Tools are no longer the constraint for a small business: off-the-shelf AI services, CRM systems and bots are within reach of a team of five. The difficulty has moved to integration — in the 2026 OECD survey, 61% of small and medium-sized firms use at least one AI-enabled application, yet three quarters of them apply it to isolated tasks. That is where the difference in results now lies.
Key takeaways
What changed is not the availability of technology but the place where money leaks. A small business used to be blocked by budgets and development time; today a ready tool can be switched on in an evening, so the losses moved elsewhere — into the decision about which process is worth changing at all.
IT has stopped being a separate line of work "for the technical people" and has become part of the company's operating model. Enquiry records, client correspondence, pricing, documents and reporting now live in systems rather than in one employee's head or in private chats.
Hence the main question of the year is not "which AI service should we plug in" but "which part of the work will we stop doing by hand". The answer comes from data about your own processes, not from tool reviews.
More than half, but the depth of use is low. The OECD report "Empowering SMEs in the Age of AI" is based on a survey of more than 2,000 small and medium-sized firms across twelve OECD countries, run between the fourth quarter of 2025 and the first quarter of 2026. The sample is not representative and is skewed towards firms already trading on digital platforms — worth keeping in mind while reading the figures.
The second picture comes from the employee side. An Ipsos survey for the U.S. Chamber of Commerce Foundation (1,070 workers at US small businesses, May 2026, a nationally representative sample) found that half of them use AI at work. Of those, 64% use it for personal productivity — drafting, summarising, brainstorming — 26% for recurring tasks and only 6% to automate workflows with minimal human involvement.
Meanwhile the gap between small and large firms is widening. The official statistics quoted in the same OECD report are more modest than the survey: in 2025 AI was used by 17% of small and 30% of medium-sized firms. The difference in adoption between small and large firms grew to 35 percentage points, up from 23 in 2023 — everyone is growing, but large firms faster.
The difference is who holds the process: an employee with a chat window open, or the system. Below is the same path of an enquiry in two versions.
| Step | We use AI | AI is built into the process |
|---|---|---|
| An enquiry arrives | The manager sees a notification and reads the thread | The enquiry lands in the CRM and fields are filled from its text |
| Understanding the request | The manager copies the text into a chat with a model | The model sets the category, the priority and the missing data |
| Replying to the client | The manager carries the answer back by hand | A draft sits in the record; the manager edits it and sends it |
| What is left afterwards | A sent message and the employee's memory | A record, a history, a task for the next contact and data for reporting |
The second version takes nothing away from the person — it removes the carrying of data between windows. How to separate what a model can take from what an employee must decide is covered in our piece on automating processes with AI.
Integration runs into preparation rather than into the model. In the OECD survey the main obstacles named are maintenance costs, 39% of answers, and the lack of time for staff training, 38%. These are costs invisible in the price of the tool that show up in the second month of use. The Japanese part of the sample shows where that preparation leads: a transformational impact is reported by 2% of firms using AI in isolated contexts, 11% of those applying it across several functions and 23% of those deploying it enterprise-wide.
An AI agent is software that performs a sequence of steps towards a goal rather than answering a single prompt. So far this is a minority pursuit: under the OECD taxonomy, complex and customised solutions including agents are used by 5% of firms for specific functions and by 3.6% across the whole organisation.
For a team of ten to twenty people a simple rule follows: an agent is not the first purchase. The process has to be described and fixed in a system first, otherwise the agent has nowhere to take data from and nowhere to put the result.
Where the data is already collected, agents save noticeable time on first-line handling of enquiries and on marketing routine — the specific tasks are covered in our piece on AI agents in marketing.
Readiness is easy to test: if today you cannot say within a minute how many enquiries came in last week and how many were closed, an agent has nothing to work with.
Because the whole client path runs through them. A CRM has stopped being a sales department tool: it ties together the source of the enquiry, the correspondence, the payment and the repeat sale, and only on top of connected data do automation and AI work.
In Uzbekistan the first contact most often happens in Telegram, so a bot here is not a toy but the entrance to the funnel. In our UzFranchise Expo project more than 7,500 visitors registered through a Telegram bot, 3,539 QR tickets were scanned at the entrance over two days, and the whole base was collected straight into a CRM with segments.
Which rules are worth switching on in a CRM before the others is a separate conversation, covered in our piece on CRM automation. What matters here is different: as long as client data sits in three unconnected systems, no tool will produce an effect.
