Five Questions to Ask Before You Build a Chatbot

Jun 15, 2026

Posted by: Craig Szelestowski

Black-and-white retro-style photograph of a robot at a desk, diligently filling out an endless paper form that unrolls off the desk and coils onto the floor. Beside it sits a small, untouched "Simple Request Form" with just four boxes. A wall sign reads "Form Processing Division." Vintage office details include venetian blinds, rolled blueprints, a rotary phone, and binders labeled "Process Optimization" and "Systems Analysis."

Image generated by ChatGPT.

Chatbots are the new shiny object. Many process improvement presentations I’ve sat through lately include a recommendation for a chatbot somewhere. “Guide the client through intake”. “Answer FAQs”. “Validate form fields”.

Sometimes that's the right answer. Often it isn't. Here's how to tell the difference.

Before your team invests in a chatbot, work through these five questions in order. The first four are about form redesign. Only the fifth is technology.

This post focuses on chatbots assisting end-users/clients in filling out forms because it is the most common AI scenario I hear leadership groups asking about. However, the same five questions apply to any AI deployment: document processing, data classification, meeting summarization, etc. Substitute your AI use case for 'chatbot' and the questions still hold.

Simple line-art icon of a tangled scribble with arrows pointing outward in multiple directions and a question mark, representing confusion or an unclear process.

1. Why is this form so complex?

Is the complexity essential? Are all the fields required for regulatory or legal reasons? Or is it simply accumulated form design debt that nobody has cleaned up in years? If a form grew to be complicated because people kept adding fields without removing old ones, the answer isn't a chatbot. It's a cleanup.

Simple line-art icon of a pair of scissors cutting a coiled red ribbon, representing cutting through red tape or unnecessary complexity.

2. Which fields can be eliminated entirely?

If nobody uses the output of a field on a form, remove the field. Don't build technology to help users fill in information that goes nowhere.

Simple line-art icon of a desktop computer tower and monitor with multiple arrows and dotted lines branching outward in different directions, representing information flowing or scattering out to many places.

3. Which information do we already have?

If the organization already knows the department code, the cost centre, and the job classification from last time—pre-populate it in the form or omit the fields altogether. Don't ask the client to provide what you already know.

Simple line-art icon of two speech bubbles connected by an arrow: the left bubble contains a tangled scribble, the right bubble contains the letters "A" and "Z," with a pencil above, representing messy text being rewritten into clear language.

4. Can we rewrite in plain language? Make it more intuitive?

Here's my favourite test: if a field needs a chatbot to explain what it means, the instructions—or even the label itself—are poorly written, not poorly understood. Rewrite the field. Then test it with real users and refine it until first-timers can get it right without help 19 times out of 20. You might find that you do not need the chatbot at all (which your users/customers probably dislike anyhow!)

Simple line-art icon of a robot/chatbot head with a speech bubble containing three dots, representing a conversation in progress.

5. What's left after questions 1-4? That's where AI belongs.

If there are required fields that are genuinely complex and expert guidance would add real value to the end-user’s experience, build the chatbot. But build it for the simplified form, not the original mess.

The organizations that get this right save resources twice: once by not building unnecessary technology, and again by having a simpler process that's cheaper to maintain and improve. The ones that skip to question 5 end up with a chatbot that protects the very complexity they should have removed in the first place.

This five-point-checklist is also a great tool to use before piloting any solution experiment. Try it out!

Key takeaways:

The first four questions are about form redesign. Only the fifth is AI.
If you can't answer questions 1-4, you're not ready for question 5.
A field that needs a chatbot to explain it is a badly written field.
Simplify for the client. Don't add technology to compensate for complexity they shouldn't face.
The cheapest chatbot is the one you didn't need to build.

These themes are covered in more depth in our virtual two-day workshop, Practical AI for Process Improvement Specialists. If you're an improvement practitioner figuring out where AI fits in your work and your method, this course is designed for you. The next delivery is October 14 - 15, 2026 — registration is open now.

Craig Szelestowski is a recovered executive, the founder of Lean Agility Inc., an instructor at the Telfer Centre for Executive Leadership, University of Ottawa, and a Subject Matter Expert at the New Jersey Institute of Technology.

LEARN MORE

These themes are covered in our virtual two-day workshop, Practical AI for Process Improvement Specialists. The next delivery is coming up:

Oct. 14 - 15, 2026

If you're a Lean practitioner wondering where AI fits in your processes and your own method, this is the course.

Navigation

Information

1 - 95 Guigues Avenue
Ottawa, Ontario
K1N 5H8

T: 613-266-4653