2. Analysis of the request flow: the agent groups a month of requests by topic and shows which repeat most often. Metric: the share of standard questions in the total flow. The steps are described in a blog article on a plan for AI in support.
3. Client booking: the agent finds a free slot, books by phone or in a messenger and records the visit in the CRM. Metric: the share of bookings without a receptionist. Solution: AI receptionist for client booking.
4. No-shows: the agent confirms each booking the day before, moves the visit at the client’s request and offers the freed slot to others. Metric: the share of no-shows. A calculation for a clinic is in the review of visit reminders.
5. Replies to reviews: the agent answers reviews and buyer questions on marketplaces within an hour. Metric: reply time to a review. More in the review of replies to marketplace reviews.
6. One agent in all messengers: the client writes where it suits them, and the dialog history stays shared. Metric: the share of messages without a reply by the end of the day. Page: “AI agent for messengers”.
Finance, accounting and documents
1. Entering source documents: the agent records invoices and delivery notes in 1C, a Russian accounting system, from emails and scans, and the accountant checks and posts them. Metric: time per document. Solution: AI agent for 1C.
2. Overdue receivables: the agent sees a late payment in the accounting system on the first day, reminds the client and prepares a summary of debtors. Metric: the amount of overdue debt at month end. A calculation is in the review of receivables control in wholesale.
3. Management accounting: the agent collects revenue, margin, debts and cash by day from the accounting system, the CRM and the bank. Metric: the day of the month when the management report is ready. More on the page about management accounting with AI.
4. Formulas and pivot tables in spreadsheets: AI writes a formula from a description, cleans an export and builds a pivot table. Metric: hours on the weekly Excel report. Ready prompts are on the page “AI for Excel”.
5. Questions about company numbers: the question “how much do we owe suppliers” gets an answer from the data in a minute, not from an accountant a day later. Metric: time from question to answer. Page: “AI for data analytics”.
6. Comparing supplier offers: the agent puts commercial offers into one table by price, terms and conditions and flags gaps with the request. Metric: time to compare. An example from a large company is in a blog article on an AI agent in procurement.
HR and training
1. Screening applications: the agent reads all applications, asks candidates the first questions and gives the recruiter a short list. Metric: time from application to a reply to the candidate. Solution: AI for recruiting.
2. Mass hiring for a chain of locations: the agent answers each candidate on the day they apply and books an interview. Metric: the share of candidates who reach an interview. A calculation is in the review of an AI recruiter for retail.
3. Training managers: in an intensive course, managers learn to set tasks for AI and check its answers on their own processes. Metric: how many tasks a manager gave to AI after the course. The program is on the page about AI training for managers.
4. Training departments to work with agents: an AI director teaches staff to hand routine to an agent and handle disputed cases, without leaving their work. Metric: the share of staff who use the agent every week. More on the page about staff training by an AI director.
Executives and management
1. Morning report: sales, leads, stock and cash from the accounting system, the CRM and the bank reach the owner in one message at the start of the day. Metric: time spent on building the summary. How it works is on the page about the daily report to the owner.
2. Contractor control: AI checks the estimates, deadlines and acceptance reports of AI and IT contractors against the contract and flags deviations. Metric: the share of stages delivered on time and on budget. More on the page about control of contractors and the IT budget.
3. A quarterly rollout plan: what to automate first, with a budget and a metric for each stage. AI collects the numbers by process, and the decision stays with the manager. Metric: the share of stages where the metric was met. Page: “AI rollout strategy: a plan for the quarter”.
4. Decisions based on data: before a decision, AI collects numbers from your systems, and a disputed idea is first tested by a pilot with a metric. Metric: the share of decisions made from a report, not from memory. The method is on the page about AI for management decisions.
Tasks AI still handles poorly
The site has no ready solution for these four tasks, and this is not a gap but a limit of the technology.
1. High-stakes decisions without a human check: a dismissal, a large payment, a refusal to a key client. A model can be wrong in a confident tone, so AI prepares the draft and the arguments, and a person signs.
2. Tasks without data: a sales forecast for a new product with no history, or an estimate of a market the company has no records about. AI knows only what it was given about the business, and without data it gives a plausible guess. What calculations need is covered on the page about data for AI.
3. Negotiations and complex conflicts: a dispute with a key client, where the history of the relationship, trust and concessions decide. AI helps collect numbers for the meeting, but a person must lead the conversation.
4. Rare non-standard tasks: things that happen once a year and differently each time. Setting up an agent costs more than doing it by hand, and there is nothing to test the metric on.
How to catch AI errors before a client sees them is described on the page about AI errors.