Often a client will ask me to review a deal (typically) that went south after the fact to see what can be gleaned from it for improvement or recovery.
Nobody asked me here.
Yet I became instantly captivated by the story of Bench, an AI-forward accounting behemoth which suddenly and completely imploded operations on December 27, 2024.
I think it was “data gluttony” that took it down.
What is data gluttony? How relates to ventures and AI, our accounting and legal industries even?
How other should we call it? It is important. "Drunk data"?
AI in many respects, at least the good ones, is pay-as-you-go software. More use, more cost.
Data gluttony (overuse of function by teams and resulting data/token costs) will kill many enterprises.
It already is/has.
It is a barrier to entry for many startups and entrepreneurs the world over, estopping much.
It slows and inhibits AI development.
Gluttony is not the precise term maybe, more like compulsive or untrained or underdersupervised or unrestricted overuse -- we need another new term for
What is it called when you have an AI-driven enterprise which requires compute and maybe a large distributed team of differing skills each using compute, semi or non supervised at all on functons, and attendant uncontrolled cost centers?
'Guardrails for data usage'? 'Data obesity'? 'Data sobriety'?
Hey as an AI founder building a global distributed team, it terrifies me.
I think it took down Bench, the Vancouver-based successful and very well-funded tech and AI-enabled accounting company. "America's largest bookkeeping service for small business"
At the end its customers went from sheer delight and excitement to,
"This sucks AAAS." (accounting as a service)
It did suck AAAS to end that way. They shut down instantly hello, by screen message.
* Tired stressed missing family and vacation year-end accountant instantly sees, INSTEAD OF their f/a tax content, entries, records, ability to MAKE ENTRIES
On among the very worst days of days, virtually, that anyone could choose for accounting.
Dec 27, 2024
In my work as a corporate and investment attorney over many years, this is a critical time in companies' functions for very large tax and accounting reasons.
While many are already out skiing on the slopes, or toasting eggnog with family, or doing "Boxing Days" at the malls,
There are F/A (finance and accounting) and Legal people (like myself) internal (and mostly external CPA firms) who plan to remain, and do remain each year, to tie up loose ends from the year during this quiet interregnum and work to "close the books."
Close it by midnight Dec 31 and it belongs to that year. Don't and it doesn't.
Every time, without exception, that I have chanced out to go to the slopes anytime between Dec 26 - Dec 31, I have received calls and done work nevertheless.
[Picture a frantic man waving off family and small children in ski kits with one hand, cell phone pressed to his other ear, clamping frantically in ski boots over to the "business center" to do something.]
Dec 27 is thus absolute high season for corporate accounting for these reasons.
For Bench to go dark then shook ripples of absolute operational and likely personal human meltdown through the 12000-35000 (!) Bench customers.
You're working through year end, four days to go, and your tool and data go "dark." Think especially of juniors who don't know so much just trained to use the tool they were given like they were trained.
News said someone called in a note suddenly.
People smarter than I in finance have been cautioning more recently about the danger of venture debt. It is legit dangerous for this reason.
[In June 2021 Bench rased a USD 60M C round which included at least 37m in equity and 23m debt. Altos Ventures, iNovia Capital, and BMO T&I, Contour Venture, Sage (surprised they did not buy) and Shopify. End figure totaled 100m w a pmv of 232m.]
The Information reported a bank called Bench's venture note in Dec 2024. This is of course related and may have been the proximate cause, but I do not believe it to be the cause. That was an after-effect.
How did they get so underwater that the bank called?
It was data.
I am sure of it. Speculation. But certain.
Their platform uniquely and intentionally melded human resource (accounting) w brand new AI tools which were likely supercharged and supercool. And always come with a lagging DATA bill.
In my project I have been surprised by later bills.
Am a huge believer that humanai, "humanaity" as I call it, is the best mode for most sound creation and use and development of AI tools. It means a human operator, expert, paired with AI tools that the human expert can best understand and exploit.
[Sidebar: AI is a thinkers game right now, ai is nothing other than thoughtware masquerading as software. Thought is king currency today. AI is a bionic intellectual power which accelerates the greatest minds and can confound or confuse others leading to repeated usage for clarity.]
Yet, unlike our previous "all you can eat" Internet based software SAAS of the 1990s, AI is different. The truth with AI is that if you allow people to use your AI-product, and even promote it as your business model, it will cost you additional money.
If they use it a lot, it can cost you a lot of money. If you build a large network on purpose and open the gates to many users, it can cost you a very great lot of money, and suddenly and suprisingly.
As I understand it, Bench's ambition sought to dominant the global market for AI-enabled human component accounting. It said it would do so with a unique human component + ai model. That's why it said it was doing its 100m C round raise in 2021.
I believe data costs took down Bench. "Data gluttony" (too many users "overusing" the tool, overserved "open bar" "Drunk AI"....totally spun out of control, your college student's access to Dad's credit card for just a small party. 12000x.
BOOM.
It wasn't the 'new CEO,' -- he came to save. Most probably to renegotiate/restructure the bills on data. It wasn't the bank calling the note, it was the reason the note was called.
Employer(dot)com bought so you will have to ask them.
More will be known in time.
Second to last metaphor:
Here in Texas we like green lawns. Many HOAs (Home Owners Associations) mandate/require/fine if not that your lawn be green and not brown.
It gets very hot in Texas. To maintain a green lawn requires a lot of watering. Yet, let your lawn turn brown and your HOA may fine you and the neighbors get mad.
Can you imagine the Christmas Party including the guy who owns the water company, and the HOA leaders?
The use case (green lawn despite heat, requiring massive water) is undergirded by an essential service (the water from the water company guy who makes more money the more of his water that is used).
Large LLMs and NVIDIA behind them are the water company.
This is the reality of AI software business at this moment in history and market development.
I view Bench as a cautionary tale in AI development.
How to build and change the world with AI with so much attendant cost overhang?
Some say AI requires that governments start UBI (Universal Basic Income) due to lost jobs.
I think that is a mistake. That would only de-energize humanity and auto-impoverishes the opportunity every individual represents.
Instead we probably need to think about UBC -- Universal Basic COMPUTE.
That humankind and enterprises can build and build and create the great AI global market, with less fear of their credit card instantly declined, their project shuttered.
Data gluttony will take AI development down. A solution is needed and stat.
Compute is the water on the green grass future of AI.
***
Speaking of lawns and AI here is Segway’s Navimow, AI-mapping assistive robot lawn mower…does not yet have voice command but needs it. SEGWAY NAVIMOW
AI COUNSEL NEWS + PODCAST is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