Since 21 January 2026 Uzbekistan has been applying law No. ZRU-1115, which added article 7-1 to the law "On Informatisation". It says plainly that in taking legally significant decisions affecting human rights and freedoms it is not allowed to rely solely on the conclusions of systems built on AI technologies.
For a small company this is a practical limit rather than an abstraction. Turning down a candidate, refusing a service to a client, deciding on a debt claim — these are decisions about a person, and a person signs off on them. The same law introduced administrative liability for unlawful processing of personal data using AI technologies.
Important A model may prepare a decision and explain it, but it cannot be the sole ground for it. In a process this looks like a mandatory human confirmation step and a record of who exactly confirmed.
By losses, not by fashion. Walk through the list and mark what you recognise in your own company.
Then cost the problem rather than the solution. A worked example: a company receives 500 enquiries a month, a hundred of them are handled too late, and the average profit per client is 100 dollars. Recovering even twenty clients is 2,000 dollars a month, and the budget conversation becomes concrete. The order of steps up to the first system is covered in our piece on where automation starts.
Check separately what happens to the data. According to the same report, almost half of firms, 46%, have no security measures or only minimal ones, and 22% have already faced a security incident. Once enquiries, correspondence and documents move into a system, access stops being a theoretical question: every employee gets their own rights rather than a shared login.
Below are the starting prices from our service pages; the final figure depends on the number of processes, integrations and the volume of data.
| Task | What it is | Price |
|---|---|---|
| Work out where to start | AI audit of processes: a list of processes, an automation map, an implementation plan | $400 |
| Take enquiries and bookings | A Telegram bot for enquiries and appointments | from $1,200 |
| Remove routine inside the company | Automation of repetitive operations | from $2,500 |
| Put sales in order | CRM implementation with a funnel and reports | from $3,000 |
| Answer from your own materials | An AI assistant that knows your product | from $3,000 |
| A system of your own | A CRM built around your way of working | from $12,000 |
The usual order is: the audit first, then one process, then connecting the systems to each other. The full list of tasks and terms is on the AI and automation page.
Syntra Systems is an IT company from Tashkent and an IT Park resident. We start not with the choice of a tool but with a review of processes: where an enquiry stalls, which data is carried by hand, and what of that can be measured before the work begins.
Then we assemble a minimum working circuit — enquiry records, systems connected to each other, and where needed a bot and a model on incoming messages. An AI audit of processes costs $400 and produces a list of processes, an automation map and an implementation plan with priorities.
If you already have a CRM and it stands idle, the task is a different one: revive the existing system rather than replace it. That is where the work most often starts — first what has already been bought, then anything new.
Let’s discuss your project
Tell us what you need, and we will estimate the timeline and cost and suggest a solution.
With a single process where the losses are visible as numbers: lost enquiries, time spent on reports, data carried by hand. Cost the problem first, choose the tool second. An AI audit of processes costs $400 and ends with a list of processes and an implementation plan with priorities.
A chatbot runs a dialogue along a script written in advance, while an agent performs a sequence of steps towards a goal: it finds data, checks it, updates a record in the system and creates a task. An agent needs access to data and somewhere to put the result, so it comes after the process already lives in a system.
It depends on what counts as use. In the 2026 OECD survey 61% of firms use at least one AI-enabled application, but 76% of them apply off-the-shelf tools to isolated tasks. The survey sample is not representative and is skewed towards firms trading on digital platforms.
Preparing the decision, yes; making the final call, no. Article 7-1 of the law on informatisation, added by law No. ZRU-1115 of 21 January 2026, prohibits relying solely on the conclusions of AI systems when taking legally significant decisions about human rights and freedoms.
It does if enquiries arrive through several channels and there are more of them than one person can hold in memory. The point is not the interface but the link: the source of the enquiry, the correspondence, the payment and the repeat sale in one place. While that data sits in separate systems, automation and AI on top of it will not work.
From our service pages: an AI audit of processes $400, a Telegram bot for enquiries and bookings from $1,200, automation of routine from $2,500, CRM implementation from $3,000, an AI assistant trained on your materials from $3,000, a custom CRM from $12,000. The final figure depends on the number of processes and integrations.
With a review of what is already there: where data never reaches the system, which fields nobody fills in, which reports nobody opens. Reviving an existing CRM is usually cheaper than replacing it, and it is a separate task with its own price on the service page.
