This is a shortened version of Making Risk Flow podcast, episode 3. You can listen to the full episode here
In this episode, Juan went through a deep dive into what digital risk flows look like for new business submissions and the benefits they drive.
Current new business process
Today, commercial insurers receive over 80% of their new business submissions via email from brokers. These emails contain the risk information in the body of the email and in the attachments. Until recently, the only way to analyse the risk and review the documentation was for humans to open each one of those emails and navigate through the attachments. Even worse, in many cases, we see it’s the most experienced underwriters who are doing this activity.
These challenges in the current new business process lead to manual risk flows where:
- Over 50% of underwriters’ time is wasted in either non-value-add activities (such as analysing out-of-appetite submissions) or manual activities (such as doing online research to pull risk data)
- Most experienced underwriters are deployed as traffic controllers, doing submission routing activities instead of being underwriting complex risks
- Underwriting leadership doesn’t have an effective way of steering the portfolio. They do not have a way to ensure underwriters are focused on the risks that enhance the portfolio, other than by sending guidance via email to all frontline underwriters.
- There are limitations in driving superior risk selection as insurers only have available the same risk information as every other competitor in the market to assess the correlation between loss behaviour and risk characteristics.
- It is difficult to identify growth opportunities. A certain trade, ie solicitors, might not be profitable as a whole, but there will be segments within the trade that are profitable. Unless you have more risk data available to segment the trade, you can neither identify nor capture these growth opportunities.
- There is no visibility in how submissions flow through the insurer, what submissions are declined or who touches each submission, which hinders insurers’ ability to identify operational improvement opportunities
What does the new business process look with digital risk flows
How to overcome all these challenges?
With digital risk flows deployed for new business submissions, the goal is to enable automated decision making on the best course of action for each submission and ensure that underwriters only work on winnable, decision-ready risks. At its highest level, it provides insurers with the capability to automatically process submissions resulting in either a decline (because it’s out of appetite or low value), a fully digitised risk ready to be straight-through-quoted (for the more homogeneous, simpler risks), or a risk that contains all required information for an underwriter to analyse (for the more complex ones).
So what do insurers need to enable this automated decision making? There are 3 steps to achieving that state:
Build a digital representation of the risk
Let’s start with the first step, how to build a digital representation of the risk. The objective here is to build a digital risk profile that contains all the information required to make decisions on the risk. The risk profile is built progressively through multiple steps of the workflow where each step requires different combinations of data – for example, to filter the risk against appetite requires a smaller number of data fields than to make the risk quote ready. The main challenge with submission digitisation is that the information contained in the broker submission is not enough to evaluate or underwrite a risk. To evaluate a risk typically underwriters need to combine data from three different sources: the broker submission, internal data and external data.
- Starting with the broker submission, it contains client details (such as client name and address), broker information (which will be important to prioritise the task), quote details (such as inception date or target premium) and risk details (such as turnover or number of employees). Then depending on the product requested there will be additional risk information such as individual cover limits or relevant risk details. All this information comes in email bodies and attachments that cannot be directly processed, but technologies such as OCR and AI can extract this information at increasingly higher levels of accuracy and precision. The challenge with extraction alone is that you get data in the form of key-value pairs or raw text, but comes with two limitations: the first one is that you need to match the extracted data to insurance entities, which is not a straightforward task and secondly, that you need to ensure high levels of accuracy and precision to be confident to make automated decisions at the back of that data, as often insurers will use the extracted data to drive straight-through-quoting without human intervention. Digital risk flows overcome these challenges in two main ways:
- the first one is the concept of data extraction confidence, which provides an estimate of how certain the extraction technology is that the value extracted is correct; Digital Risk Flows include minimum confidence thresholds that only use data extracted over that confidence threshold.
- the second way is enabling higher levels of certainty. Digital risk flows use external data to check that the extracted values are within the expected range; for example, when extracting a company turnover from a submission, this value can be compared to the turnover value from external sources
- The second source of data to digitise submissions is internal insurers’ data. This data is critical to put the risk in the context of the insurer’s portfolio. For example, as part of the clearance process insurers need to understand whether they have existing exposure to that client. Or to understand the accumulation implications of this new exposure. Or to retrieve the historical success ratio with the producing broker to assess the likelihood of binding and be able to prioritise the submission accordingly. All this information is required to make automated decisions on the submission, and require pulling data from insurers’ internal systems.
- The third source of data to digitise submissions is external data. External data is required to analyse the risk, for example pulling peril scores such as wind, fire or subsidence. External data also enables insurers to achieve superior risk selection, allowing insurers to identify more nuanced correlations between risk data and loss behaviour, and resulting in a more granular risk appetite.
There is a challenge to integrate these three sources of risk information. All of them rely on being able to identify the client. More often than not the client name contained in the submission is not an exact match to the legal name of the client company. It can be incomplete, it can be the trading name rather than the legal name or can be misspelt. If that is the case insurers will fail at combining these sources, or even worse, will pull information from the wrong client.
So there is one last crucial digital risk flow capability to overcome these challenges and integrate these three sources. This capability is being able to resolve the client name to a unique identifier. This can be, for example, the Companies House Company Registration Number in the UK or the Handelsregisternummer in Germany. Once insurers have this unique identifier they can be sure the clearance or accumulation checks are correct. And that the external data is applied correctly. This step requires advanced Machine Learning capabilities to transform a string of text extracted from the submission into that unique identifier.
Automatically evaluate the risk
The goal of this stage is to assess risk characteristics such as appetite fit, priority or complexity, and to provide guidance to underwriters on what areas of the risk require special attention, which is a key step to enable underwriting by exception.
