What Small Businesses Must Prepare Before Adding AI to a Website or Workflow
What must a small business prepare before implementing AI in a website or connected business workflow?
Prepare one clearly defined business problem, a mapped workflow, usable data, systems that can exchange the required information, a named owner, boundaries for data and human review, and a measurable pilot. Do not start by choosing a tool. First establish what the AI should improve, how the work happens now, what information it will use, who approves important outputs and how success will be judged.
Prepare one clearly defined business problem, a mapped workflow, usable data, systems that can exchange the required information, a named owner, boundaries for data and human review, and a measurable pilot. Do not start by choosing a tool. First establish what the AI should improve, how the work happens now, what information it will use, who approves important outputs and how success will be judged.
Start with the business decision, not the AI tool
AI readiness guidance recommends defining the business bottleneck and desired result before selecting a tool. Write a short decision statement: “When this event happens, this person currently does this work, and we want to improve this outcome.” Add one measure that will show whether the outcome improved, such as completion time, correction work, response consistency or the number of enquiries needing staff follow-up.
Sources: AI Readiness: A Small Business Guide | 1TC.
Use-case prioritisation can be based on impact and feasibility. For a first website or connected-workflow test, prefer a task with a worthwhile benefit and a practical path to testing it without redesigning the whole business.
Sources: AI Readiness: What Every Business Should Do Before Implementing AI - Atlantic | Tomorrow’s Office.
- Name one bottleneck rather than a broad ambition such as “use AI everywhere”.
- Describe the desired business result in plain language.
- Choose a measure before reviewing products or commissioning development.
- Keep the first decision tied to one workflow, customer journey or staff task.
Map the current workflow and decide whether AI is appropriate
Before changing a workflow, map the work as it is actually completed rather than as it is assumed to happen. Start with the trigger, then record the inputs, decisions, hand-offs, exceptions, output and the person responsible at each point. Include what staff do when information is incomplete, a customer asks an unusual question or a connected system is unavailable.
Ordinary automation is usually the clearer first option when a task follows stable, explicit rules, such as routing a form by postcode or sending a reminder after a fixed event. AI may be worth testing when a task involves interpreting language, classifying requests, extracting information, summarising material or recognising patterns that would be awkward to express as fixed rules. If an incorrect result could create a material customer, financial or operational problem, retain human approval in the first pilot.
- Record the trigger that starts the work.
- List the minimum inputs needed for a sound result.
- Show every hand-off between the website, inbox, CRM and staff.
- List normal decisions and unusual exceptions separately.
- Record the current output and baseline performance.
- Choose rules-based automation where rules are predictable; consider AI where interpretation is required.
Prepare the information and connections the task depends on
AI readiness guidance describes a business as ready when its data is reliable, its tools can exchange information cleanly and efficiently, and its processes have clear owners. For the chosen task, identify the smallest set of information needed to produce a useful output rather than making every customer record or document available by default. Check a sample for missing fields, duplicate records, outdated wording, inconsistent definitions and unclear source ownership.
Sources: Is Your Business AI-Ready? A Readiness Checklist.
Connected cloud systems can make implementation easier, but an API or integration does not automatically guarantee compatibility. Test the actual path the pilot needs: whether the correct record can be found, whether fields mean the same thing in both systems, whether updates arrive where expected and whether staff can see failures. Common operational readiness gaps include messy CRM data, disconnected tools, conflicting definitions, unclear ownership and missing performance baselines.
Sources: AI Readiness: A Small Business Guide | 1TC; Is Your Business AI-Ready? A Readiness Checklist.
- Define approved information sources for the pilot.
- Remove or correct obvious duplicates and stale records in the selected sample.
- Document what each required field means and who maintains it.
- Test the specific connection needed for the workflow.
- Confirm that the pilot can detect or report a failed hand-off.
- Limit access to information necessary for the stated task.
Assign an owner and define human review
Clear process ownership is one of the readiness conditions identified in AI readiness guidance. Name one business owner who is accountable for the workflow result, exception handling and the decision to continue, change or stop the pilot. The owner can gather input from staff, but accountability should not be spread so widely that no one resolves errors or reviews performance.
Sources: Is Your Business AI-Ready? A Readiness Checklist.
Decide in advance which outputs may assist staff, which require human approval and which actions the pilot must never perform automatically. Write a fallback path that lets staff return to the previous manual process when a result is uncertain, wrong, unavailable or outside the pilot rules.
- Assign a named workflow owner.
- Define which staff may view inputs and outputs.
- Mark outputs that require review before use.
- Keep consequential decisions and commitments human-approved during the pilot.
- Document how staff report errors and exceptions.
- Keep a usable manual fallback path.
Run a narrow pilot before expanding
AI readiness guidance recommends using a scored checklist to identify operational gaps and address them before scaling. Run the first test on a frequent but manageable task with a limited group of users, a defined start and end date, and a way to restore the existing process. Record the pre-pilot baseline before switching on the test so the business can compare the same work before and during the pilot.
Sources: Is Your Business AI-Ready? A Readiness Checklist.
AI readiness guidance warns that projects can stall and results can fall short when necessary elements are not in place. Test routine examples, incomplete inputs, unusual requests and connection failures before expanding access or automating further actions. At the review point, choose deliberately between continuing, narrowing, redesigning or stopping; expansion should be earned rather than treated as the default next step.
