“AI will deflect our clients.”
Everybody who’s within the assist trade has both heard a colleague say this or has thought-about it themselves. When the assist neighborhood began leaping on the AI bandwagon, the primary focus was — and infrequently nonetheless is — how the assist demand might be lowered. How can we do extra with fewer individuals? The important thing metric turns into “deflection.”
The implication of this deflection strategy is that this can be very company-centered. On this context, AI is the trail to decreasing assist value. After all, each firm can be glad to maintain its prices low, however I might argue that for assist, this shouldn’t be the first focus.
When deflection turns into the principle purpose, buyer interactions are a fallback — solely for use when completely crucial. The assist group out of the blue turns into the uncommon useful resource that we have to defend from these unwelcome interactions.
AI assistants develop into the henchmen who do the soiled work, sending clients down rabbit holes and limitless loops. Assist efficiency metrics are nice, so long as the assist groups don’t have to have interaction.
After all, that is an exaggerated picture, but it surely represents a really dominant assist narrative. At WooCommerce, we additionally purchased into it at first. Once we first applied our AI instruments, our first metric was deflection.
Nonetheless, as quickly as we requested ourselves the query of what we actually wished to attain with our AI instruments, it clicked that deflection despatched us down the mistaken path. Now, we at all times ask a query extra central to our personal concept of assist: does it make the shopper expertise higher?
That is the place we would like our true AI purpose to lie: not as a strategy to make it tougher to contact us, however as a means to assist our clients sooner and extra successfully. Each time we’re evaluating our use of AI options, we at all times come again to that query: does it make the shopper expertise higher?
Our concentrate on the shopper expertise has resulted in a number of strategic adjustments in how we offer assist.
Conducting high quality assessments and steady enchancment
First, now we have ongoing opinions of the standard of our AI replies. On a weekly foundation, we evaluate no less than 250 AI replies (roughly 20% of the amount we ship out). Everybody in our assist group will get concerned with these opinions, so AI reply high quality is repeatedly improved with a big number of views and areas of experience.
We evaluate how properly our AI software categorises the assist request and the way properly it selects sources, and we consider the precise replies it provides. This permits us to often tweak our prompting, sources, and configuration.
And it’s paying off. Once we began this over a yr in the past, we determined to evaluate and price the AI assistants primarily based on how useful and problem-appropriate their responses had been. We began at round 60% helpfulness. As of April 2025, our AI responses are useful over 90% of the time.
Our group rigorously opinions suggestions that we obtain about our AI-powered replies and the shopper expertise in coping with our AI assistants. Whereas we can’t at all times make prompt adjustments, we analyse the suggestions and make knowledgeable choices primarily based on this evaluation.
For instance: When clients submit a assist request, the primary reply they obtain is an AI-generated message containing ideas and attainable resolutions to the request. Clients indicated that it was not clear they may merely reply to the AI electronic mail to get in contact with a human. Consequently, we modified our template to make that extra clear. Whereas the quantity of instances this suggestions is given is now smaller, we are going to preserve tweaking our template till nobody mentions this anymore.
Rethinking assist sources and documentation
We’ve rethought our total strategy to public details about our product. We used to have over 1,000 inner paperwork on learn how to configure and troubleshoot our merchandise. Not solely had this develop into tough to course of for our group members (sure, we constructed an inner AI assistant for our inner docs), we additionally realized that shielding a lot of this data behind our assist group created an unnecessarily poor buyer expertise.
So, we reviewed each single a type of docs and shifted one of the best sources into our public documentation. The transfer empowered our clients to search out extra and higher data for themselves — and it broadened the variety of sources our AI assistant might pull high quality data from.
Among the shifted documentation contains:
Rigorously testing and implementing AI assistants
Earlier than launching a brand new assistant, we take care to check and iterate towards a variety of recognized assist situations. As soon as launched, we’re not afraid to disable AI assistants if the outcomes aren’t acceptable.
As an example, after implementing our AI answer in emails, we discovered that the automated replies for WooPayments questions weren’t assembly our requirements. We had been satisfied that sending out AI-generated emails for WooPayments would’ve created a worse buyer expertise than what we had on the time.
Consequently, we excluded AI replies from any WooPayments ticket, although WooPayments consists of ~25% of our whole quantity. If our foremost concern was to scale back the variety of assist requests to our group, this could’ve been an unwise strategic determination.
In early March 2025, we arrange a brand new AI assistant that targeted on these assist requests particularly. We’re lastly assured that our software can deal with the distinctiveness and the complexity of our WooPayments product.
Offering clear strains to human assist
We work with the fixed assumption that AI assistants aren’t excellent — very similar to none of us are both. So, we need to make it as straightforward as attainable for individuals to point that they need to attain out to a human. In our emails, we invite individuals to easily hit the reply button to speak with a human. On our assist web page, clients can contact us immediately.
Subsequent to that, we’ve additionally designed our assistants to err on the aspect of asking for clarification or handing inquiries to a human in the event that they aren’t certain. That system isn’t excellent, however in consequence, we’ve vastly lowered the variety of instances when considered one of our AI assistants very confidently provides the completely mistaken reply.
At all times specializing in bettering the shopper expertise
We’re comfy transport an imperfect answer. Once we launched the AI reply in our emails, the correctness of our AI assistant was not but the place we wished it to be. However the query we stored coming again to was: “Does it make the shopper expertise higher?”
Earlier than AI-powered electronic mail replies, we despatched out a static automated message that mentioned one thing alongside the strains of: “Hey, we’ve acquired your request and can get again to you inside 24 hours. When you wait, take a look at our docs [insert generic landing page] and our troubleshooting information [even though we weren’t sure whether you actually needed it].”
The e-mail AI assistant modified a largely ineffective reply to an prompt electronic mail that provided particular documentation and tried to reply your particular person questions. Throughout the time clients waited, our assistant provided a useful reply in additional than half of the instances it despatched one out. So, it wasn’t an ideal buyer expertise, but it surely was positively higher than what we had earlier than.
What’s subsequent?
Individuals throughout all industries, demographics, and geographies are getting used to processing data with the assistance of AI. A lot of our clients look to an AI assistant for assist with processing documentation on WooCommerce.com slightly than needing to seek for the correct reply themselves.
Within the first full week of 2025, we launched a conversational AI assistant on our assist web page to do exactly that. Our subsequent step is to carry this assistant to many extra locations on our web site: our bot will develop into your ever-present AI assistant to get assistance on WooCommerce.com.
Whereas it will initially be an AI-only answer, our part two purpose is offering a clean path to chatting with a human. We are going to ship the AI-only answer first, although, as a result of we consider it creates a greater buyer expertise than what we’ve had earlier than.
We’d love to listen to about your experiences! We evaluate every bit of suggestions that we obtain, and there are many methods to get in contact:
- Leaving suggestions in your assist interplay,
- Tagging me on the Woo neighborhood Slack,
- or commenting on this submit.
Tell us what you suppose and the way we might enhance your expertise. We are able to’t wait to listen to from you.
Job Thomas is the Chief Buyer Officer at WooCommerce, the place he and his group allow clients to thrive with Woo. Earlier than becoming a member of Woo in 2014, Job labored within the educational world, targeted on academic philosophy and curriculum improvement. He makes use of that background to empower clients to run their shops. Job is Belgian, however has been residing in Cape City, South Africa, for over a decade.
