8 AI Terms I Love (part 1)
Here are a few AI terms that those actually doing the work will identify with and the wonderful rabbit holes they will take you down for insights to actions. While they aren’t necessarily technical terms, they apply to the different levels of conversation our teams have when discussing AI.
#1: SPEED. A quick mention to a word that helps drive our teams success. SPEED. Speed is the new leverage and I would also say Freedom. Since the days of Appirio the word that drove the teams to such success was SPEED. Go fast without fear, break things (without impacting the client or their customers of course), learn fast, develop fast (fast and clean), think fast (and smart), and fast is what Appirio delivered on. What a wild ride.
2. CORRELATION: Technique that shows how strongly and whether pairs of quantitative (measuring) variables are related. ie. Quantitative can be measured and analyzed vs Qualitative which is non-numerical. It indicates how changes in one variable correspond to changes in another, but it does not necessarily imply causation. (think weight to calorie intake - do the number of daily calories consumed and body weight have a relationship for X individual?)
3. HARMONIZATION: How many times have I heard this said within the RFP or pre-sales process. Here is a nice quick pull from AI: Data harmonization in technology is the process of collecting, transforming, and integrating data from various sources into a consistent, accurate, and meaningful unified format. It involves standardizing different data structures, formats, and terminologies to create a single, comparable, and usable dataset, thereby improving data quality, fostering better analysis and decision-making, and enabling seamless data sharing across systems and organizations. That is pretty much what we are talking about here. Great word.
4. INTEROPERABLE: Means that different information systems, software, devices, and applications can exchange, access, integrate, and cooperatively use data with a shared and consistent understanding of its context and meaning. Interoperable systems can seamlessly share and process data across organizational boundaries without requiring manual intervention, which leads to improved efficiency, better decision-making, and greater innovation.
5. INTENT ANALYSIS: This is an interesting one. It is what someone or something wants or means. It is the purpose or intention behind words. This is something that is popular with Chatbots for humans and IoT with Assets. What is this person, asset trying to tell me? Where do I take them (Case or Work Order? which record type? process A, B or C?) For humans you can use Sentiment Analysis which is, does this person express a positive or negative or neutral sentiment. For assets you need to set your sensors etc to provide your target with enough information to make a decision for action. No easy task as there is very often multiple IoT platforms feeding into the target.
6. GROUNDED: This is just a good concept to know for context. It’s about bridging the gap between the abstract and real world knowledge. It’s specific to AI LLMs and refers to the process of connecting the AI system's abstract knowledge to real-world information, contexts, and specific data sources. Add some human bias and data bias and you have a genuine conundrum. A quandary if you will, a real life dilemma. Have fun out there.
7. IDENTITY RESOLUTION: I chose to put this term at the end as it tends to get complex but I love the concept and delivering on the success of this requirement. Before diving into the identification resolution of your customers or assets, I would suggest visually mapping out your customers journey with clear milestones and their associated KPI mappings. This starts with the Vision, which is supported by the Guiding Principles and KPIs.
Guiding Principles may include: Customer centric approach, leveraging tech innovations, data insights, agile mindset. Pick your own but 5 components is the goal.
KPIs can be Operational (FTFR, MTTR, Travel Time, Tech Utilization Rate), Customer-Focused (repeat visit rates, CSAT, NPS), Financiual & Asset ($ per work order, Rev per tech, SC renewal rate), Miscellaneous (there may be grey areas like improved user experience that are harder to measure but you need to work these out for measures via your CoE team).
The customer journey may include the customers interactions from where they go for their information to make decisions and take action, data per interaction needed (per channel), data needed for audience segmentation, data quality per source. Somethings that has a lot of complexity regarding assets is IoT and pulling the right data from the asset to the target so that it can be consumed, processes and actionable.
I would also define what a customer 360 means to you, AND what it means to your customers? Do you know your customers from different sources? Can you segment your customers by their individual attributes? age, activity preferences (ie. weightlifting to running) and other bits of information allow AI to do its job. Segmentation of true data is key here so lets dive in a bit.
Process: Combine data from sources into a single profile based on user-identified resolution rules in a ruleset. (Rulesets covered below). A very basic example - but think source with first name(Tom) and email, second source with first name(Thomas) and email, third source full name and phone number but different email = unified individual. Add on complete order purchase history, service case history, marketing engagement, etc. Use resolution rulesets to match and reconciliation rules are added to profiles across various data streams to provide a unique ID (from all these multiple sources). You now have a Unified Profile.
Now take this concept and apply it to Assets. Hence my reference to rabbit holes. Work from the Asset to your customers. Now you are on your way to a true Customer 360 Model.
Note: I speak on this in the MDM article for companies that have 3+ source systems that hold customer data.
Other mentions to explore:
Formula Functions (4): Text manipulation, Type conversions, Data calculations, Logical expressions.
(FQK) Fully Qualified Keys: To avoid key conflicts when data from different sources are ingested and harmonized in a data cloud model. FQKs consist of Source Key & Key Qualifier.
8. RULESETS: Let’s revisit Rulesets. The Goal of Rulesets is to match data across multiple systems using match rules and reconciliation rules about a specific object such as an individual or asset. While match rules are used to link together data into a unified profile, Reconciliation Rules determine the logic for the data selection. (ex. if an email address is available from 2 sources, a reconciliation rule helps the unified profile know which one to display.)
Note: you may select reconciliation rules at object or field levels.
More to come, see you soon.