SDG 6: Clean Water and Sanitation
SDG 7: Affordable and Clean Energy
SDG 9: Industry, Innovation and Infrastructure
SDG 11: Sustainable Cities and Communities
2. Project Details
Company or Institution
Lili.ai: weak-signal driven project management
General description of the AI solution
As stated by the UN Sustainable Development Goals, the development of quality, reliable, sustainable, and resilient transformative infrastructure is critical to society's ability to innovate and the well-being of the economy. Yet, according to McKinsey, these large-scale infrastructures face an average cost increase of 80% of the original value and an average delay of 20 months. One of the problems (and perhaps also one of the solutions) is that complex projects generate a large amount of rambling, unstructured documentation, making it impossible with current non AI tools to effectively search for information or factually track the progress of these large projects. To cope with this massive and ever-changing documentation, Lili.ai has built a pre-trained intelligent system that leverages recent advances in ML/NLP (machine learning/natural language processing) and Lili.ai's 5 years of expertise in applying AI to large-scale projects.
Lili Search is the only platform specialized in projects that extracts knowledge from heterogeneous and disjointed project documents (Gantt charts, project business records, project charter, meeting minutes, emails, etc.) and transforms them into a semantic representation of project data; and performs searches with relevant interfaces. This representation is used with semantic search engines and data-driven ML to: (1) understand the chronological evolution of a situation, (2) detect early signals and alerts of risks, (3) early dispute avoidance; and (4) increase transparency on the reality of project execution.
Lili Search is currently used by major players to reduce risks on multi-million dollar projects: hydroelectric plants, bridges, transportation and highways. The results include improved budget predictability and accelerated fact-based decision making. Lili.ai's technology has been recognized by top 30 out of 10,000 AI Xprize teams, CogX AI Rising star awards and more.
We hope that by getting involved with IRCAI, we will help improve transparency and efficiency in the delivery of projects contributing to the Sustainable Development Goals.
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
Lili.ai brings radical improvements compared to the state of the art of AI applied to project management, and compared to existing commercial tools. This breakthrough has been recognized by the world leaders in project management (our customers), by AI specialists at the Global AI Competition (Xprize and CogX AI) and project portfolio management (PPM) editors (Oracle Construction Innovation Lab).
State of the art: Several works in the scientific literature have studied how AI could revolutionize project management in terms of: (i) natural language processing, semantics and knowledge capitalization; and (ii) project management optimization under uncertainty at the project level. Yet, few (if any) current tools seem to have built-in intelligence to explore past project knowledge. Instead, these tools still rely on individual experience and intuition.
Existing tools are limited in their scope and use of AI: traditional PPM solutions (Microsoft Project, Primavera, …) use a rigid project model without word processing; data-driven Ediscovery solutions (Forecast, Nplan, …) are limited due to their generalist approach. Companies interested in large projects have used generalist machine learning libraries to build their in-house solutions, usually at a prohibitive cost (creating a qualitative annotated dataset).
Our solution (TRL8, 30 projects analyzed with world leaders in energy, transportation and construction) is built at the intersection of natural language processing (based on symbolic rules, morpho-syntactic analysis, information-extraction, …), deep learning (transformers and active learning), knowledge representation and search; combined with a unique focus on project management practice. This allows us to develop specialized resources and expertise: dynamic ontology and annotation schemas, but also a club of project managers to think with us about the features and the approach. Among the achievements we are most proud of: the detection of a milestone that will be missed 6 months before the project health indicator turns red.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
Our goal is to help improve transparency, collaboration and efficiency in the realization of transformative projects (new clean energy infrastructure, new water pipelines, etc.) thereby contributing to the delivery of SDGs. One of the most scalable and attractive aspects of Lili's technology is that it disrupts project management without imposing any change in habits, as Lili.ai processes existing daily documentation (reports, meeting minutes, emails, scans, etc.).
Since most of these mega-projects are funded by the World Bank, European Bank, etc., the only way to disrupt project management at scale is to have these organizations tie funding to transparent, collaborative and efficient project delivery obligations with technologies that allow for automatic auditing of project documentation for compliance violations (no child labor for example) and early detection of risks to secure project delivery and ensure efficient spending of public money.
Transparency: Lili allows to understand the chronological evolution of a situation and to increase transparency on the reality of the project execution. This can be measured by a reduction in time and an increase in the quality of the project audit.
Collaborative delivery: Lili allows to detect signals and early warnings of risks. This can be measured by the number of disputes/arbitrations (legal fees) avoided on projects with Lili.
Efficiency: Lili.ai can be seen as a shared and collective project memory by accumulating knowledge over time and across portfolios and domains. This can be measured by the number of early warnings detected by Lili.ai.
Thanks to Lili.ai, valuable time and money will be saved (the average duration of mega-projects is 80%; the average delay is 50%); with these additional resources, international institutions will be able to build more new infrastructure for underserved communities, and they will have access to it sooner.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
Lili has been successfully used for auditing 30 major projects for world leaders in energy, transportation and construction. It has been used by more than 45 project managers of large groups. In some cases, Lili was able to detect delays up to 6 months in advance on major milestones for instance. In addition to formal evaluation strategy, we have collected satisfaction letters from these groups in support of our solution.
Within its role of support for big project management, Lili's impact can be measured by delays reduction relative to project duration. The time and resources can also be measured against a control group of projects not using Lili.
With scaling up in terms of number of projects, industries and countries, Lili aims to make its model a requirement for all major projects. By integrating policy makers and major stakeholders, we can ensure that projects funded with taxpayer money are being audited to guarantee transparency and reusability.
The major obstacles for scaling include the accessibility of project data which takes time and effort to acquire. Furthermore, the latency in reaching international adoptability of Lili's approach due to difficulty of changing in-place procedures.
Since 2016, Lili team has been working along world-leader companies in complex project management to extract insights about project failure, gather data, observe and interview specialists, create and adapt internal tools, test algorithms and validate the results. This experience is necessary for anyone to make use of AI technologies to address sustainable project management. Lili's ambition is to eliminate this barrier for new stakeholders and create a center of excellence grouping practitioners, decision makers, companies and organizations to share best practices and models for increased transparency and efficiency around an open standard for sustainable project management.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
Fairness and data bias in AI. Lili uses machine learning and natural language processing algorithms at the core of its data-driven approach to decision-making. It is used to find patterns in data and make predictions to assist humans. To mitigate racial and gender biases that may be present in training data, Lili closely follows active research in the field of algorithmic fairness and computational ethics in artificial intelligence which is rapidly developing to address this issue. However, the tasks performed by Lili mainly involve analyses and predictions related to project organization and artifacts without directly involving people which limits the impact of inherent biases.
Personal data. Lili is a subcontractor as per GDPR Rules and acts on behalf of and under the instructions from the controller. Lili manipulates professional documents only and does not include personal data, and no functionality offered by Lili focuses on personal data. All collected data are pseudonymized according to a strict procedure and are encrypted end-to-end during transfer for real-time processing. Processing is carried out with the consent of persons involved in the projects and also after the signature of an ethical chart with the client. We make sure project actors are aware of the analysis of documents generated by them.
We believe that transparency is necessary for trustworthiness and wide adoption of AI tools. We therefore focus on explainable approaches to AI in order to produce results understandable by humans. This includes the use of symbolic rule-based systems, and probing techniques in deep learning which extract understandable rules from numeric representations.
Green computing. Project data is also filtered to restrict its size and impact on training resources. We also make use of pre-trained publicly-available models to avoid retraining and wasting of compute resources.