Gifting and AI adoption
B2B software products are "gifts", given by CIOs to business teams. There is an inherent problem with that. As LLMs have upended software economics, driving usage has become a big problem for vendors
Gifts are violations
A few weeks ago I got my wife some gifts. She sort of liked them but not that much, and said “I feel violated”!
This was in reference to a conversation we’d had sometime ago, perhaps along with some other friends, where we had discussed gifting as an act of violation. By gifting someone something, you are forcing them to (even if temporarily) own that thing, whether they like it or not.
Sometimes they might actually like it (in which case they could have bought it themselves using your cash gift) and keep it, sometimes they might keep it out of guilt, and sometimes they might recycle it. Irrespective of what they do with it, the fact remains that they had to deal with something they didn’t choose to own.
And then, of course, if you ask any economist, they will tell you that gifts are a “dead weight loss” on the economy.
The adoption challenge with B2B Tech
I was reminded of this gifting analogy when I was thinking about why a lot of AI-for-data products are struggling for adoption. Of late I’ve been having a lot of conversations with both vendors and clients in the space, and there have been very few happy stories. And this affects both sides - in the absence of adoption, clients suffer by not getting value from a tool they’ve invested in. Vendors suffer by not getting revenue - since at least a part of it in AI-for-data products is usage based.
And AI-for-data is not the first class of products which has struggled for adoption. In the same domain, just a few years ago, dashboards had the same problem - companies would struggle for adoption of dashboards. I’m possibly tired of saying this but, when I started Babbage Insight, I interviewed ~100 CXOs on how they get insight from data. Fewer than 5 actually used dashboards.
You have this issue elsewhere as well - I’ve had old and new clients talk about driving salesforce adoption, driving github adoption, driving Jira adoption, getting finance teams to use Anaplan effectively, etc. etc. That distribution doesn’t imply traction isn’t necessarily a data or AI thing - it has been an ever-present problem in B2B tech.
B2B Tech has an adoption problem because most software products are “gifts” and gifts, by definition, are violations.
The basic problem is that apart from a handful of products that are marketed to actual users (think devtools, my old Babbage Insight, etc.), most B2B Tech products are essentially “gifts”. Gifts bought by centralised teams such as CIOs and BI and data engineering, and “gifted” to actual users (business / sales / product / … ) to use.
And there is no surprise that all the problems with gifting (which most people who are not economists don’t talk about, since it’s not “polite”) haunt B2B sales also.
Most users, in most cases, had no say in the decision-making process. In a lot of cases, they didn’t even know that this tool was coming, and the way they worked needed to be changed based on this. On top of this, typically most “normal business managers” are steady-state types who like to be left alone to do their work, and are resistant to change - especially changes imposed on them by others.
So there is no surprise that a lot of B2B Tech products struggle for adoption. The people who are expected to adapt and adopt never bought the product in the first place — someone else bought it for them. That the buyer has bought is no guarantee that the user uses.
The economics of B2B software, and AI
The B2B software buying / selling process hasn’t changed using AI. Users have seldom been the buyers for such products in the past. So why has the lack of usage suddenly become a problem now? It has to with pricing, and economics.
In the grand old days, software was a zero marginal cost product. The cost of making an additional copy of software was pretty close to zero (basically the cost of the CD on which it was burnt), leaving gross margins close to 100%. The first change happened with the emergence of cloud-hosted software. Now, the vendor had to bear the cost of hosting and serving the software, and SaaS meant that vendors assumed the cost of maintenance as well. This dropped the gross margin, but it was still high.
Typically software products were sold per-licence, or per-”seat”! The near-zero marginal cost meant that it didn’t matter how much of the software someone used - every user was profitable. And per seat pricing was a convenient proxy for charging more money from larger customers. Salespeople were used to selling this way. CIOs were used to buying this way. And once the CIO had paid for a certain number of licences, whether they used it or not was none of the salesperson’s business. “Driving adoption” was an entirely internal problem.
LLMs have completely upended the economics of software, and thus pricing, and thus adoption.
Thanks to token costs, every use of a piece of AI-enabled software now costs the vendor real money. Gross margins of software have thus plummeted. More importantly, it has had a massive impact on pricing. A flat price doesn’t work any more - light users will end up cross-subsidising heavy users, and soon users will discover the inherent adverse selection, driving margins worse.
LLMs have completely upended software economics, so now adoption is also a vendor problem
So most vendors of AI-native software products have changed the pricing model. Pricing is now a function of “usage” - however it is defined. Some vendors simply charge a markup on total tokens consumed. Some others charge a “fee per use case” while passing on token costs in addition. Others charge based on the overall quantity of data scanned (another great proxy for usage). There might still be a per-user fee, but that is usually small compared to the overall costs that incorporate usage.
No use, no pay
The upside of usage based billing is that you continue to be profitable when a user uses more of your product than expected. The downside of usage based billing is that when the user doesn’t use your product, you don’t get paid at all!
So driving adoption and usage, which so far had been the buyer’s (typically the CIO’s) problem, is now the vendor’s problem! And from my conversations, they have been dealing with this in a few ways:
Customer success teams are suddenly way more important. Earlier, they were primarily focussed on renewals and not using the customers. Now the customer success team is way more invested in constantly driving usage - there is no revenue without this.
I don’t know if the profile of people being hired for customer success teams has changed, though.Every vendor now has a team of forward deployed engineers to make sure the product is tightly integrated with the customers’ processes and workflows. The faster you get your customers productive, the faster you now get paid.
The one place where I’m not yet seeing that much effort (and where I’ve started talking to both clients and vendors to step in and help out) is working with the client’s business and data teams, and helping them get past the roadblocks that prevent them from adopting.
Outside of working past the “I didn’t buy this tool” issue, this includes (but is not limited to) helping clients figure out how to give the AI-for-data products the right kind of context (to “train” them), guarding against “context decay”, making sure both the data and business teams have confidence in the outputs of the analysis (IMHO the biggest blocker to adoption, leaving aside the “gift” thing), helping teams know how to get maximum RoI out of the product, etc.
I’m starting to do this specifically for AI-for-data products (since that’s “where I come from”), but will soon expand into driving usage and adoption of other kinds of AI products as well.