Designing Enterprise Chatbots - Dear Enterprise chatbot: what is our CEO’s annual salary?

This morning, my brokerage sent me a survey asking me about their new Chatbot Ted (or was it Suzy?). The survey was simple enough. The 5-6 questions asked me if I used other chatbots and virtual assistants like Siri, Cortana, Google Assistant, etc. It also touched on security concerns I had about sharing personal details with a the brokerage's Bot.

The past year has seen tremendous advances in chatbots integrated with machine learning, cognitive computing and Artificial Intelligence (AI) techniques. The bots are moving from quirky, esoteric tech tools that amuse us to serious business productivity tools that enhance the UI/UX design and usability of systems.

Organizations are evaluating bots as the front-line of customer service to address routine queries before (expensive) humans are engaged. Most major banks, brokerage houses, Airlines and eCommerce companies have rolled out cutely named Bots and virtual assistants that pop up when we try to engage with their portals. Many of these bots are being designed to take on basic human interactions like the recent demo of ‘AI Assistant’ by Google's Sundar Pichai that went viral.



In the demo, Google showed off its Assistant having a human-like conversation with folks at a hair salon and a small restaurant. The rather simple use-case also highlights the complexity of human communications, even for seemingly mundane tasks like making reservations that we take for granted while dealing with other humans, even those who have different sounding accents. Enterprise Chatbots that blur the line between human and system interactions, are also starting to appear in the corporate world.

Motivated by such viral videos, some CxOs, are asking their internal Business-technology teams to evaluate Bots for their internal users. Typical corporate Use-Cases include service help-desk functions for IT, HR, payroll and other shared services. Incidentally, these are also areas where organizations have been attempting to minimize manual interactions by enabling employee self-service techniques including searchable databases, FAQs, personalized portals and internal social-networking.

Technologies to power chatbots include commercial and open source tools that enable Machine Learning (ML) algorithms and natural language understanding to learn appropriate answers to user queries over time. While enabling and integrating chatbot tools in a corporate landscape can take some time and effort, training the Bot to learn the corporate context may require a lot more effort that shouldn’t be underestimated.

Hi HR-Bot, what is our CEO’s annual salary?

Design and training an enterprise chatbot requires functional context and access to enterprise data; data that may exist in silos. An FAQ of typical queries to the sales help-desk may include queries about monthly targets, details of product catalog, positioning and even sales data for the prior months. Such data may be business sensitive, and restricted even within the teams. For example, the sales team in the mid-west may not want their sales numbers exposed to other account teams, leave alone published to rest of the organization without aggregation or masking.

The design for an enterprise chatbot may also have to restrict information based on roles of the person querying. It should recognize that the senior executive asking if “margins of ACME account have improved since last month?” is authorized to review such information.

An account management team I once worked with was paranoid about the queries in a FAQs that could expose 'salary ranges' for some roles. Teams like that will certainly not appreciate queries to a Bot that ask "what’s our CEO’s salary?" Never mind the fact that most of us can just google the information from public sources; and a well designed bot enabled by machine-learning will eventually learn to search for that information on the internet.

The design of a machine-learning enterprise Chatbot also needs to guide (read: control) it to stay in the context of its enterprise domain. Many of us also continue to learn from experiments in the social media; like Microsoft’s ‘innocent’ chatbot, Tay that was ‘corrupted’ by Twitter and digirati in less than 24-hours.

Advances in AI, ML and NLP are pushing the envelope, and promising productivity gains by enabling self-service. A well-designed Chatbot, enabled in a specific functional context – like an IT, Claims or Benefits service desk – can aid productivity and also employee engagement while minimizing manual effort in responding to queries. However, given the current challenges in human interactions with systems, chatbots from Banks, brokerages and eCommerce companies are not being designed to be truly Machine Learning tools. At least not yet. While they respond cutely to routine questions, they don’t really ‘learn’ from queries posted by random users on the web.

Bottomline:  While there is a lot of promise that cool enterprise Chatbots hold, wider adoption in large organizations will also have to go hand-in-hand with organizational design.

---

Thanks for reading! Please click on Like, or Share, Tweet and Comment below to continue this conversation or share your favorite 'trend to watch' | Reposted on my linkedin blog | 

Rolling out Robotics, Automation and RPA? Prepare for the initial heavy-lifting!

