How your assistant remembers

When a customer writes “yes, the second one” or just “no”, your assistant has to know what they are answering. That sounds like one thing, but it is really two, and they behave differently: what was said a moment ago in this conversation, and what you already know about this person from every conversation before it.

Both are on by default. There is nothing to set up for the first one, and the second one is a single switch.

Your assistant follows the conversation it is having on every plan. The switch described below controls whether it also recalls earlier conversations.

---

The conversation it is having

Your assistant reads back over the recent messages in the thread before it answers. So a multi-step exchange works the way a person would expect:

Who	Message
Your assistant	”…the substitute cannot be from your own classification. Do you have one lined up for the next time you travel?”
Customer	”no”
Your assistant	”No problem — would you like me to suggest a few members who have subbed before?”

It picks up whatever it can see: the question it just asked, the three options it just offered, the booking details collected so far. This is what lets it ask for a date, then a time, then a name, without starting over each message.

It works the same on every channel — WhatsApp, Instagram, Messenger, your website chat — because all of them run the same assistant. A voice call follows its own conversation the same way, for the length of the call.

A few things it deliberately does not count as part of the conversation:

• A message that failed to send. Your customer never saw it, so your assistant does not act as though they did.
• Anything you have had deleted. Messages removed under your data-retention window, or by a deletion request, are gone from its memory too — see what we delete.
• A very old thread — sometimes. It depends on whether it has a summary of the customer; see where the two meet below.

Replies you sent by hand count. If you took a conversation over from the Inbox and promised a customer a discount, your assistant can see that when it picks the thread back up — so it will not contradict you.

---

What it knows about the customer

The second kind of memory is a short, plain-language summary of everything that person has ever said to you, kept per customer rather than per conversation. It is what lets a returning customer say “has my order shipped?” on WhatsApp when they placed the order on a phone call.

You can read it yourself: open Customer 360, choose a person, and it is the Customer memory panel at the top of their page. It is written by your assistant after conversations, not by you.

This is the part controlled by the Customer memory switch under Features on the Config screen:

• On — your assistant recalls earlier conversations with this person, on any channel.
• Off — it still follows the conversation it is having, but each new conversation starts without any history of the person. Your uploaded documents are searched either way.

Telling it something yourself

If you need your assistant to know something now — “offer up to 10% on the annual plan until the 31st” — you do not have to wait for it to work that out. Add a note to the deal and set it to Visible to AI rather than Private note, and it applies from their very next message. Notes with an expiry stop applying on their own, so “until the 31st” needs no follow-up from you. See your knowledge base for the documents side of this.

---

Where the two meet

When a thread has been quiet for days, your assistant does not resume it mid-sentence — but what happens next depends on whether it has a summary of that person, and the two cases are genuinely different:

	It has a summary of the customer	It does not
When	Customer memory is on and the person has talked to you before	The switch is off, or this is a brand-new customer whose first conversation has not been summarised yet
A dormant thread	Starts fresh — it leans on the summary plus the last message or two, rather than picking up a three-week-old abandoned booking	Carries on from where the thread left off

That second row is deliberate. With no summary, the recent messages are the only memory your assistant has, and trimming the only memory it has would leave it answering blind. So it keeps them.

The practical consequence: if a customer picks up a month-old thread and your assistant seems to remember more than you expected, that is this — and it is usually harmless. If you would rather it started clean every time, turn Customer memory on; it gives your assistant something better to fall back to.

---

Testing it

The Test the chat bot panel on a bot’s page follows the conversation exactly as a real customer’s would, so you can walk a multi-step flow all the way through. Clear starts a fresh conversation.

To see how a returning customer would be greeted, open Add returning-customer memory and paste in a summary — that is the second kind of memory, supplied by hand, and nothing you type there is saved to any real customer.

---

Troubleshooting

It answered my follow-up as though it were a new conversation

If the messages were days apart and Customer memory is on, this is deliberate — a dormant thread starts fresh rather than resuming (above). Send the question with its context in one message.

If the messages were minutes apart, check whether the earlier ones actually reached Kooday — open the conversation in the Inbox and confirm they are there. Your assistant remembers what is in the thread; a message that never arrived is not in it.

It repeated its greeting mid-conversation

Your assistant greets people it does not recognise. If it is greeting a customer you know it should recognise, the problem is identity, not memory — the number they messaged from is probably not the number on their record. Open the person in Customer 360 and check which phone number is stored.

It does not remember a customer between conversations

Turn on Customer memory under Features on the Config screen. With it off, each new conversation starts without any history of the person.

Note that the summary is written after a conversation, so the very first exchange with a brand-new customer has nothing to recall yet.

It knows something about a customer I want removed

Your assistant keeps no separate copy of anything — its memory is the messages themselves, so removing those removes it from the memory too. Messages age out under your retention window automatically; for a specific deletion request, follow how to delete your data.

If what you want is not removal but a correction — the customer’s situation has changed — the faster route is to tell your assistant directly with a note, as above. A current instruction outranks a stale summary.