It’s been a while since our last blog post, so I wanted to write an update on what we have been up to and our progress.
During the past couple of years, we have been building tools to help teams working in expert fields find the information and expertise they need. These fields include law, engineering, consulting and medicine, and our aim was to craft an approach that was relevant for any profession where expert knowledge is needed to interpret information and make a decision. We wanted to make it possible for an expert to scale their expertise.
The work initially centred on how we can leverage language models to help teams manage, find, and exploit unstructured or semi-structured information in large-scale companies and organisations. What we observed was a workflow pattern that repeats across industries. Whether it’s an engineer looking for the latest planning document, a lawyer reviewing a large swath of contracts, or a consultant preparing a proposal for a complex client project, they all need fast access to the expertise buried inside in-house content, to help them deliver their best work on new projects. Additionally this expertise is not just the content itself, but also what to do with that content.
Being able to surface this expertise quickly and easily has several upsides: It reduces the incidence of repeated mistakes and re-invented wheels; it increases quality and efficiency in key revenue-generating tasks like project bidding; and it empowers junior employees with the expertise embedded in the work of more senior colleagues.
Traditionally, companies attacked this problem with knowledge management and search software. However, many such approaches have fallen short of aspirations. Typically, they are either too broad, trying to shoehorn a diverse set of needs into an overly rigid structure, or require too much additional work to maintain, often leading to a lack of adoption. The big opportunity we saw was that machine learning could now help solve this, by learning from your work and behaviour to guide you to the right information, all in a package that is configurable to your needs.
Since then, we set out on a journey of understanding how this could be delivered. We have encountered our fair share of challenges, including complex deployment environments, strict compliance and security regulations, legacy systems, and the overall complexity of introducing a new way of doing things to organisations that have long established practices. Some of these challenges were to be expected and some exceeded our expectations, but throughout we have sought to understand the key components that would be necessary to provide our clients with an efficient and scalable path to solving them, both in terms of cost and effort.
We are therefore very excited to announce that a distillation of this process is now available in a new expanded version of our product. One which aims to address the challenges we have encountered in a way that we believe will help many others.
What we have arrived at is a flexible, intuitive platform that lets individuals and organisations train the solution to find relevant knowledge “blocks” locked in documents, then save them for easy reuse. These blocks could be clauses in contracts, sections in research papers, or methodologies in a proposal.
Storing all these blocks centrally, but as separate entities, has two main advantages: First, individual blocks (or small collections of them) can be easily found and re-used as reference points or scaffolding for new work. Second, blocks can be analysed and inter-compared en masse, to identify trends and create reports. Whether this be a report on research trends in an R&D department or contract analysis for a due diligence process.
The software is also simple to use: You don’t need a computer science degree. Instead, it allows any expert in his or her field to train the system directly, simply by using it. What we have seen is that for certain industries there are some specialised solutions that are optimised for specific tasks, with little or no human involvement. This approach can be efficient if the cost of failure is low or the volume of usage for that single use case is high enough to justify central manual correction by data scientists. However, when this approach fails it falls short of delivering expected results, and the transparency and correct-ability that many professionals want and need to trust a system is missing.
Lateral’s software solves this with a tight corrective feedback loop, giving the user a direct sense of what the system is learning, but also allowing for efficient human supervision in those mission-critical cases where the only acceptable pass rate for the “last 20%” is 100%. It also keeps all your active work in one manageable place, rather than across many separate tabs, windows, excel spreadsheets and word docs.
In short, Lateral is optimised for subject matter experts — engineers, researchers, consultants, accountants, risk analysts and medical professionals — who regularly need to interpret, extract and synthesise insights from large volumes of information, in order to deliver value to current and future customers.
In our experience, we also found that clients are very keen to try compelling new tools to increase their productivity, but often need to navigate hard security and compliance requirements for third party software, especially in larger enterprises. To meet this need, we’ve engineered a deployment process that aims to make it easy to validate the benefit our tool can provide you, while accounting for a diverse array of security requirements over time. Simply put, clients can test the software in Lateral’s hosted environment with a small sample of content, then deploy to their own production environment using Kubernetes or Gravity. Importantly each step aims to deliver results, that help motivate and justify the next step.
Finally, as we aim to help you make your expertise scalable, we see a huge opportunity of connecting the models generated by Lateral to other software. Using our API, Lateral software can be extended and integrated into other workflow software, e.g. robotic process automation (RPA), document management, or productivity solutions.
Overall, our mission is to help people leverage the collective expertise of organisations, in order to increase the speed and quality of their work. This product is a major next step in that direction as it gives teams the ability to train and correct the system directly across multiple levels of its interpretation pipeline.
If you work in a field where such a tool can help you, or are building tools to support this audience, please get in touch or request access. We are excited to show you what’s possible and welcome you and your teams onboard this exciting journey.