Before You Buy an AI Agent, Try Automating Your Workflows First

The Agent Obsession

I was in a strategy meeting with a manufacturing client last week. The VP of operations said: "We need an AI agent to optimize our supply chain." The CTO nodded. The CFO asked how much it would cost. They budgeted £500,000 for an AI agent.

I asked a simple question: what are the top three decisions in your supply chain that cost you money when they go wrong?

The VP listed them. Then I asked: are those decisions being made wrong because you lack intelligence, or because your process is broken?

Silence. Then: "The process is broken. We have people making decisions with bad information because the information does not flow between systems."

So I said: before you build an AI agent, fix the process. Before you spend £500,000 on intelligence, spend £50,000 on workflow automation. You will probably solve 80 percent of the problem.

They did not want to hear it. They wanted the shiny new agent. But that is not how this works.

Why Companies Jump to Agents

I understand the appeal. AI agents are exciting. They are new. They are impressive. They feel like the future. Workflow automation is boring. It is not new. It does not feel current. So companies skip straight to agents.

This is a mistake. Most business problems are not intelligence problems. They are process problems. Someone has to manually copy data from one system to another. Someone has to call another department to get information. Someone has to fill out a form that should be automatic. Someone has to wait for approval because the approval routing is broken.

These are not problems that intelligence solves. These are problems that process solves.

Practical tip: The next time someone says "we need an AI agent to solve X," ask: is X a problem because we lack information, or because our process is broken? If it is the latter, automate the process first.

The 80/20 Rule for Business Problems

I have audited workflows in dozens of organizations. I have found that roughly 80 percent of workflow problems are process problems. 20 percent are intelligence problems.

Process problems: information does not flow to where it is needed. Approvals get stuck. Data is duplicated in multiple systems. Work is duplicated. Steps are done in the wrong order. Rules are enforced inconsistently.

Intelligence problems: we need to predict something (will this customer churn?). We need to classify something (is this email spam?). We need to optimize something (what is the best price?). These are the problems AI actually solves.

Most organizations are struggling with process. They see their struggle as an AI problem because AI is trendy. That is a category error.

The Workflow Audit Framework

Before you buy an agent, audit your workflows. Here is how I do it.

Step 1: Identify Your Bottleneck Workflows

Not all workflows matter equally. Some workflows are bottlenecks. They slow down important work. These are the ones to focus on.

How do you identify them? Ask: what workflow, if it were faster, would have the biggest impact on the business? What workflow do we complain about most often? What workflow requires people to do things that seem unnecessary? What workflow involves the most handoffs between people?

Usually there are 3 to 5 workflows that matter most. Start there. Do not try to audit everything.

Step 2: Document the Current State

Walk through the workflow step by step. You will be surprised what you find.

Get someone who does this work every day. Have them walk you through. What is step 1? What is step 2? What information is needed at each step? Where does it come from? What systems are involved? What happens if something goes wrong?

This usually takes 2 to 3 hours per workflow. Write it all down. Diagram it. You are looking for inefficiencies.

Practical tip: Do not ask the person what the workflow is. Have them walk you through a recent example. "Show me the last three times you did this." Watch them do it. You will see waste you would not see if you just asked them to describe it.

Step 3: Identify the Waste

Common waste patterns:

Duplicate data entry: someone enters data in one system, then enters the same data in another system. Huge waste. The data should flow automatically.

Manual approvals: someone has to manually review and approve something that could have auto-approval rules. Huge waste. Define the rules and let the system decide.

System disconnects: someone has to manually check one system, then manually input into another system. Huge waste. Systems should be integrated.

Unnecessary steps: steps that nobody remembers why they exist. They are there because that is how it was always done. Huge waste. Delete them.

Sequential steps that could be parallel: someone finishes step 1, sends it to person B for step 2, who sends it to person C for step 3. All sequential. Could be parallel. Moderate waste. Reorganize the workflow.

Bad information flow: person A needs information from person B, but person B does not know. So person A wastes time asking. Information should flow automatically to where it is needed.

Step 4: Quantify the Impact

How much time does this workflow waste? How many people are involved? How often does it happen?

If the workflow takes 2 hours per day and involves 2 people, that is 10 hours per week. That is 520 hours per year. At £40 per hour, that is £20,000 per year of wasted labor. Just on one workflow.

Most companies have 5 to 10 workflows with significant waste. The total waste is often hundreds of thousands of pounds per year.

Simple Automation First

Once you have identified the waste, automate it. Start with simple automation. If/then rules. Email triggers. API integrations. Scheduled syncs.

Do not immediately jump to AI. Nine times out of ten, simple automation solves the problem.

Practical tip: For each workflow, list the waste. For each waste item, ask: is there a simple automation that solves this? Example: "data entered in CRM needs to sync to accounting system" → solution is an API integration or a scheduled sync. This is simple automation. No AI needed.

