I recently sat down with father and son team, Andrew and Elie Rashbass to talk about their new venture, ScultureAI. This startup helps organisations reinforce their cultural values through the application of AI to the many micro-interactions that take place amongst employees every day. ScultureAI helps employees adapt their behaviour reflected in their communications such as emails and messaging to better match the values of their workplace.
Andrew, Chairman and Founder, brings his extensive experience of managing The Economist Group, Reuters and Euromoney Institutional Investor PLC to the business as well as his continuing work with the Cambridge-based van der Schaar Lab and their AI initiatives. Elie, CEO and Co-Founder, has previously worked in investment banking alongside holding a master’s in psychology. Their innovative approach to harnessing AI for organisational improvement presents an interesting new application for this technology.
Martin: Could you give us a brief overview of ScultureAI and the background to its formation?
Andrew: As an ex-CEO of some large organisations and through talking with other CEOs, I’ve realised that while having a well-defined business culture is important, trying to embed it in the everyday activities of employees is often a stumbling block. Putting more posters in company elevators is not the answer. Through my collaboration with the van der Schaar Lab in Cambridge and my teaching at Harvard, I realised that AI has the potential to help us do this. Offering training to employees on embedding a company’s culture in their activities may be helpful but it is often too remote from their everyday workflows. Coaching is a better approach but has scaling issues. This is where using AI to provide a level of coaching support in the moment an employee is working has such potential.
As a result, we have now built a fantastic team of experts to develop this proposition which we recently launched.
Martin: So how does an organisation use ScultureAI?
Elie: The first stage is for us to better understand the company’s culture and values and what they are trying to achieve. This requires some upfront work on our side to dig into these variables so that our AI-based coaching reflects the organisation’s objectives in terms of their operating principles and how they expect employees to represent the business both internally and externally.
Once we’ve identified these values, we can program them into our proprietary model which uses LLMs as its base operating engine. We’ve designed our system architecture to be model agnostic and can incorporate a range of open source LLMs as well as offerings from larger vendors such as OpenAI and Anthropic
By integrating with products such as Gmail, Outlook, Teams, Slack, and later Zoom, we can embed ScultureAI into the communication tools employees are using every day. Our coaching app asks questions to get employees to reflect on how they can better reflect the broader aims and values of the organisation in their everyday interactions. Using the ScultureAI bot is entirely voluntary on the part of the employee, and we consider this vital if it is to gain acceptance amongst the workforce. We believe the value will be demonstrated very quickly, encouraging staff to use it whenever they think it might help. Examples of the types of questions it might ask, which will depend on the organisation’s desired culture, include emphasising inclusion alongside ambition or pushing boundaries and innovation while maintaining a focus on delivery.
We aim to preserve a user’s style of writing while suggesting where behavioural changes might be helpful. Our product is grounded in behavioural science as building employee buy-in and demonstrating value is crucial if user-acceptance is to be achieved.
We also recognise that different communication platforms have varying dynamics. Email is usually a one-to-one form of messaging while Slack and Teams often incorporate group discussions with multiple views and objectives being expressed in single threads. Therefore, we will coach users on Slack as teams as well as individuals, where appropriate.
Martin: Is there a danger that employees might become annoyed if they feel their messages are constantly being evaluated?
Andrew: We were very aware of this possibility when we designed our product, so users will have the power to turn off suggestions if they wish. We also want to avoid coaching fatigue, so our belief is that individuals will only seek ScultureAI’s coaching several times a week where certain messages and communications are particularly relevant to furthering and reinforcing organisational culture.
Trust is vital with such deployments. It is important that employees realise no coaching interactions are stored, and none are reported back to management except in aggregate and anonymised form.
Martin: Is your system designed to learn from its interactions so it can improve over time?
Andrew: Because we don’t store any of the interactions, we do not have a direct feedback loop in the way that LLMs are trained, for example. However, we have developed an alternative approach that, we believe, will achieve a form of learning. We have developed a number of proprietary evaluators that measure the quality of our coaching across 100 different dimensions. This data then helps us with deployments for future clients and make it easier for us to tune and customize our models.
Martin: From a business perspective, how do you build a competitive advantage in a sector where many technology companies have very deep pockets and proprietary frontier models?
Andrew: On one level, our niche technical expertise and deep understanding of behavioural motivators gives us an advantage over bigger companies that do not understand this intersection between AI and human psychology. Also, our evolving model of learning that develops as we grow also makes it difficult for others to develop equivalent products. I think there is also the advantage of having our product embedded in corporate workflows and the tools that employees are using every day. This gives us a degree of technological lock-in. Additionally, as mentioned earlier we do not rely on any single LLM so are not vulnerable on that front.
Martin: What plans do you have for the next couple of years?
Andrew: We’re developing our offering so it can be used in a broader range of applications as well as for other types of events such as internal meetings, document preparation and external meetings such as with clients or VCs where culture is shaped and propagated.
In terms of sales channels, we are expanding our direct sales but also looking to work with companies in the cultural advisory space which we see as an exciting opportunity.
