top of page

From dead ends to clear paths: How I redesigned a chatbot fallback for BMO Assist

The challenge: Help users succeed on their first try

BMO launched a new chatbot, BMO Assist, to support everyday banking. With 760,000+ users and 1M+ sessions, traffic was strong. But half of those chats ended in failure.
 

When the bot wasn’t confident, it said: “I’m sorry, I didn’t understand that.”

That left customers stranded without guidance or next steps.

The problem: fallback wasn’t just a copy issue — it was a strategic gap

While the long-term plan was to improve intent recognition through training, I proposed a content-led solution that could help immediately:

​

“What if fallback wasn’t the end of the conversation—but a second chance?”

​

That’s how the fall-forward model was born.

The goal: measure outcomes, not just AI accuracy

We set clear targets to shift focus from precision to progress:

  • Raise the response success rate from 45% to at least 80%.

  • Reduce exits after fallback.

  • Improve positive feedback.

  • Define success in terms of customers completing tasks, not just the AI guessing right.

A poor user experience. The chatbot couldn't match a response even if it were close. 

My approach: turn fallback into fall-forward

I owned the content strategy for a new model that gave customers choices instead of dead ends.

  1. Create options, not apologies

    • Short acknowledgement → “Try one of these” → 2–4 clear CTAs → safe escalation.

  2. Design a scalable CTA system (300+ options):

    • Verb + noun → “Lock card”

    • Noun only → “Account balance”

    • Question → “Why am I receiving alerts?”

  3. Work across functions

    • With PM: set new KPIs.

    • With design: tested fit within UI.

    • With engineer/QA: spec’d behaviour and edge cases.

old experience.png
old experience (2).png

Real examples

share_361692831610448279 (1).png
share_1993223009169839150 (1).png
share_5974216662990745890.png

The results? Users didn’t just stop at fallback. They moved forward.

Within the first week of launching the Fall-Forward Model, the chatbot’s response success rate jumped from 45% to 86.5%—nearly doubling its effectiveness overnight.

13,150

Fall-forward responses invoked in the 1st week

90%

Increase in correct responses 

43%

 Responses contributed to total thumbs up

Why it matters: building trust in a new channel

For many people, BMO Assist was their first time trying digital self-serve with the bank. A poor fallback would have eroded confidence quickly. By replacing a dead end with clear choices, we made the channel feel more reliable and worth coming back to.
 

Lesson learned: Sometimes the fastest way to improve AI isn’t retraining — it’s better content strategy.

bottom of page