- Appetite: the first decision is whether the risk is aligned to the underwriting strategy and will result in a quote, or otherwise the insurer will not be willing to quote and will result in a decline. Sometimes as high as 60-70% of submissions do not result in a quote, and as we discussed in the first episode, a key source of waste in current underwriting processes is having underwriters analyse risks that are outside of underwriting appetite. One of the main challenges in appetite analysis is that appetite is often driven by data points that are neither in the submission nor directly available from external sources. A great example is the assessment of the client’s activity, which constitutes one of the main drivers in appetite decisions. In the submission, brokers provide a summary of the client activity that is frequently not enough to assess appetite fit. For example, it might describe a client as “an engineering company”, which is not at enough level of granularity to assess appetite. Typically you’ll need to know whether the client is a construction engineer, a structural engineer or a software engineering firm. Or whether they export to the US or not. This level of detail can only be inferred from the analysis of different sources, such as the company website, news articles or registries of associations such as the Institution of Structural Engineers. Once all the information is available, insurers can use a number of business rules to assess whether a client is within appetite and the submission will result in a quote.
- Submission priority: Priority is the score that defines the level of attractiveness of a given submission. This level of attractiveness can be driven by a number of factors including the strategic fit of the risk (for example, if a certain insurer wants to grow its professional services book), the propensity to bind (for example, using proxies as the broker tier), or the lifetime value of the risk (including factors such as the predicted retention rate and profitability for a given profile of risk). The rationale behind prioritising submissions is twofold:
- Firstly, from a commercial perspective, it drives an uptake in conversion. Typically faster turnaround time drives a higher conversion rate. Making sure underwriters work first on the opportunities that can convert further increases the likelihood of binding that business. It is futile to be working on a submission for a broker with a low historical conversion rate for that given line of business. Prioritising submissions is also a powerful tool to drive underwriter behaviour aligned to the commercial strategy. For example, for the top quartile of most attractive risks, some insurers ask underwriters to call the broker within an hour of receiving the submission to show interest and maximise the chances of binding that submission.
- Secondly, from an underwriting perspective, it provides underwriting executives effective tools to steer the shape of the portfolio. If the CUO wants to grow a certain segment, and those types of submissions are prioritised, the book will reshape accordingly to the underwriting strategy
- The complexity of the risk: this can be driven by factors such as the company turnover, the number of assets or the legal structure of the client. This is crucial to deploy effective operating models where simpler submissions are either automated or sent to more junior underwriters, whereas more complex submissions are sent to the more experienced underwriters. Only by deploying this type of segmented operating model insurers can address one key area of inefficiencies, which is the underutilisation of talent. This happens when you have experienced underwriters looking at simple risks because those are the ones received in their inbox; at that point, there is really not a good choice: they either underwrite it themselves, wasting their precious time, or forward to a more junior underwriting, which means somebody else needs to analyse the risk from scratch, only worsening the quote turnaround time.
- Guidance to underwriters: following the principles of underwriting by exception, insurers want underwriters to have access to all risk details, but guide their attention to the areas that require attention or review. For example, if a client has two CCJs, this fact should be highlighted to the underwriter. Another example would be to use this type of underwriting guidance to effectively deploy a nuanced pricing strategy, where underwriters do not need to rely on reading emails from the underwriting leadership and remembering instructions, but rather is naturally front of mind as they review a given risk. Those types of rules can be easily created once the risk characteristics are fully digitised.
Route the risk to different destinations
And the third dimension of digital risk flows is the ability to route risks to different destinations based on the result of the previous evaluation phase.
In an optimal operating model, insurers need to send different submissions to different destinations based on the risk characteristics and the results of the evaluation phase. A common challenge with the current manual flows is that submissions bounce 2, 3, or 4 times between underwriters before they find the right underwriter to make a decision on the risk. This creates significant inefficiencies, as time is wasted and it only adds to the turnaround time to produce a quote back to the broker and client. And, as importantly, these inefficiencies are often hidden to management as they happen as email exchanges between underwriters in a way that is difficult to monitor and therefore difficult to address.
With digital risk flows, insurers can push the simpler, more homogeneous risks directly to a rating engine or policy admin system for straight-through-underwriting, so that a quote is produced and sent to the broker without human intervention at all. Other simpler risks that require some level of underwriter judgement can be routed to more junior underwriters. And more complex risks or those that require higher levels of underwriting authority can be sent to more senior underwriters.
From a system perspective, there is also the need to route submissions to different downstream systems. Digital risk flows enable insurers to determine what risk data fields need to flow to what systems. The set of downstream systems also varies depending on the result of the risk evaluation. For example, for out-of-appetite submissions insurers might only want to route the risk to a CRM to capture the opportunity and to a data lake for analysis purposes. But for an in-appetite submission, insurers will additionally need to push the risk data to the policy admin system as the system of record, or to an underwriters’ workbench.
Exception handling
One last key component that is crucial throughout digital risk flows is exception handling. Some submissions might not contain all the information required to analyse the risk or underwrite or not all the information will be able to be extracted from the submission at the required level of confidence to fulfil the target schema.
In those cases, digital risk flows require exception handling capabilities to enable a human to review the risk, address the blocker and enable the risk to flow back into the digital workflow. This is referred to as the Human in the Loop capability.
This Human in the Loop allows a human to review the list of submissions that require human intervention and act on them. Some of them will require the human to find a certain data point that the extraction process has not managed to find. Some of them might require the human to confirm a candidate for a given schema field that the extraction has identified but at a lower confidence level than the defined threshold. Sometimes it might even require the human to get in touch with the broker asking for required risk information that was not included in the submission. Once the human acts on these submissions, they flow back through the automated digital risk workflows.
In summary, there are three stages of digital risk flows: building a digital representation of the risk, automatically evaluating the risk characteristics in the context of the insurer’s strategy and routing the risk to different destinations.
