1
00:00:03,680 --> 00:00:09,358
Domain experts often collaborate with AI experts 
to realize their AI application ideas.

2
00:00:09,358 --> 00:00:13,727
These client-AI expert collaborations have become increasingly widespread.

3
00:00:13,727 --> 00:00:22,266
Client-AI expert collaboration begins with the client, who outlines their needs, expectations, and available resources through pre-collaboration planning.

4
00:00:22,266 --> 00:00:25,920
The pre-collaboration plan forms the basis for the discussion with

5
00:00:25,920 --> 00:00:31,914
AI experts during which the stakeholders 
collaboratively iterate the plan into a concrete and actionable one.

6
00:00:31,914 --> 00:00:37,445
The experience during this iteration heavily depends on the quality of the pre-collaboration plan.

7
00:00:37,445 --> 00:00:43,168
But because the client often lacks knowledge 
about AI and about the information needs of the AI expert

8
00:00:43,168 --> 00:00:47,780
coming up with a well-defined pre-collaboration plan 
is often challenging for the client.

9
00:00:47,780 --> 00:00:53,249
It can be mentally taxing and can even lead to misinterpretation of intents.

10
00:00:53,249 --> 00:00:59,200
To address this issue, we introduced PlanTogether, 
a system that provides guidance tailored to

11
00:00:59,200 --> 00:01:02,065
each client as they outline a pre-collaboration plan.

12
00:01:02,065 --> 00:01:06,876
For an illustration of how the system can aid a client 
during pre-collaboration planning

13
00:01:06,876 --> 00:01:10,160
let's imagine Sarah, a manager at a used car dealership

14
00:01:10,160 --> 00:01:13,294
who wants to avoid overpaying when purchasing used cars.

15
00:01:13,294 --> 00:01:16,499
She decides to hire an AI expert to solve her problem

16
00:01:16,499 --> 00:01:23,206
and uses PlanTogether to develop and concretize 
her AI application idea before entering discussions with AI expert.

17
00:01:23,206 --> 00:01:27,518
Sarah opens up PlanTogether and starts with the ‘Project Objective’ section.

18
00:01:27,518 --> 00:01:31,393
She notices the overview panel on top, which starts empty.

19
00:01:31,393 --> 00:01:39,588
The system first asks her about ‘Task Performed by AI Applications: 
“Please explain the task the AI application needs to perform.”

20
00:01:39,588 --> 00:01:44,050
But she finds out she cannot answer the 
question without any tips and suggestions

21
00:01:44,050 --> 00:01:48,461
so she looks at the questions that can help her with 
this question and moves to 'End Goal'.

22
00:01:48,461 --> 00:01:53,706
Sarah now sees the question “Describe the end goal of your AI application.”

23
00:01:53,706 --> 00:01:59,994
For this question, she types in “Make better decisions on used car purchases.”

24
00:01:59,994 --> 00:02:05,157
Afterward, Sarah wants to confirm 
whether her answer meets the intentions of the question.

25
00:02:05,157 --> 00:02:13,843
Sarah clicks on the AI chatbot and asks, “Would my answer for End Goal be helpful for AI experts in charge of AI application development?”

26
00:02:13,843 --> 00:02:20,204
The chatbot responds, 
“Instead of simply stating ‘Make a better decision on the used car purchases,’

27
00:02:20,204 --> 00:02:26,891
clarify what the better decision indicates to suggest 
how the AI application can improve the decision making.”

28
00:02:26,891 --> 00:02:31,045
Based on the response, Sarah modifies her answer to

29
00:02:31,045 --> 00:02:35,091
“Avoid overpaying when purchasing used cars from their previous owners.”

30
00:02:35,091 --> 00:02:38,505
Then, she clicks on “Save & Continue” to move on.

31
00:02:38,505 --> 00:02:43,477
Sarah eventually returns to the earlier question about 
'Task Performed by AI Application'.

32
00:02:43,477 --> 00:02:47,271
This time, she sees the tip for answering the question

33
00:02:47,271 --> 00:02:54,937
“Focus on how the AI model will enhance decision-making 
in used car purchases by evaluating price accuracy and negotiation outcomes.”

34
00:02:54,937 --> 00:03:00,000
with a blue highlight on the phrase: 
“enhance decision-making in used car purchases.”

35
00:03:00,000 --> 00:03:06,158
Hovering over the phrase, she sees that this tip draws information from the previously answered ‘End Goal’

36
00:03:06,158 --> 00:03:09,733
telling her the relevance of her previous answer to that question.

37
00:03:09,733 --> 00:03:15,443
She clicks on the downward arrow button 
and obtains answer suggestions tailored to the current plan.

