The Secrets of Successful AI Startups. Who’s Making Money in AI? Part II

Cross the AI Commercial Divide to the Enterprise

Those startups that are doing well generally understand the nature of AI technology and the opportunity in the enterprise. But more than that AI startups that are starting to scale have all crossed the commercial divide from a technical world to the enterprise. They have learnt:

1. Don’t move too fast and break things; embrace Responsible AI

This culture popularised by Silicon Valley works well in a business to consumer (B2C) world where the consequences of a bug in an application are relatively limited. Software developers globally have embraced the lean startup and agile methodology. But this means live applications can often have bugs and break. This doesn’t really work well in the enterprise.

This is especially true in highly regulated enterprises such as financial services, agriculture or pharmaceuticals. Technology is driving greater more and more regulation. Europe introduced GDPR regulation in 2018 that provides rights to individuals as to the use of their personal data by companies. There will be significant fines — up to 2% of a company’s revenue — for the misuse of consumers’ personal data. And in the financial industry the response of “whoops, I guess the app didn’t work” doesn’t work when you are dealing with real financial data and financial transactions. Revolut a new UK challenger bank that relies heavily on AI has found itself in hot waters with the regulators over alleged compliance lapses. And in a world where the consequences of automated decision can be life-changing, such as autonomous vehicle crashing or critical health care treatment diagnoses, you better make sure you are confident in the reliability and accuracy of your automated decision making.

Corporates are increasingly putting in place board governance and oversight to manage reputational risks to their firm from the use of AI. For example if the datasets used to train AI algorithms have sample biases then a company’s brand can suffer if they are seen as being discriminatory. We all saw the recent headlines where algorithms were shown to be much better at recognising the gender of white males compared to other ethnic groups. Or the recent case from Amazon where they abandoned their hiring recommendations systems as the machine learning simply mirrored the fact that hiring historically was heavily male, white and young.

Increasingly startups will need to offer enterprises some level of assurance around the risks of their AI offering. Does your AI startup technology provide explainability for fully automated decisions that have legal effect? For example if your technology can be used to automate hiring decisions then you will need to explain how the algorithms work under GDPR. You need to demonstrate that the algorithm is not biased against people based on protected classes such as gender, age, socio-demographics or health challenged (feel free to take a look at our introductory readings on the topic of explainable AI). The Information Commissioners Office in the UK recently released a discussion paper that identifies eight AI specific risk areas that enterprises are likely need to manage including (a) fairness and transparency in profiling which is especially concerned with bias and discrimination, (b) accuracy of the AI models, (c) the level of automated decision making be it fully automated or human in the loop, (d) security and cyber risks and (e) tradeoffs in accuracy versus privacy versus explainability.

Startups and corporates are going to need to be really well versed in this topic which is increasingly being referred to as Responsible AI practices. As a startup don’t move too fast and break too many things. It could get you in a lot trouble. And ensure you are embracing and demonstrating Responsible AI practices (a topic I will write more on later).

2. Solve really high value use cases, not nice to haves

In the past few years we have seen a tidal wave of consumer mobile applications that addresses any imaginable consumer need. Apps can be built in weeks, launched and consumer traffic bought by placing ads on Google and Facebook. But this approach doesn’t work in the enterprise. There is something of a zero sum game where there is little appetite from the chief information officer (CIO) and other executive leaders to embrace, yet, another technology solution. We all recognise work frustrations trying to rollout the latest application from HR or finance or sales or marketing. We struggle to remember our passwords. None of the new applications work the same. And then we certainly often can’t remember where we stored that file be it on the project or my personal cloud folder. Technology leaders are frustrated trying to integrate more technologies into their existing and often fraying legacy databases and technology platforms.

