I wrote this blog post because as a Manager of Products and Solutions with a Systems Integrator that specialises in the field of Document Process Automation, I have recently completed a project to evaluate a SaaS Invoice Processing solution that uses machine learning. This is meant to be a short update and if I receive sufficient feedback, I would be happy to produce a more in-depth analysis.
I’ve spent much of the last decade working with various document capture solutions, honing my skills during that time to a level where I believe I can tackle most projects given to me.
Like many things in life, automating the extraction of data from an unstructured document is a matter of balance. If you deliver too “light” a project, you run the risk of the customer being disappointed at the level of automation (everyone wants and expects “straight-through automation” right?). Embark on too “heavy” a project and you potentially run into a myriad of issues such as false positives, an over-complex project that is difficult to maintain, and worst of all, over-runs on the budget for time and cost and maybe a project that never quite delivers on its promises.
Much of the setup of an automated Document Processing system is the expectation-setting exercise that our Sales Consultants undertake with the client when the business case is being calculated.
After spending 6 months training and familiarising myself with a leading SaaS solution for Invoice Process Automation, I feel I’m well placed to summarize the differences between traditional, on-premise solutions and the SaaS solution I’ve worked with:
1. SaaS eliminates all the infrastructure setup – Let’s cover the obvious one first: no servers, little liaison with customer’s IT, no thick-client installs, no scanner setups either.
2. Fast and Standardised setup via configuration screens – In place of a “project workbench” or technical setup environment for a traditional, on premise system, SaaS solutions are configured via browser screens, all intuitive and straightforward. The downside is that although the choice of settings is extremely large, the configuration is somewhat prescribed.
3. Auto-learning of invoice layouts – Access to an enormous “library” of learned layouts. This is the biggest difference. We have deployed traditional on-premise (i.e. non-SaaS) document capture systems with “machine learning” and essentially, these combine some pre-set rules with example invoices that are learned by the system. The learning process consists of a technician or experienced user manually keying the fields to be learned so that the system can analyse where on the document the data is to be extracted from. Once a user performs the same manual indexing on the next few examples of the “to-be-learned” invoice, the system will then perform this indexing automatically. This process is sometimes referred to as “learn by example” and, in my experience, it works well. The SaaS solution I’ve evaluated also requires that the end-user manually indexes any missed fields, the learning happens behind the scenes.
The big difference in the two systems is the “learned library” of invoice layouts. The SaaS system anonymises the invoices and submits them to a global library of invoices containing examples from all customers using the system. This for me represents the power of a SaaS solution. Customers benefit from the experiences of the entire customer group and the more the system is used, the larger this library becomes and the greater the benefit to all users of the system.
Crucially, unlike traditional premise systems, the SaaS solution requires no specific training on a new customer’s supplier invoices. In other words, new customers can be up and running within hours of deploying the system because the system is already setup for most of the invoice layouts it’s likely to encounter with a new customer. Personally, I really like this approach; much as I enjoy the pre-deployment setup and analysis required by traditional systems, I’ve often had the feeling that I’m simply re-inventing the wheel for each and every project. Having decided to automate their process, most end customers simply want to get on with it and to realise the benefits straight away; this is certainly the case during the Covid19 pandemic.
4. Performance – Most of the invoice processing solutions delivered by Open Connections fall into the range of 10,000 – 200,000 invoices per annum, i.e. accounting for financial period end or seasonal fluctuations, a peak volume of 1,000 invoices per day. At that level and based on my experience, the data extraction performance between SaaS and on-premise systems are very similar. Naturally, an on-premise system, configured with contemporary server technology should be faster, but the reality is that the brute data-centre power behind SaaS solutions means that SaaS performance won’t be lacking.
5. Access Control and Data Privacy – The SaaS solution conforms to Data Processing standards for privacy and access control is good. As a qualified and experienced IT professional, I don’t see any issues in this regard.
6. Reporting and Analytics – Naturally, with a SaaS solution, you need to accept that because you don’t have direct access to the underlying database tables, reporting will be what you get out of the box. Having stated that, I was impressed with the scope and performance of the dashboard and reports. It certainly exceeds what in my experience, most AP Managers need, to monitor their Invoice Process workflow.
7. Additional Capabilities – Unlike the classic on-premise Invoice Processing solutions, SaaS solutions often have additional modules such as Invoice Approval, GL coding and three-way matching exceptions handling. I found these to be good enough and the ability to access approvals via mobile app is super-compelling as it allows approvals to be made when out of office. Customers can select the modules they need; it may be that the ERP system’s GL process is sub-optimal and so this can be switched to the SaaS solution.
In summary, having analysed and tested a leading SaaS Invoice Processing solution and after years deploying conventional on-premise solutions across a range of sectors, I must admit to being “sold” on the SaaS approach. For many end customers, a SaaS global machine learning database is definitely the way forward. The convenience, cost, risk, and performance/accuracy is every bit as good and in many situations is better than traditional on-premise systems.
If you would like more information, please contact firstname.lastname@example.org