Book a demo Start free

Texa AI

Experiments

Try a changed procedure set on part of your live conversations, against the one you run now, and keep whichever does better.

Help centreUpdated on

An experiment splits new conversations between the procedure set an inbox runs now (A) and a changed set (B), so you can see which does better with real customers.

Experiments run on live conversations, so take an inbox live first. See Choose how Texa AI works in each inbox.

Start one

  1. Save your changed procedures as a new set and approve it. See Procedures.
  2. Go to Texa AI → Procedures → Experiments and choose New experiment.
  3. Give it a Name and write What you expect, so you can read the result against it later.
  4. Choose the Inboxes. They must run the same set, which becomes A.
  5. Choose Set B, and the Share of new conversations for B, such as 50%.
  6. Under Stop B by itself when, set the limits: handover rate higher by more than, CSAT lower by more than, and safety refusals higher by more than. These are checked every hour, once each arm has enough closed conversations.
  7. Start it.

Each conversation stays on its arm until it ends.

Read the results

For each arm, the page shows conversations, how many Texa AI resolved, how many were handed over, safety refusals and satisfaction. Results appear 7 days after the first conversations close, when they settle.

Finish

  • Finish, keep A: the inbox goes on with the set it ran.
  • Finish, switch to B: the inbox moves to the new set.

If B breaks one of your limits, it’s stopped by itself and every new conversation goes to A.

Still need help?

More in Texa AI

  • How Texa AI works What Texa AI does for your customers, where its answers come from, what it will never do, and how you stay in control.
  • The go-live checklist The steps the Overview checks before Texa AI answers customers on its own, and what to do for each one.
  • Texa AI's name, tone and languages Give Texa AI a name, a tone, a greeting and an email sign-off, choose its languages and voice, and set how long conversations are kept.
  • Choose how Texa AI works in each inbox Turn Texa AI off, on in shadow mode or live for each inbox, go live in steps, add testers, and set its procedures, languages and hours.
  • Hand-over routes Decide which team gets a conversation when Texa AI hands it over, with what priority, labels and ticket, and what happens if nobody answers in time.
  • Knowledge: teach Texa AI your answers Give Texa AI your FAQs, prices, fees and policies by typing them, uploading files, reading your website or using your help centre, then approve them.
  • The knowledge pack Tell Texa AI how your business works, what it is not, what it may and may never promise, and when it must hand to a person.
  • Knowledge gaps and suggestions See what Texa AI couldn't answer, write the missing answers, and approve improvements drafted from how your team answered.
  • Procedures Tell Texa AI how to handle each kind of request, in your own words or from conversations your team handled well, then check, approve and improve it.
  • Simulations Let AI-played customers talk to Texa AI in the sandbox, against your procedures and test answers, before a real customer does.
  • Actions: connect your systems Let Texa AI look things up and make changes in your own systems, with ready-made connectors or your own API, under rules you set.
  • Texa Relay for systems behind your firewall Let Texa AI look up records in your core banking or database without opening your network, through a relay that runs inside it.
  • Quality checks Run checks on every conversation, by words or in plain English, get alerts when something needs attention, and review Texa AI's conversations.
  • Safety Stop Texa AI at once with the kill switch, download or delete a customer's data, see where your data lives, and read every change made to Texa AI.
  • Texa AI inside your current helpdesk Keep Zendesk, Freshdesk, Intercom, HubSpot, Salesforce or Zoho, and add Texa AI to it as an agent of its own.