Prescient AI reference
Chat reference
Reference for the Prescient AI chat, with the topics that an answer can read, the header buttons, the status lines, the messages, and each limit.
This page lists the parts of the Prescient AI screen and of a chat, with their limits. Prescient AI is the first item of the product sidebar.
Data that an answer reads
Section titled “Data that an answer reads”An answer reads the data of the workspace. The table lists the kinds of data.
| Topic | What an answer can use |
|---|---|
| Knowledge base | How the product and the models work. An answer from the knowledge base lists its articles under Sources. |
| Models, channels, and tactics | Which models, channels, and tactics your workspace has. |
| Reported performance | The figures that your sales sources and ad platforms report, as totals or over time. |
| Modeled attribution | The outcome that your models attribute to paid media, as totals or over time, by channel, tactic, or campaign. |
| Base and Halo | How the modeled outcome of each channel divides into Base and Halo. |
| Saturation | The saturation curve of one channel, tactic, or campaign. |
| Model fit and model health | How closely a model follows the reported totals, and whether each model is up to date. |
| Data health | Whether the data of each connection is up to date. |
| Media plans | Your Media Forecaster plans and the results of one plan. |
| A new plan | A proposal for a Media Forecaster plan, which you can change and run in the chat. |
Prescient AI writes each answer for the question, so the docs cannot list each possible answer.
The Prescient AI screen
Section titled “The Prescient AI screen”The rules for the greeting, the recommendation card, and the reply pills.
| Rule | Value |
|---|---|
| Rows on the recommendation card | Up to 3 rows that gain from more spend, and 1 row with the label biggest cut. On a short screen the card shows 2 rows that gain. |
| Recommendation card | The card shows only when the daily plan found a positive gain and one or more channels gain from more spend. |
| Period of the recommendation | The next 4 weeks when your models are weekly. The next 30 days when your models are daily. |
| Build this plan | You can use Build this plan one time in each chat. |
| Last week pill | The pill shows the channel or tactic whose modeled outcome changed most in absolute terms. A row needs a change of 0.5% or more. |
| Last week, daily models | The 7 days up to the last modeled day, compared with the 7 days before. |
| Last week, weekly models | The last complete model week, compared with the week before. |
| Connection with no fix on your side | A failure for which the product has no fix on your side gets a short note, not a row with buttons. The note shows after the failure has lasted 24 hours. A failure with no recorded start time shows at once. |
| Model that is behind | A daily model is behind when its last modeled day is 10 or more days old. A weekly model is behind at 21 or more days. |
| Screen after the first message | The greeting and the card stay at the top of the chat, without the reply pills, until you leave the page. A chat that you open again does not show them. |
Reply pills
Section titled “Reply pills”| Label | What it shows |
|---|---|
| Last week | The pill names one channel or tactic and the change of its modeled outcome from the week before. The pill sends a question about the cause of the change, not the text of its label. |
Buttons
Section titled “Buttons”| Button | What it does |
|---|---|
| Submit | Sends the message. |
| Stop | Stops the answer. |
| History | Opens the list Recent chats. |
| New chat | Opens a new chat with the Prescient AI screen. |
| Save to library | Saves the conversation as it is now to the Library. |
The message box also has a model picker. The first option is the default.
Status lines
Section titled “Status lines”| Status | What it means |
|---|---|
| Thinking… | Prescient AI is preparing the answer. |
| Working… | Prescient AI is working, and no step or text is on the screen yet. When a step starts, a line for the step shows in its place, for example Checking your attribution…. |
Limits
Section titled “Limits”| Rule | Value |
|---|---|
| Message length | 8,000 characters. A message contains text only. |
| Message rate | 20 messages in 60 seconds for each person. |
| Time for one answer | 300 seconds. A longer answer stops with the message This answer didn’t finish. Ask again. |
| Date range of an answer | The last 90 days, unless your question names a period. |
| Longest period of an answer | About 2 years (731 days). An answer for a longer period covers 731 days. |
| Daily figures over time | Up to 92 days in one answer. |
| Recent chats | The 50 chats with the most recent activity. The page reads the list when it loads, so a new chat shows in the list after the next load of the page. |
| Sending while a plan card is prepared | After you select Build this plan, the message box reads Creating your plan… until the plan card is ready. You can type, and you cannot send. |
| Chat title | The first 80 characters of your first message. |
| Who can open a chat | Only the person who started the chat. To show a chat to your team, save it to the Library as a shared item. |
| Controls that a chat does not have | A chat has no control to rename the chat, delete the chat, copy an answer, or rate an answer. |
Messages
Section titled “Messages”| Message | What to do |
|---|---|
| Something went wrong. Please try again. | Send the message again. This message also shows when a message is too long or when you send messages too fast. |
| This answer didn’t finish. Ask again. | The answer reached the time limit. Send the question again, or ask a narrower question. |
Related
Section titled “Related”- Prescient AI: the guides and the concepts of the product area.
- Ask a question: the steps to ask and to follow up.
- Find an earlier chat: the steps for the history.
- Where the daily recommendation comes from: the plan behind the recommendation card.