Where AI Should and Should Not Sit in a Small Business Process
Most owners are not asking whether AI is good or bad. Instead, they are asking something more practical: where should it sit in the business?
That is the right question. AI is rarely a success or failure on its own. More often, it is well placed or badly placed.
A badly placed tool drafts a reply that nobody checks. A well-placed tool prepares the reply, while a person decides what to send.
This article offers five questions for assessing any process step. It then sorts those steps into three zones and explains what meaningful checking involves.
Why AI placement matters more than the tool
The same AI tool can support one step and create risk in another. The difference is where it sits and who remains responsible.
For example, summarising a long email thread may be easy to check. Deciding whether a complaint deserves a refund carries more consequence. The tool may be the same, but the decision is different.
The UK Government AI Playbook was written for government organisations, not small firms. Even so, two principles provide useful reference points.
- Maintain meaningful human control at the right stage.
- Use the right tool for the job.
The Playbook also notes that an established technology may sometimes be a better fit than AI.
Therefore, the useful question is not simply whether to use AI. It is which steps suit it and which should remain with people.
Five questions to ask before adding AI
Choose one step in a process, then ask these five questions. They are Davina Connect’s practical test, not a research finding.
1. Is the step clear and repeatable?
AI works better with a recognisable pattern. If the step changes every time, or nobody agrees how it works, it is not ready. A tool will simply repeat the confusion.
2. What would a mistake cost?
A wrong internal summary may be easy to spot and correct. A wrong customer message is harder to take back. As the consequence rises, human involvement should increase.
3. Does the step require judgement, care or accountability?
Some steps depend on trust. A difficult conversation, safeguarding concern or decision affecting someone’s job needs a person who can explain and own the outcome.
4. Is the information accurate and accessible?
AI works from the information available to it. If records are stale, incomplete or scattered, its output may reflect those gaps. Therefore, examine the data before choosing the tool.
5. Who checks the result?
Every AI-supported step needs a named owner. The checker also needs enough knowledge, authority and time to challenge the output.
Three zones for AI in a small business process
After asking the five questions, most process steps can be placed in one of three zones. These examples are illustrations, not universal rules.
Green: AI can support the step
Green steps are clear, lower in risk and easy to check. AI might summarise notes, sort incoming messages or draft a first version for editing. It may also remind somebody that a follow-up is due.
Amber: AI prepares, a person decides
In amber steps, AI can complete the groundwork, but a person makes the decision. Examples include drafting a complaint response, preparing a quotation or suggesting enquiry priorities.
In each case, the person reads, changes and owns the result.
Red: the decision stays with a person
Red steps depend heavily on judgement, care or accountability. Examples may include safeguarding, hiring, dismissal, unusual pricing or sensitive customer conversations.
AI may support administration around these steps. However, it should not make the decision.
The zones are not permanent. A step may move as the business learns more about the tool, its errors and its effect on customers.
What meaningful human checking involves
Many businesses say that a person reviews every AI output. However, a review only protects the process when it is meaningful.
The ICO warns about automation bias. This happens when people accept a system’s suggestion without properly challenging it.
Its human-review guidance says reviewers need appropriate knowledge, experience, authority and independence. It also recommends manageable workloads and records of decisions that people override.
The guidance is currently under review following the Data (Use and Access) Act. Therefore, businesses should recheck it when designing higher-impact uses.
For a small business, ask three questions. Does the checker understand the subject, have permission to disagree and have enough time?
If one busy person approves fifty drafts without reading them properly, the process does not have meaningful oversight.
A simple way to start
You do not need a large policy manual before examining one process. Start with a short, shared record.
- List each step in one process, such as handling an enquiry or sending an invoice.
- Use the five questions to mark each step green, amber or red.
- Name one person who owns every AI-supported step and its checking process.
- Review the decisions after one month and move steps when the evidence supports it.
The Department for Science, Innovation and Technology now provides two useful resources for organisations introducing AI.
AI Management Essentials is aimed mainly at SMEs and start-ups. It offers a self-assessment of the organisational processes supporting responsible AI use.
The AI Risk Management Toolkit, published in September 2026, provides a broader starting point for understanding and managing AI risks.
An example of placing AI within one process
This example is invented to demonstrate the method. It is not a client story.
Imagine a small firm handling customer complaints. Its owner maps five steps: receive, log, investigate, decide and reply.
Receiving and logging are clear and easy to check, so the owner marks them green. AI can sort messages and extract key details.
Investigating and replying are amber. AI can gather the history and prepare a draft, but a person checks and edits the work.
The final decision is red because it affects the customer and the firm’s reputation. A person decides the outcome and remains accountable for it.
Nothing here requires a large system. It requires a clear process and visible responsibility.
Map the process before placing the AI
Placement is one part of a wider picture. First, map the process. Next, check the information. Only then can the technology earn its place.
Continue with the most relevant article for your situation:
- AI Enquiry Assistants: Where They Fit and Where They Don’t
- When AI Creates More Work
- How Can AI Support Domiciliary Care Providers Without Replacing Human Care?
- What Happens After a Customer Enquiry? Map the Process Before You Automate It
- Why Your CRM Data Is Out of Date Before the Week Ends
See how connected lead management works in the Bridging Loans Broker case study.
Frequently asked questions
Where should AI sit in a small business process?
Place it in steps that are clear, proportionate in risk and easy to check. It can also prepare work for a person. Keep consequential decisions with people who can explain and own them.
Which business tasks should stay human?
Keep tasks requiring care, judgement or accountability with people. Examples may include safeguarding, employment decisions, unusual pricing and sensitive conversations. Each business should identify its own high-consequence steps.
Does somebody need to check every AI output?
The level of checking should reflect the potential consequence. Customer-facing or decision-support outputs need meaningful review. The reviewer must understand the subject and have the authority to reject the result.
What is automation bias?
Automation bias occurs when people rely on a system’s recommendation without applying enough judgement. A human approval step is not meaningful when the reviewer simply accepts every output.
Does a small business need an AI policy?
A short written record is a sensible starting point. State which tools may be used, what information they may receive, who checks outputs and which decisions remain human. Obtain specialist advice when required.
How should a small business begin using AI?
Choose one process and list its steps. Apply the five questions, assign each step a zone and name an owner. Test the arrangement on a small scale before expanding it.
Start with the business
Before adding AI, understand how the work flows and where responsibility sits. The right place for AI then becomes clearer, as does the place where it does not belong.
The Davina Connect Business Review examines people, information, systems and processes together. Recommendations do not begin with a tool.
