TL;DR
- AI adoption does not move at one speed. It moves at the speed of each industry's diffusion curve, and that curve is set by asset intensity, regulation and the cost of being wrong.
- The finance function is automating faster than almost anything except software engineering, because accounting is a set of rules with a predictable set of exceptions.
- Asset-heavy industries look slow from the outside and are usually being sensible. When one bad recommendation can destroy a crop cycle or a plant, caution is a form of competence.
- Where your industry sits on that curve should decide what you do next: build data infrastructure, run a co-build, or wait. It should not be decided by what the vendor in front of you sells.
Our business model is unique. We build ventures in partnerships with midmarket companies in regulated and asset-heavy industries. But these ventures don't start as ventures on Day 1. They usually start as co-build/co-sell motions or consulting projects that we start for them. Occasionally our pre-built IPs come into play. (Related read: Buying Vs. Becoming AI) The distinction between renting a capability and owning one is the argument in Buying vs. Becoming AI.
Through all these motions, we meet executives from a wide range of industries. In the last year, we have spent time with executives from biotech, paper, finance, HVAC, agri, healthcare, municipal tax, energy, and a few other industries. This gives us a rich perspective on how AI diffuses in each domain. (Related read: Building on Boring Strength)
The short of it: Each industry moves at its own pace, for several reasons.
Why finance is automating first
The finance function is being automated at a speed second only to software engineering. Counterintuitive as it may sound, accounting for instance, is a set of rules with a predictable set of exceptions. The complexity is human-level but not so complicated that AI will fail. I anticipate that within the next 12 months several accounting functions will be automated by at least 70%. Almost no one writes code in startups these days. They simply watch AI write code. Posting transactions and closing books will follow this pattern. We anticipate that foundational models will include these capabilities out of the box.
Let's take two asset-heavy industries: HVAC and Agri.
HVAC: the technology is not software
I was at a recent HVAC conference with over 50,000 professionals. It was chilling to see just 50 (out of 1000) booths occupied by tech. The biggest tech vendor booths belonged to aggregators and marketplaces. So, not really tech. HVAC technology isn't software; it involves heat pumps and similar equipment. That's what I understood. The loudest announcement about AI compared purchase orders (POs) with invoices—a problem the insurance industry solved 2 years ago. AI may be magical, but diffusion absorbs the shock and propagates the innovation over the years.
Agriculture: the product is alive
Agriculture is a different beast. The product is a living organism. Living things are complex and complicated. There is already significant innovation in this area, including activating lights of specific wavelengths, targeted nutrient delivery, and moisture control. However, none of these systems communicate with each other or recommend holistic interventions.
Moative has built interesting intellectual property in this area (‘State-Space Models') that understands how underlying systems of confounding variables behave, by observing other visible variables. It's applicable in Agri but diffusion plays a role. An AI engine that recommends holistically is also a weapon that can destroy the entire system if its not understood well. In agriculture, the failure of one crop batch can lead to the failure of the entire business.
Reading the curve before you spend
Moative's business model is malleable to the effects of diffusion. Is everyone rushing to do AI in an industry? Consulting might be a good idea but launching a startup is not. If an industry is two years behind, now is the right time to invest in data infrastructure and engineering to harness the flow of data and context. That's what we would do. Is an industry technically ready, but held back by past scars from moving forward? We will seek a visionary company that thinks from first principles and experiments with us. Potentially this would mean a co-build motion. (Related read: Don’t Sell to Private Equity)
Where to actually start
If you are thinking of where to start with AI for your business, the right first step is to understand the state of tech, how it is diffusing in your industry, and assess whether accelerating that diffusion provides a competitive advantage. Measure the risks you are taking and choose a partner willing to partake in that endeavor, through models that proportionately distribute the risks and rewards. (Related read: Don’t Sell to Private Equity) The gaps worth finding before any of that spend happens are set out in Find the five gaps before spending on AI.
If you have questions, write to me. I am on ash@moative.com
Enjoy your Sunday!
Top reads
Where Moative fits
Moative reads the diffusion curve before it proposes anything, because the same capability is a good idea in one industry and a waste of a year in another. That is why our engagements range from consulting through co-build to joint venture rather than arriving as one fixed offer. In healthcare revenue cycle management, where the document flow is dense and the exception patterns repeat, the curve is further along than most operators assume. In asset-heavy industries it is usually earlier than the vendors claim.
Frequently asked questions
Why is AI adoption slower in traditional industries?
Because the cost of a wrong answer is physical rather than cosmetic. In agriculture a bad recommendation can cost a crop cycle; in asset-heavy operations it can cost equipment or uptime. Slower adoption in those settings is usually a rational response to blast radius, not a cultural failing.
Which business function is automating fastest with AI?
Outside software engineering, the finance function. Accounting is a set of rules with a predictable set of exceptions, the vocabulary is standardized, and the output reconciles or it does not, which gives you a cheap and immediate correctness check.
Why is there so little AI in HVAC?
Because HVAC technology is mechanical rather than digital. The engineering that differentiates a vendor is in heat pumps and refrigerant, not in software, so the technology presence at industry events is small and weighted toward marketplaces and aggregators.
What is a state-space model and why does it matter in agriculture?
It is an approach that infers how an underlying system of confounding variables is behaving by observing the variables you can measure. In agriculture that matters because lighting, nutrients and moisture are controlled separately today and no system reasons across all of them at once.
How do I know if my industry is ready for AI?
Look at three things: whether your data is already flowing and captured, whether comparable firms in your sector have run real deployments rather than pilots, and what a wrong answer costs you. If the data is not flowing, that is the investment to make now, regardless of what models can do.
Should a midmarket company build, buy or partner on AI?
It depends on where the industry sits on the diffusion curve. If the sector is already moving, buying and integrating is reasonable. If it is roughly two years behind, invest in data infrastructure first. If it is technically ready but scarred by failed programs, a co-build with shared risk tends to be the only structure that gets started at all.