Sources: AI Readiness: What Every Business Should Do Before Implementing AI - Atlantic | Tomorrow’s Office.
- Choose one contained task and a limited pilot audience.
- Set a fixed review date.
- Capture a baseline for the selected measure.
- Keep the previous process available.
- Test exceptions rather than only ideal examples.
- Review quality, staff effort and operational impact before expansion.
Use this go-live readiness check
AI readiness guidance identifies reliable data, clean information exchange and clear process ownership as core conditions. Use the checklist below before allowing the pilot to affect a live website, customer enquiry path or connected operational workflow. An unanswered item is a reason to narrow or delay the use case, not necessarily a reason to abandon AI altogether.
Sources: Is Your Business AI-Ready? A Readiness Checklist.
Operational gaps commonly include disconnected tools, conflicting definitions, unclear ownership and missing performance baselines. Resolve the gaps that directly affect the chosen task before broadening the pilot or removing human review points.
Sources: Is Your Business AI-Ready? A Readiness Checklist.
- The business problem and desired result are written down.
- The current workflow, hand-offs and exceptions are mapped.
- The minimum required information is identified and checked.
- The required system path has been tested.
- One owner is accountable for the workflow.
- Human review points and prohibited automated actions are defined.
- Staff have a fallback path.
- A pre-pilot baseline exists.
- A review date and expansion decision rule are set.
AI Website or Workflow Preparation Decision Tool
Use this table to turn a broad AI idea into a controlled preparation decision. Complete each row for one workflow before selecting a product or broadening a pilot.
| Preparation area | Question to answer | Minimum evidence of readiness | If unclear |
|---|---|---|---|
| Business outcome | What bottleneck are we improving? | A written desired result and one measure. | Narrow the idea to one observable task. |
| Workflow map | How does the work happen today? | The trigger, inputs, decisions, hand-offs, exceptions and output are recorded. | Map the work with the people who complete it. |
| Information | What data is needed? | Minimum fields and approved sources are identified and checked. | Clean a limited sample or choose a simpler use case. |
| Connections | Can the required systems exchange the right information? | The actual end-to-end path has been tested. | Keep the pilot manual or reduce the connection scope. |
| Accountability | Who owns the result? | One person is accountable for outcomes and exceptions. | Assign an owner before starting. |
| Human review | What must people approve? | Review points, prohibited actions and fallback steps are written. | Keep automation at an assistive stage. |
| Pilot decision | How will we judge the test? | A baseline, review date and continue-or-stop rule exist. | Do not expand until a decision rule is set. |
This is practical planning guidance, not a universal technical, legal or security standard. Adapt the level of control to the information and consequences involved in the chosen workflow.
Frequently asked questions
Do we need perfect data before trying AI?
No. Prepare reliable information for the narrow task you intend to test. If the selected data is too incomplete or inconsistent to support that task, narrow the use case, improve the information or delay the pilot.
Does an API mean our website and CRM will work properly with AI?
No. An API or integration can make a connection possible, but it does not prove that records, definitions, permissions or updates will work correctly for your chosen workflow. Test the actual end-to-end path you need.
Should we use AI when rules-based automation can do the job?
Usually start with ordinary automation when the task has stable, explicit rules. Consider AI where the task needs language interpretation, classification, extraction, summarisation or pattern recognition and people can review important results.
What is the safest first AI pilot?
A controlled first pilot is narrow, measurable and reversible. It uses a known workflow, has a named owner, retains human review where needed, records a baseline and preserves a manual fallback path.
Related guidance
When should this approach not be used?
A small business does not need a perfect technology stack before trying AI, but it does need enough operational control to test one useful task safely. The sensible first move is a narrow, measurable and reversible pilot built around a real bottleneck, not a broad website or workflow rollout driven by enthusiasm for AI.
What follow-up questions matter most?
- Do we need perfect data before trying AI?
- No. Prepare reliable information for the narrow task you intend to test. If the selected data is too incomplete or inconsistent to support that task, narrow the use case, improve the information or delay the pilot.
- Does an API mean our website and CRM will work properly with AI?
- No. An API or integration can make a connection possible, but it does not prove that records, definitions, permissions or updates will work correctly for your chosen workflow. Test the actual end-to-end path you need.
- Should we use AI when rules-based automation can do the job?
- Usually start with ordinary automation when the task has stable, explicit rules. Consider AI where the task needs language interpretation, classification, extraction, summarisation or pattern recognition and people can review important results.
- What is the safest first AI pilot?
- A controlled first pilot is narrow, measurable and reversible. It uses a known workflow, has a named owner, retains human review where needed, records a baseline and preserves a manual fallback path.
What steps does this workflow follow?
Prepare a Small-Business AI Pilot
- Define one business problem: Write down the bottleneck, desired result and one measure that will show whether the selected task improved.
- Map the current work: Document the trigger, inputs, decisions, hand-offs, exceptions, output and current performance for the task.
- Check information and systems: Identify the minimum required data, check its quality and test the specific website or system connection required by the pilot.
- Set ownership and review: Name an accountable owner, define access boundaries, require review for important outputs and preserve a fallback process.
- Run and assess a narrow pilot: Test a limited workflow, compare results with the baseline and decide whether to continue, narrow, redesign or stop.