To get the attention of a CIO and a head of department, such as a chief marketing officer (CMO), your solution better be addressing a really important problem. The type of problem it should be addressing should be one where the manager is going to bed worrying about it and their bonus plan is clearly tied to it. “Nice to haves” do not work in the enterprise. I was chatting recently with a CTO of an AI startup who has spent two years building an AI product suite to drive enterprise intelligence but the sales aren’t coming. Why? Not high enough value yet.

An example of a really high value use is from HireVue. They worked with Unilever to save over 50,000 hours in candidate interview time and delivered over £1M annual savings and improved candidate diversity with machine analysis of video-based interviewing. That’s a lot of money. Reinfer, a British startup that uses advanced NLP algorithms to sift through billions of emails and messaging to determine what people are communicating about, recently completed a pilot with a major international bank. It identified major issues in post-trade operations by analysing mailboxes with the use of machine learning with the potential for millions in operational savings. These are really high value use cases.

3. Master B2B enterprise sales and learn calculated patience

Startups need to ensure they master enterprise sales. These skills are at a premium for AI suppliers as it takes time and hands-on experience to master consultative sales . The most valuable training course I took early in my career was the SPIN sales methodology — S(ituation), P(roblem), I(mplication), N(eed). Selling requires time to identify stakeholders, get a meeting with those stakeholders, assess the current business situation, ask questions to identify critical problems, assess the implications of those problems across multiple departments, build a consensus as to the implications andt he need. And even then there is no guarantee that your solution to this need will get budgetary prioritisation.

A typical corporate manager has a to-do list that lasts from here to kingdom come. For a startup solving a high value use case need is critical so don’t just base your company on a “product hunch.”

And patience is at a premium when a sales cycle can last 12–18 months. Many AI startups run out of money or assign precious resources to sales opportunities that are not qualified. Make sure you qualify out opportunities quickly if you think they will ultimately not lead anywhere. Re:infer found the sales cycle was really really long but they stuck with it identifying a really high value use case, finding an internal champion, and completing a successful sales pilot.

4. Translate AI for the real world

Startups need to be able to explain the value of their unique algorithm, technology, product or solution in a language that is readily understandable by a business audience is critical. They need to be bi-lingual. With many AI startups being founded by young, wicked smart and highly technical minds there is often a communications gap with the “suits.” We see executives’ eyes glaze over all the time as the minutiae of this cutting-edge technology is explained. And if little context is provided as to how this technology can help drive a business forward — by impacting revenues, efficiency and customer service — the executive will lose patience rapidly. It is critical that startups speak the language of technology and business. They need to translate between technical and commercial languages.

5. Lower the barriers to trial (a pilot)

It is important that startups make it really simple to implement a pilot. AI solutions often require much data and much time to train the machine learning models. And AI often requires significant engineering integration into back-end technologies to work well. It can take months to acquire, clean and wrangle data then months to use it to train models. In enterprises that are very busy there is little appetite to do a lot of heavy lifting to setup a pilot. Companies want quick proof of concepts. DigitalGenius has done a good job on this front. Their customer service technology works on top of existing enterprise platforms, such as Salesforce, and they have minimised the amount of time it takes to train the technology to answer customer queries. They have also made it very simple to integrate into existing work-flows providing human-over-the-loop decision making. KLM, the Dutch airline, claims to respond to over 50% of customer enquiries on social media by implementing a machine learning chatbot with DigitalGenius.

6. Technical founders need to hire business people.

Startups need to ensure they have the skills to cross the commercial divide from the world of techies to the enterprise. The Harvard Business Review recently reported that the startups most likely to succeed have technical founders who have quickly hired business people. “One theory for why technical skills seem to matter more for a founder is simply that the average technical founder has better business skills than the average business-trained founder has technical skills.” But blending the DNA of founders with commercial people often requires alchemy. I have lost count of the number of times that I have seen startups struggling to get this cultural fit right. The most common is technical founders will hire sales “farmers” instead of sales “hunters.” Farmers don’t know how to go out and hunt for business which is what is required for a startup . The are used to big, fat marketing departments bringing them a torrent of leads to harvest.