AI-based forecasting in IBM Planning Analytics
Leading research and design to deliver a minimum viable experience
Project context and goals
Project context and goals
For a financial plan manager, generating forecasts and budgets is typically a tedious and manual process that involves accessing multiple data assets across multiple systems to estimate forecasts.
​
The goal of this project was to augment users' forecasting tasks by providing statistically accurate forecasts as a starting point.
The business goal in this endeavour was to gain competitive advantage for our product, IBM Planning Analytics, by providing an integrated AI-driven forecasting capability.
My role
My role
Lead designer and researcher
Team
Visual Designer, Product Manager, Development Manager, Development Architect, 6 Developers
Product
IBM Planning Analytics - a financial and operational planning​ software
Preview chart for forecast
Initial research and early insights
Through a quick competitive analysis, we realized most competitors did not offer a quick integrated forecasting capability.
We interviewed 6-8 customers to map their as-is forecasting process and uncover opportunities for improvement.
-
Smaller to medium-sized companies lack statistical and data science expertise, and rely on estimates for forecasts
​
-
Larger companies tend to have in-house solutions to support statistically accurate forecasts
​
-
Most customers forecast bottom-up, instead of top-down, implying that a quick forecast could be valuable to most users
"Hope for something that gives us a baseline, using historical data. We spend a lot of time bringing data from Excel to Planning Analytics...with this I'm looking to improve our productivity more than accuracy [of forecast]"
​
- A small-sized company
Primary user need identified
User
Need
Outcome
Line of business managers
​
need a way to apply statistical algorithms to their forecast,
​
so that the accuracy of their forecast improves and they finish forecasting on schedule.
Requirements scoping workshop for first release (MVP) with project stakeholders
MVP ideation
The cross-disciplinary team agreed that our first release would focus on providing a quick and statistically accurate forecast to our users. Together we solidified the concept through several iterations and created a prototype.
Feedback for prototype
Over 2 conference workshops with upto 70 participants, a remote customer-group session with 13 participants and 5 one-on-one interviews, we improved the prototype based on user feedback.
"I like forecasting at the click of a button”
"I wish it would tell me how forecast was calculated.. in plain language"
"I wish we could ignore some data"
"Will the average user understand [statistical details]?"
User feedback from workshop
IBM Planning Analytics workshop for AI-forecasting at IBM Data and AI Forum - Miami, 2019
Survey results
Final designs for first release
Using historical data, user runs an AI-based forecast for two items
Usability tests
75
SUS score
I planned, recruited and conducted usability tests with 10 customers to evaluate the MVP offering before it was released.
​
Smaller usability issues were fixed by the team right away, while the larger issues and enhancements have shaped the roadmap.
Retrospective
With the market-trends and IBM's AI focus, it was critical to get this capability into our product early. However, with misalignment of goals, and organizational changes, it was a challenge for the team to collaborate effectively and deliver rapidly. Research findings played a key role on this project in resolving conflicts and grounded the decisions firmly in user needs.
​
This project was certainly a learning exercise of building mutual trust between Design and Development. We worked together on solutions to communicate more effectively and truly collaborate to deliver the best experience to our users.
​
Next steps for this project are to expand this forecasting capability beyond the minimum viable experience.