Most write-ups about Tidio Lyro explain what it is. This one assumes you have already decided to try it and now need it answering real tickets by Friday. That is a different job, and it comes down to a handful of steps that have surprisingly little to do with clicking around a dashboard.
Picture a skincare brand with six support agents handling about 450 conversations a week. Roughly 60% of those are order status, shipping times and returns. That is the ground Lyro should own. The rest, from ingredient complaints to wholesale enquiries, needs a clean route to a human. Get the split right and the setup takes a day. Get it wrong and you spend three months apologising to customers who got a confident, fluent, completely incorrect answer.
Sort your tickets into three buckets before you open the dashboard
Pull your last 200 tickets and write down the recurring ones. A typical ecommerce list looks like this:
- Where is my order? (usually with a tracking number pasted in)
- Can I change the delivery address after ordering?
- Do you ship to Ireland, and what does it cost?
- How do I return a sale item?
- Is this product safe for sensitive skin?
- Can I get an invoice with a VAT number?
- My bottle arrived cracked. This is the second time.
Now sort them into three buckets. Lyro-only is anything factual and repeatable: shipping costs, return windows, opening hours, order tracking links. Lyro-assisted is where it gathers information and hands over: a cracked bottle needs a photo, an order number and a human decision on the refund. Human-only covers complaints, legal threats, wholesale pricing and anything medical-adjacent.
Write that list down. It becomes the spec you check your setup against, and it stops the classic mistake of switching everything on at once and hoping.
Step 1: Connect only the channels you can actually staff
Lyro will happily sit in live chat, email, Messenger and Instagram. The temptation is to enable all of them on day one. Don’t. Every channel multiplies the ways a conversation can stall, and a customer who gets an AI answer at 11pm on Instagram and no follow-up until Tuesday is worse off than one who got no answer at all.
Start with live chat and email, where you already have someone watching. Internal handoff to Slack is worth setting up early, because that is where your team will actually see escalations. If you want a clear-eyed view of how Lyro behaves as a Slack support agent, read that before you wire it in, since the tone that works in a customer chat is not always the tone your team wants in their own workspace.
Step 2: Feed Lyro the content you’d hand a new hire
This is the step that decides whether the whole project works. Lyro answers from the sources you connect, so your answers are only as good as the pages behind them.
Point it at a help centre, not your entire website
A 30-page help centre with clean, single-topic articles beats 400 website pages every time. Marketing copy is written to persuade, not to answer, and it produces answers like “our generous 30-day policy” when the customer needs a date. If your help content lives inside a blog, consider rebuilding the top 15 articles into standalone questions before you connect anything.
Delete the contradictions first
Search your own site for the words “returns” and “shipping”. You will almost certainly find two pages disagreeing. One says 30 days, an old FAQ says 14, and a 2023 blog post mentions free returns you scrapped last spring. Lyro cannot know which one is current. It will pick one, and it will pick it consistently, in front of customers. Fix the source pages before training, not after.
Step 3: Write the handoff rules before you need them
Define the triggers that push a conversation to a person. Ours looked like this: any mention of a refund over £150, any message containing a chargeback or legal threat, any customer who has asked the same question twice, and any conversation where the order number cannot be found after two attempts.
Then decide what the human receives. A bare “customer needs help” notification is useless. Push across the transcript, the customer’s name and email, the order number if Lyro captured it, and one line summarising what was already tried. That single sentence saves your agent from asking the customer to repeat themselves, which is the fastest way to lose someone’s patience.
Step 4: Give Lyro a brief, not a vibe
“Be friendly and helpful” teaches nothing. Write rules an intern could follow on their first day: keep replies under three sentences, never use exclamation marks in a complaint thread, always confirm the order number before quoting a delivery date, and end with a question whenever you still need information. If you have ever worked through a step-by-step guide to getting better results from an AI assistant, the principle is identical: specific instructions outperform warm intentions every time.
Add one line that matters more than the rest: if you are not certain, say so and offer to find a colleague. A slightly hesitant answer costs you a little trust. A confident wrong answer about a delivery date costs you the order.
Step 5: Run a 20-ticket test before launch
Pull twenty real tickets from last month, a deliberate mix of easy, messy and unpleasant. Paste them into a test conversation one at a time and score each as resolved, escalated correctly, or wrong. Anything under 15 out of 20 means you go back to the content, not live.
Keep the failures in a list with a short note next to each: “quoted EU shipping as free”, “couldn’t find the returns page”. That list is your to-do sheet for the next two weeks, and it is far more useful than any dashboard score.
Step 6: Read transcripts every Monday for the first month
Thirty minutes, once a week, reading what Lyro actually said. You are looking for three recurring themes, not one-off oddities. In practice you will find things like customers asking about a product line that was discontinued, or a phrase your customers use that your help centre never mentions. Fix those three themes each week and the resolution rate climbs steadily.
This is also where you catch the failure that surveys miss: answers that are technically true but unhelpful. “Returns are accepted within 30 days” is correct and useless if the customer asked how to start one.
Where Lyro still slips
It handles structured, factual questions well and struggles with anything that requires reading between the lines. Sarcasm, multiple questions in one message, and conversations that jump topics mid-thread are all reliably weaker spots. There’s a fuller picture of what this AI support agent gets right and where it fumbles if you want to know which corners to reinforce with human cover.
The four numbers worth tracking after 30 days
- Deflection rate on your Lyro-only bucket. On FAQ-heavy stores, 30% to 50% is a realistic first-month range. Anything above 60% usually means it is closing conversations it should have escalated.
- Escalation accuracy: what share of handoffs genuinely needed a human. Below 70% and your triggers are too twitchy.
- First response time, split between Lyro and human conversations. This is the number your customers feel most.
- Repeat contact rate: how often the same customer comes back within 48 hours. A rising number is the clearest sign Lyro answered something it shouldn’t have.
If you only have time for one habit from all of this, make it the Monday transcript read. Every team that gets Lyro working well does that, and every team that quietly turns it off stopped doing it somewhere around week three.