Our Investment in CANDIS and the Case for “Enhancing AI”
We’ve recently announced our investment in CANDIS, a Berlin based company leveraging Artificial Intelligence and Machine Learning to automate bookkeeping tasks.
CANDIS has a few things in common with other early-stage companies we have been fortunate enough to partner with:
- A Strong founding team with relevant domain experience
- A very large market, currently very labour intensive and addressed by tax-advisors, accountants and the businesses themselves
- A compelling ‘why now?’: Invoices and payments going digital, standards pushed down by regulatory bodies across the EU, increased accessibility to banks via APIs / PSD2 etc. (There were also the points around maturity of machine vision and AI tech I will discuss below)
- Very promising SaaS metrics indicative of product-market fit. Especially given how lean the company is running
Certain things that came up during the due diligence process with CANDIS and in general, over the past year or so, got me thinking about the different ‘buckets’ of AI/ML based companies and the reaction they get from the market.
An important context for the text to follow, CANDIS is taking a two-pronged approach to the market. They sell directly to companies but also to tax-advisors and through them, to the companies they serve. Companies invite their accountants and accountants invite other companies they work with.
This kind of distribution model is only made possible by the fact both sides perceive the product as a net-gain and not as a threat.
The above became even more apparent when we started calling tax-advisors who are already using CANDIS — they REALLY love the product and the productivity gains they’re getting (which in such a low-margin, service business, impacts their bottom-line quite significantly).
This is the part which I found interesting and perhaps different from other AI companies we’ve seen. The very same people who (in theory) should have pushed back as it’s their job that’s being automated, are the people who help sell the product. Why?
AI for X
The pitch — “AI for X” — is probably the most frequent pitch I’ve gotten from teams in the past year (followed closely by “blockchain for X” maybe).
Being the ex-consultant that I am, the more pitches I’ve seen, the more I started thinking about them on a 2 x 2 matrix as follows:
Fake vs. Real
Fortune magazine recently ran a great article by my partner Arif about this exact topic which is saving me some work here. In short, ‘Real’ AI companies build a moat around a proprietary data set with a strong feedback loop between data and output and a software that self-improves over time (CANDIS falls beautifully into this category).
Now, tech-enabled service companies and good old product/UI focused companies win categories all the time so being a ‘real’ AI company isn’t sufficient to win by any means but I do believe that automation and insights from proprietary models and access to data are becoming an important edge for companies going forward.
‘Real’ AI companies per the definition above are actually quite rare. From the pitches I’ve seen in the past year or so, less than 10% fall into this category.
Enhance vs. Replace
While ‘fake’ vs. ‘real’ AI is a fundamental technology trait (arguably, inherent to the problem space and team capabilities), the Y axis of the matrix above is a product/go-to-market choice as the same technology can often be applied to REPLACE humans or AUGMENT them.
Consider for example Autonomous trucking/driving companies building a fully automated vehicle (e.g. Kodiak Robotics) vs. Tesla’s assistive approach with their Autopilot vs. autopilots on commercial aircrafts where the first versions came out as early as 1912.
These products take a very different path to market that is impacted by technology but also, and maybe more so, by regulation and business considerations.
Much has been written about the ‘future of workplace’ and the Skynet- powered-robot-overlords coming to take your job so I’m not going to go into that and instead, focus on how this ‘enhance vs. replace’ choice is impacting market acceptance.
It’s interesting to observe tech companies, often selling very similar solutions within within same industries, trying these two different approaches to market.
So — is replacing better?
I used to believe that a company with a ‘real’ tech edge will generate a far superior enterprise value by applying that edge to an end-to-end product and capitalize on the entire value chain while companies based on ‘weaker’ tech will end up selling a signal/model/SaaS/dashboard based solutions to existing players in the value chain.
This line of reasoning is based on the assumption that the tech edge will manifest in higher margins, lower prices and therefore an improved ability to grab and defend a meaningful position in the market. There’s also another implicit assumption here which is that a company who provides an on-par or better service at a lower price-point (due to tech in this case) will attract more customers.
Again and again I’m reminded that this isn’t necessarily true as customers (companies, but really, people within companies) buy on many other considerations other than price or ‘efficiency’ alone.
If you’re a founder pursuing an ‘AI for X’ path — I think it would be wise to recall that the task your company is automating is currently performed by a human being. That person does not like feeling redundant or obsolete. If you’re selling to an enterprise customer, it is very likely that this exact person would be the first within the organization to be called upon to examine or budget your technology.
And if you ARE pursuing a full ‘replace’ model out of the gate, gear up properly and expect resistance..
I’ve seen this play out a few times in recent years in different industries (e.g. autonomous software testing, AI for field tech or automated fraud prevention) and it’s common to see founders course-correct after their initial encounter with the real-world buyer beyond the first batch of design partners and beta customers (who often have a burning need or level of comfort with tech that doesn’t represent the market as a whole). This encounter forces founders to crystalize their go-to-market and choose between a full-stack offering (replace) and a product built around empowering that buyer (enhance).
In some industries, a ‘replace’ play is simply too difficult to shoot for which is why, for example, we see companies pursuing a decision-support FDA approval in healthcare or a blended-learning model in education.
Having said that, the same could have been said for driving a decade ago so I’m certain we will see more founders pursuing a full ‘replace’ model in difficult industries going forward.
Back to CANDIS
In a market where several competitors are pursuing an aggressive ‘replace’ playbook, CANDIS (so far at least) is able to walk a very fine line and build an AI engine that automates much of the human accountant work but productize it in a way that resonates with accountants and advisors instead of alienating them.
It will be interesting to see how this market evolves as I personally believe it’s a market that is very likely to experience a massive change in the near future. I’m very glad to work with team CANDIS on defining this future and excited for what’s to come. Wish us luck..