When You Add AI

Only after you have automated the workflow should you consider adding AI. And only for the parts where intelligence actually helps.

Example: automated workflow sends a customer support ticket to the right department based on keywords. This is simple automation. But the assignment is sometimes wrong because some keywords are ambiguous. So you add AI. The AI learns from historical ticket assignments. It predicts the right department even for ambiguous keywords. This is where AI adds value.

But if you had not automated the initial routing, you would be drowning in manual work before the AI ever got deployed.

Real Example: The Expense Reimbursement Nightmare

I worked with a consulting firm that had a terrible expense reimbursement workflow. Employees submitted expenses. HR reviewed them for policy violations. Finance approved them. Accounting paid them. The whole process took four weeks.

Employees were frustrated. Finance staff was drowning in review. Accounting was processing late payments.

Someone suggested an AI system to process expenses. I suggested we first audit the workflow. What we found:

Expenses were submitted in email (some) or a spreadsheet (others) or a form (sometimes). Someone had to collect all of them. Someone had to manually check each one against policy. Many expenses had missing receipts or unclear business purpose. Someone had to email the employee. Someone had to re-submit. Then it went to Finance. Then Accounting. Then payment.

The waste was not intelligence. It was process. So we automated it:

Built a simple form. Expenses submitted in one place. Simple auto-checks: receipt attached? Amount less than policy limit? Straightforward approvals routed to the right person. Payroll integration so payment is automatic.

Processing time dropped from four weeks to four days. Staff time dropped by 70 percent. Cost to implement: £15,000. Timeline: 8 weeks.

We did not add AI. We did not need it. The problem was process, not intelligence.

The Agent Question: When Do You Actually Need One?

You actually need an AI agent when:

You have a decision that requires judgment based on multiple data sources. You have a process that needs to adapt based on changing circumstances. You have a problem where simple rules are not enough.

Example: a loan approval process. Rules say "approve if FICO score above 700 and debt-to-income below 40 percent." But what if someone has a FICO of 650 but strong income growth? Simple rules fail. An AI agent can consider multiple factors and make better decisions. This is where agents add value.

But most organizations are not at that level of sophistication. They are still struggling with basic workflow automation. Do that first.

The Build vs Buy vs Configure Decision

Here is how I think about it:

Can you solve it with simple automation (Zapier, Make.com, spreadsheet formulas)? Do that. Weeks to implement. Thousands to invest.

Can you solve it with a configurable software package (HubSpot, Shopify, accounting software)? Do that. Buy it from a vendor. Let them maintain it.

Only if the answer to both is no, do you consider building something custom. And you consider an agent only after you have tried everything else.

The Most Valuable Automation Project Is the Most Boring One

I will leave you with this observation: the most valuable automation projects I have been part of are the most boring ones. Nobody gets excited about automating expense reporting. Nobody writes case studies about it. But it saves companies hundreds of thousands of pounds.

The flashy stuff (AI agents, machine learning models) gets the attention. But if your workflows are broken, flashy technology makes things worse.

Do the boring work first. Automate your workflows. Once they are running smoothly, then experiment with agents if you need to.

Richard Batt has delivered 120+ AI and automation projects across 15+ industries. He helps businesses deploy AI that actually works, with battle-tested tools, templates, and implementation roadmaps. Featured in InfoWorld and WSJ.

Frequently Asked Questions

How long does it take to implement AI automation in a small business?

Most single-process automations take 1-5 days to implement and start delivering ROI within 30-90 days. Complex multi-system integrations take 2-8 weeks. The key is starting with one well-defined process, proving the value, then expanding.

Do I need technical skills to automate business processes?

Not for most automations. Tools like Zapier, Make.com, and N8N use visual builders that require no coding. About 80% of small business automation can be done without a developer. For the remaining 20%, you need someone comfortable with APIs and basic scripting.

Where should a business start with AI implementation?

Start with a process audit. Identify tasks that are high-volume, rule-based, and time-consuming. The best first automation is one that saves measurable time within 30 days. Across 120+ projects, the highest-ROI starting points are usually customer onboarding, invoice processing, and report generation.

How do I calculate ROI on an AI investment?

Measure the hours spent on the process before automation, multiply by fully loaded hourly cost, then subtract the tool cost. Most small business automations cost £50-500/month and save 5-20 hours per week. That typically means 300-1000% ROI in year one.

Which AI tools are best for business use in 2026?

It depends on the use case. For content and communication, Claude and ChatGPT lead. For data analysis, Gemini and GPT work well with spreadsheets. For automation, Zapier, Make.com, and N8N connect AI to your existing tools. The best tool is the one your team will actually use and maintain.

Put This Into Practice

I use versions of these approaches with my clients every week. The full templates, prompts, and implementation guides, covering the edge cases and variations you will hit in practice, are available inside the AI Ops Vault. It is your AI department for $97/month.

Want a personalised implementation plan first? Book your AI Roadmap session and I will map the fastest path from where you are now to working AI automation.