38
00:03:15,443 --> 00:03:20,524
Reading the tip and the suggestion, 
she decides to accept the second suggestion

39
00:03:20,524 --> 00:03:28,371
“Predict accurate used car prices to ensure used car buyers do not overpay by comparing historical sale data and negotiation outcomes.”

40
00:03:28,371 --> 00:03:30,000
and she continues.

41
00:03:30,000 --> 00:03:33,313
Sarah proceeds to the question about ‘Domain Metrics’.

42
00:03:33,313 --> 00:03:40,039
She is unfamiliar with the notion of ‘domain metrics’ 
and initially lacks ideas about how to approach this question.

43
00:03:40,039 --> 00:03:42,987
Sarah sees a tip for answering the question

44
00:03:42,987 --> 00:03:48,411
“Focus on how you measure the 
used car dealer’s decision-making performance in used car purchases.”

45
00:03:48,411 --> 00:03:52,922
Hovering over the highlighted phrase “used car dealer’s decision making”

46
00:03:52,922 --> 00:03:57,964
Sarah sees a reference to her previous answer for 
‘Target User Groups & Demographics’.

47
00:03:57,964 --> 00:04:03,594
Similarly, she hovers over “used car purchases” 
to review her answer for ‘End Goal’.

48
00:04:03,594 --> 00:04:07,269
She re-reads the question in the context of her previous answers.

49
00:04:07,269 --> 00:04:13,907
After some thought, she decides to get concrete examples 
by clicking on the expand button and sees the suggestions

50
00:04:13,907 --> 00:04:17,194
“Cost to Market” and “Profit Margin.”

51
00:04:17,194 --> 00:04:21,894
Through the examples, 
Sarah understands how she should approach this problem.

52
00:04:21,894 --> 00:04:25,438
Sarah decides to double-check her understanding through web search.

53
00:04:25,438 --> 00:04:29,840
She finds the definition “custom metrics that are specific to your product”

54
00:04:29,840 --> 00:04:35,661
and the importance of capturing user expectations and experience, 
which confirms her understanding.

55
00:04:35,661 --> 00:04:40,532
Knowing that maximizing profit margins is the 
end goal of the target users

56
00:04:40,532 --> 00:04:48,078
and since “price gap” aligns with the problem of overpayment, 
she accepts the suggested answer, “Profit Margin,” and moves on.

57
00:04:48,078 --> 00:04:53,091
To highlight the improvement of client experience, 
let's look at the case of Rick

58
00:04:53,091 --> 00:04:56,965
a used car dealership manager in the same situation as Sarah.

59
00:04:56,965 --> 00:05:03,314
But instead of using PlanTogether, 
he uses a web survey without personalized tips and suggestions.

60
00:05:03,314 --> 00:05:09,383
Rick also makes it to the question about ‘Domain Metrics’, 
unfamiliar with the notion of ‘domain metrics’.

61
00:05:09,383 --> 00:05:12,827
He opens up a search engine and searches for “domain metric”

62
00:05:12,827 --> 00:05:19,106
which eventually leads to the same definition that Sarah saw:
 “custom metrics that are specific to your product”

63
00:05:19,106 --> 00:05:24,828
However, Rick has difficulty linking the information and his situation 
into a coherent answer

64
00:05:24,828 --> 00:05:28,195
he wants to see some examples to get him started.

65
00:05:28,195 --> 00:05:34,413
He searches for “domain metric examples” 
and finds examples in domains such as marketing and education.

66
00:05:34,413 --> 00:05:40,023
Still, because the problem of used car purchases is specific 
and relatively uncommon

67
00:05:40,023 --> 00:05:45,000
Rick finds it difficult to link his problem to the examples to formulate an answer.

68
00:05:45,000 --> 00:05:50,119
Rick ends up putting in a lot of time and effort before arriving at an answer.

69
00:05:50,119 --> 00:05:54,130
Now let’s take a look at how PlanTogether works.

70
00:05:54,130 --> 00:06:00,322
PlanTogether represents information dependencies 
within the pre-collaboration plan as a Planning Information Graph.

71
00:06:00,322 --> 00:06:08,774
It uses this graph to extract and link relevant questions and information 
to help the client clearly see and leverage connections between information.

72
00:06:08,774 --> 00:06:11,468
Based on this Planning Information Graph

73
00:06:11,468 --> 00:06:17,708
the guidance generator presents personalized tips and suggestions 
for each node question and overview summaries.

74
00:06:17,708 --> 00:06:24,483
Through a user study, we confirm that the tips and suggestions 
as well as the progression overview included in PlanTogether

75
00:06:24,483 --> 00:06:30,444
help clients navigate information dependencies 
and write concrete actionable plans reflecting their domain expertise.