How I build AI workflows
Most businesses do not need a big AI programme. They need one or two well-chosen workflows that take repetitive work off a team, built properly. This is how I approach that, at Herald and with smaller businesses: what is worth automating, what makes it work, and what I have learned along the way.
Background
Experience: AI at Herald
Herald is an AI-enabled placement platform that helps insurance brokerages and carriers write more business together. As Head of AI, I lead the AI behind it, including agents that work inside brokers’ email to extract client data and send submissions, and AI that extracts data from files and emails.
- Working inside the tools people already use, like email
- Extracting data from files and emails, checked before it is used
- End-to-end workflows rather than isolated demos
- Engineering for a demanding, regulated industry
Workflows I build most often
The best first projects are frequent, text-heavy tasks with a clear definition of done.
AI document processing
Invoices, forms, contracts and PDFs turned into clean, checked data in your systems.
AI email automation
An agent that reads every incoming email, sorts it, drafts the reply and turns requests into tasks.
AI knowledge assistant
Answers to your team's questions from your own procedures and documents, with sources.
Customer follow-ups
Quote follow-ups, reminders, review requests and re-activation messages sent at the right moment, in your tone.
Scheduling and requests
Booking, rescheduling and common questions handled by chat, email or phone, also outside office hours.
Reporting and data entry
Weekly reports, CRM updates and copy-paste work between tools, done automatically and checked.
What makes AI workflows work
Most AI projects that disappoint started too big. These are the principles I follow, at Herald and elsewhere.
Start with one workflow
Not a transformation programme. One task, one pilot, one clear number to compare.
Use the tools you already have
The AI works in the email, CRM, accounting software and agenda people already use, not in a new platform.
Keep a person in the loop
The AI drafts and proposes. A person approves anything that matters, until the workflow has proven reliable.
Measure before and after
Hours spent, response times and error rates, measured before the change, so the result is a number and not an impression.
Where I have built AI
Commercial insurance
As Head of AI at Herald, an AI-enabled placement platform for insurance brokerages and carriers.
Energy
At NOX Energy, which I co-founded: software that makes heat pumps grid-flexible, built from scratch.
Healthcare practices
Patient communication and practice admin, where data protection matters most.
Professional services
Accounting, legal, real estate and agencies: teams that handle a lot of documents, emails and repeat requests.
How I usually work
- 1
Understand the work
Look at how the task is done today, how often, by whom, and what a good result looks like. Often a simple automation is enough and no AI is needed.
- 2
Pick one workflow
Choose the task with the most repetitive hours and a clear definition of done, and measure it before changing anything.
- 3
Build and test on real cases
Build the workflow on real data and tools. At first a person approves the AI’s work, so nothing goes out unchecked.
- 4
Measure, then expand
Compare before and after. Let it run where it has proven reliable, then move on to the next workflow.
Frequently asked questions
What is an AI workflow?
An AI workflow is a business process where AI handles the steps that used to need a person to read, write or decide: reading an incoming email, extracting data from a document, drafting a reply. The rest of the process, like updating systems and sending messages, is automated around it, and a person approves what matters.
How do I implement AI in a small business?
Start with one frequent, text-heavy task, measure how long it takes today, and run a pilot where a person approves the AI’s work. When the numbers are good, let it run and move to the next workflow. My 6-step playbook on the blog explains each step.
How much does AI workflow automation cost?
Mostly the work of building it: connecting the tools, handling exceptions and testing on real cases. Running costs are usually small, often cents per email or document. I break this down in a separate article on what AI costs a small business.
How long does a first AI workflow take?
A focused pilot on one workflow can usually run on real work within a few weeks. Bigger integrations take longer.
What is the difference between an AI workflow and an AI agent?
An AI agent is the part that reads, decides and acts, using tools like your email, CRM or agenda. An AI workflow is the whole process around it: triggers, checks, approvals and integrations. Most useful setups are workflows with one or more agents inside.
Is my data safe? What about GDPR?
It should be designed with GDPR in mind from the start: data kept in the EU where possible, access limited to what the AI needs, and a person approving sensitive actions. The setup also needs documenting for your privacy records.
Which tools can an AI workflow connect to?
Most tools with an API or email access: Google Workspace, Microsoft 365, CRMs, ERPs, accounting software, agendas, Slack or Teams, databases and spreadsheets.
Related articles
How to implement AI in your business: a 6-step playbook for SMEs
Article · Playbook · 5 min read
How much does AI cost for a small business? What you actually pay for
Article · Costs · 4 min read
What building AI agents in insurance taught me about AI workflows
Article · From the field · 4 min read
Thinking about AI for your company, or working on something similar? I’m happy to talk it through. You can send me an email or book a call.