Blog: Artificial Intelligence for ERP – Toolbox
I previously wrote about the challenges SAP has been having getting customers to make the switch to the newer technology offerings, and some of the reasons for this reticence. I also called out the fact that SAP’s next big bet is that AI/ML through Leonardo may be a bit of ‘last ditch’ effort to hold on to and grow their business. In this second part I will explore some of the application options most relevant to ERP and CRM and explain why I think some of the approaches may or may not work.
When RPA and OCR are suggested as AI applications
When we look at the appetite for artificial intelligence and machine learning in the ERP and CRM space is most often couched in robotic process automation based processes. You’ll see this called out by vendors who offer RPA’s but as is pointed out by William Vorhies in a piece on DataScienceCentral, “RPA at best is a way to coordinate the actions of various AI/ML inputs, and more typically [is] a system of automation driven by ordinary human-defined rules.”
To my mind if you’re a genuine AI/ML developer you won’t credit an RPA as a real application of AI/ML irrespective of what the vendors might say. So I think on face value, while we can say that automation and RPA, in particular, are rule-based and nicely accelerates data collection, management and validation, it isn’t really doing anything except following a tree of CASE or IF… THEN… ELSE… instructions or rules.
In the past life, I have even see OCR models and technology be posited as using AI and ML to improve the quality of the OCR of scanned documents, but even this I am not sure, qualifies as a good example of AI/ML in action.
What is probably most interesting, is where those RPA implementations are being done, almost certainly, some if not all of those ‘could’ be candidates for an AI-based solution that trains itself based on periodic feedback and decisions by real people.
One of the most obvious approaches that could be considered for Artificial Intelligence has to be around image recognition and category management and classification.Such capabilities have just started to appear in other ERP applications and as such, it is a new phenomenon, but the possibilities artificial intelligence adds to the ERP systems must be said to be unlimited.
Automation of data entry and streamlining task-based workflows are obvious on face-value if we look at those data entry examples they might start with duplicate record checking say for employee record creation based on a profile image. The use of fuzzy matching logic in interactive data entry helps in not only being more intuitive but also using a broad swathe of search and match criteria for new records against existing data stored in the system.
Microsoft recently revealed what it referred to as artificial intelligence through Cognitive Services as a part of Microsoft Dynamics NAV 2018 ‘Tenerife’.
Cognitive Services which was part of Microsoft Project Oxford comprising intelligent API’s that work across the platform and deliver intelligent data decisions. The examples cited include automatic face recognition, speech and image processing and Language Understanding Intelligent Services (LUIS). For HR and HCM applications in ERP installations this could be useful, but perhaps less useful for the likes of SAP as they push all their HCM efforts in the direction of SuccessFactors. LUIS and speech handling may also be useful for handling inbound order entry and inquiry services, all of which today are largely handled by customer self-service by way of complex order entry screens.
Personally, though, I can’t see this scaling in relevance beyond the one offline items, imagine ordering all your groceries verbally for a weekly home delivery shop? It would be a cumbersome nightmare when you batch up all the items you want. But maybe I am not thinking big enough here…
Where the voice and video services may also be helpful is in the back-office in other areas like inventory management, where a picker or packer actually yells at a mobile device rather than tapping a screen or keyboard. Even this though, likely will use voice recognition, which like optical character recognition is somewhat simplistic. Training of these applications improves their recognition of commands, words and phrases but where is the real intelligence of responding in a human-like way to what is being heard untrained?
Mid-tier ERP – EPICOR claims manufacturers want digital assistant-like interactive interfaces like those found on mobile phones and smart home devices. They launched the Epicor Virtual Agent (EVA which includes built-in artificial intelligence designed to simplify interactions with ERP systems, as well as automate tasks to speed up operations. Powered by Natural Language Processing (NLP), users can access EVA from mobile devices and the application delivers targeted information.
Examples cited, include visualisation of spare parts with the key information and suggestions for ‘next best action’ options. Of course, such requests would also serve up pricing or availability. An additional aspect is delivering notifications, alerts and reports based on triggering events in the systems or peripheral landscape says from machines and IoT sensors. Combined with ERP data, you start to see the evolution of ERP based eco-system that pushes insights and suggestions rather than waiting for someone to run a report or inquiry.
Such recommendations and prescriptions could feed very nicely into scheduling preventive maintenance for machines or taking systems out of production or offline for repair.
The obvious CRM options are around making recommendations on alternates, complementary options and cross-sell and up-sell recommendations in real-time – in much the same way that we see Amazon shopping carts tell us that customers bought x and y while buying ABC. AI-powered recommendation engines based on behaviours and market basket analysis holds great promise for ERP and CRM just as they have shown themselves to be useful in routine online shopping cart management.
AI for MRP
Take this down to the inventory management for raw materials and manufacturing output it becomes easy to realise that in actual fact artificial intelligence in many respects could be the greater displacer of traditional MRP even.
Artificial Intelligence can optimise the MRP decision process
We see in early rules based Expert Systems where they were used for planning that they were often nothing more than branch and leaf decision trees based, often on mathematical and arithmetic calculations that could tell you that things would be late or complete early based on dependencies. These are perfect opportunities for training artificial intelligence applications especially for Just in Time (JIT) and KANBAN production. The AI models can tweak things like recipes, batches and lot sizes as alternatives for inventory model selection and in turn, learn how to do them better based on results and feedback.
Many of the criticisms and complaints about a lack of precision and bluntness in MRP could be prevented by applying AI techniques to the MRP process, often an integral piece in the ERP puzzle.