Putting AI into a process worth improving.
An AI demonstration can be convincing in a few minutes. A useful business process has to survive incomplete records, unusual requests, changing responsibilities and an ordinary Tuesday when the expert is away.
That is where I would put the effort in 2024: taking a promising experiment and making it dependable inside a real customer process.
Context changes the quality of the answer
Salesforce's September 2023 announcement of the Einstein 1 Platform brings its data and AI work together around the platform's metadata framework. The direction matters because a business answer depends on the meaning and permissions of the information behind it.
A draft service reply needs more than the customer's latest message. It may need the relevant product, the current case history and an approved policy. If any of those are outdated, a polished reply can still be wrong.
Choose the sources deliberately. Name the people who maintain them. Make it possible for the reviewer to see what the proposed answer relies on. When the necessary information is absent, the process should make that absence visible.
Design the whole task
Map what happens before and after the AI step. Who selects the case? Who checks the draft? Where is the approved response saved? What happens when the draft cannot be used?
The answer may be a modest change to a service workflow rather than a new interface. Keeping the work close to the record can save more effort than generating a faster paragraph that somebody then has to copy, check and file manually.
Count the review work too
Measure the time needed to finish the task, including corrections. Track whether customers need to contact the team again. Compare results across ordinary and difficult cases, rather than judging the process on the best examples.
Keep a small regression set of representative requests and repeat it when instructions, source material or configuration change. Assign an owner who can pause the feature if quality falls.
My priority for 2024 would be one process where better context and a carefully placed AI step create a result the team can defend. That gives leadership something concrete to fund, improve or stop.
Context: Salesforce's September 2023 Einstein 1 Platform announcement.
The annual outlook series
2019 · 2020 · 2021 · 2022 · 2023 · 2024 · 2025 · 2026
Annual series edition: 5 January. This edition was prepared or revised on 8 September 2026.
