QuickBooks Online (QBO) has recently announced their new AI agents, which are designed to be act as personal assistants as you use the program.
This could be anything from categorizing transactions, recognizing and “resolving” anomalies, monitoring cash flow, analyzing financial data and forecasts, and “track every customer opportunity in the sales cycle”.
This is a huge ask on three fronts: depth of interaction, accuracy, and of course expectations.
Traditional Analysis vs Assisted Insight
I won’t speak to the AI assisted data entry features but will instead focus on the Insight Methodology that I have come to know in over 50 years of data mining and custom reporting experience.
This is what I have learned:
- Everyone uses QuickBooks differently and everybody views their data differently.
- Older business owners craft static views of their business after years of refinement.
- The metrics users look at can change over time, as the business changes
- Generational users of today look for “throw away” data that appears on demand as a tile with a single figure or quick comparison.
The current AI implementation in QuickBooks Online ostensibly contains LLM (Large Language Modules) for certain accounting transactions, data entry, and for analyzing certain sets of data.
Analyzing structured data fits in the current capabilities of AI and should serve the smaller QuickBooks customers well.
But for larger organizations, it certainly won’t replace the manner in which you have traditionally viewed your data metrics, for the reasons stated above.
Will it augment your experience? Yes.
Our Experience using AI
We have produced the most successful 3rd party applications for QuickBooks and we always look for ways to inculcate new technologies into our QQube Data Warehouse.
Over a year ago we explored AI to see how we could benefit from its promise – all with a skeptical eye. To date we have yet to implement the work we reaped from our investment.
That last statement might give you pause about QuickBooks and AI.
But let’s discuss the details.
- Using CHATGPT and several derivative applications, we find that the information is incorrect and/or inconsistent about 35% of the time. We have had to rip up what we have done – in some cases, dozens of times.
- It takes a while to learn how to ask the correct questions
- You MUST interact in small chunks. AI is just not capable of taking a wide swath of information, and report on it correctly.
- We HAVE gotten many important improvements and ideas from our inquiries.
- At the same time, we found that a portion of our own home-grown implementation was more efficient than what was suggested by AI. This alone led to frustration and consternation.
Expectations for QuickBooks AI
AI for QuickBooks is a TOOL. It is not the definitive answer to analysis as one size does not fit all. I remember opening up Excel when CoPilot was first introduced and saw the warning “Results May Not Be Accurate”. What is the point?
AI is a fact of life, whether for QuickBooks or any other technological and life practice implementation.
Our advice, from our own experience, is to ask and get suggestions, but not to take everything as gospel. Start with small pieces and requests, and learn what works and what doesn’t. As Rambo said in his movie: “I’ve Always Believed the Mind is the Best Weapon”.