I recently posted a couple of requests on my LinkedIn feed looking for people with expertise in Robotic Process Automation (#RPA link). I got a few good leads and some of my peers reached out asking what I was doing in this space. I figured it was a topic for a Pulse article, building on the theme of Digital Strategy Execution that I had blogged about earlier (link).

Automation, especially emerging tools enabled by RPA is getting a lot of attention among technology and business executives.

The Use Cases for automation using RPA tools are certainly compelling. Most organizations are likely to encounter gaps in existing processes that evolve over time. Many require users to perform repeated tasks like entering data from printed document or validating data in one system against other systems. RPA tools can help automate mundane, repetitive tasks. Putting together a business case to invest in tools and resources sounds compelling. However, the heavy lifting begins after the business buys into the promise of RPA. 

Case in Point

As an Enterprise Architect responsible for Corporate Functions, I get involved in technology transformations and reviewing new tools and technologies. The finance business unit was reviewing quick-win automation techniques while planning for a major system re-engineering.

The group receives thousands of invoices from vendors and suppliers across the globe. While they had integrated the invoice processing with a few large vendors, the long tail still involves thousands of invoices coming in as faxes and emails with attachments. The business engaged an outsourcer whose team used a semi-manual approach to process the invoices. They take inputs from the mails and reconcile the invoices against the original Purchase Order in the source system, after which the invoices are cleared for processing and payment in another financial system. The process has many variations; for example, some of the invoices are for single items while a large number have multiple line-items.

The long-term roadmap involves re-engineering the process to automate the integrations and workflow, which is likely to take more than a year. In the meantime, the team wants to reduce the manual processing that is fraught with errors.

This jumps out as a textbook-case for automation using RPA Robots (Bots); after a quick review, the business sponsors signed off on automation. RPA analysts were engaged to review the current steps, and began to work with functional SMEs to plan for automation.

And then the wait began; and some began to wonder why.

The reason was obvious. This was the first time the business unit was rolling out an RPA solution. Before the initial euphoria died down, we began reviewing the design to enable the foundational elements and infrastructure.

The details of implementing robotics lie beneath the iceberg

Our experience thus far has been in line with the data quoted in this PWC report (link), that gives a dose of reality. The report highlights how “The enterprises do not always feel that they have received sufficient information about how long time it actually takes to create just the right foundation.”

While the Business Analysts and RPA developers review the process to automate, RPA Architects and design teams need to review some of the guiding principles. Typical questions that need to be addressed include:

• Procuring tools – How long does your procurement team take while approving a new vendor solution? If you happen to be in a large enterprise, you will have to work through your ‘procurement process’ before you acquire the first licensed copy of the chosen tool (UIPath, Blue Prism, Automation Anywhere or others) 
• IT Governance - Does your IS policy allows system accounts for non-human users? What is the process to enable application specific access for the Bot accounts? A Bot will require a network account, user-id, application and other credentials. The first time you introduce Bot accounts in your landscape, you will have to review the policies that govern accounts used by such non-humans. 
• Business policies – An invoice processed for payment triggered by a Bot will require it to login into the Financial system. The Bot will need credentials similar to a human triggering invoice-processing. Does your corporate compliance policy allow for non-human accounts to login and process transactions on your financial (or HR, or Legal) systems? Do you have the same level of tractability for activities performed by Bot-accounts? Do you plan to onboard and offboard Bot accounts periodically? 
• Basic SDLC – How do you plan to test your Bots? Do you have the systems, environments and test data to validate the Bots? Most business users at enterprises require new systems and processes to be rigorously tested before it goes live. Business stakeholders will expect the same rigor in validating the operations of a Bot before you let them loose in your landscape. 
• Managing Bots – Who is going to operate and monitor the activities of the Bots? Robots can be scheduled to run periodically or triggered by events. However, just like other systems, they will periodically fail or generate exceptions. You will need to extend the support and service model to the operations of Bots.

Some of these questions may sound trivial, and may be a non-issue in smaller organizations or startups. However, stakeholders in larger organizations may not articulate such Non Functional Requirements (NFRs) but will nevertheless expect attention to detail when it comes Bots that operate with live financial, customer, employee or other corporate data.

Bottomline: Before the first time you decide to roll-out production grade RPA solutions that begin to process your enterprise’s live financial, accounting or procurement data, you need to analyze and agree on the foundation elements of automation.
________________________________________
Thanks for reading! Please click on Like, or Share, Tweet and Comment below to continue this conversation or share your experiences with AI, Automation and Bots.