During the past weeks I had a couple of observations and conversations that lead me to thinking that sometimes software vendors underestimate the power that their machine learning based systems could have to improve the lives and experiences of employees and customers.
From various vendors in various lines of business, from process mining and automation via application performance monitoring to vendors of conversational AI and pretty much everything in between I hear something like the following:
“Our machine learning based system continuously analyses the process/interactions and detects anomalies. From there on it identifies the patterns and can make suggestions how these anomalies can be avoided or resolved.”
Of course, this is paraphrased, but you get the meaning.
Here are some examples.
A real life scenario that I once encountered is as follows: A global B2B e-commerce solution using synchronous pricing is set up in a template approach. It is using one single ERP system for pricing. This is a pretty common B2B scenario, as it is often not feasible to replicate all prices to the e-commerce system, due to the sheer amount of product – customer combinations that are possible.
In this scenario, adding a product to the shopping cart involves multiple calls from the e-commerce solution to the ERP system to establish the price.
The pilot country site is close to the country that hosts the ERP system. Implementation of the e-commerce solution happened in the country that hosts the ERP system for all e-commerce sites. Deployment into the target countries can happen only after the testing phase, which is clearly suboptimal.
Adding a product to the cart is reasonably performant for the pilot country.
The second country is on the other side of the world.
After deploying the site to this country, it immediately became clear that adding products to the cart is far too slow to be acceptable.
Now, one could think that the development team should have seen the issue coming. It hadn’t. Most application performance management (APM) solutions wouldn’t have, either.
A simpler issue is the automated scoring of opportunities, which gives a value, a score, that represents a systematic estimation of the win likelihood of assessed opportunities. Or ticket routing based upon (bad) sentiment.
Another conversation was about chatbots. “Based on the past customer conversations, our AI will identify bots that should be built”.
Do not get me wrong, these abilities are already very valuable.
Just that they are too short of a jump. These are not enough!
Let me tell you why.
We deploy technology to get solutions for pertinent problems, aka challenges. The solution is not to inform someone that there is a problem but to either avoid it altogether or to actually solve it.
Building on the examples above, better solutions would look as follows.
An APM solution measures application performance to identify potential bottlenecks, ideally already before deployment, e.g., in a test phase. Most of them do that fine. Some of them even have predictive capabilities in a sense that they can warn before a critical situation arises. Even better, an APM system that predicts that excessive runtimes will arise for operations in given circumstances before a significant usability incident occurs would be incredibly useful. The scenario above was caused by a high number of calls between the systems in combination with a high latency due to the long distance routing. Situations like this one can be identified and therefore solved before they occur.
Scoring of opportunities already helps salespersons, now doubt. Still, additionally offering the actions that could be performed to improve the likelihood for successfully closing opportunities is far more helpful. Or in the case of the ticket routing based upon (bad) sentiment: suggesting which actions could be taken to improve the situation would help many agents.
In both cases some of the necessary activities could even be triggered and executed automatically.
Here are some examples that look even further and are partly already in place.
The best known one is probably delivered by Oracle. Oracle has enhanced their flagship RDBMS to something that the company calls the autonomous database. Oracle defines an autonomous database as “a cloud database that uses machine learning to automate database tuning, security, backups, updates, and other routine management tasks traditionally performed by DBAs. Unlike a conventional database, an autonomous database performs all these tasks and more without human intervention.”
In essence, Oracle delivered a database engine that uses available log- and meta data and a trained model to identify incidents to resolve them, freeing DBA time for more complex tasks.
Similarly, Ray Gerber, Chief Solutions Officer at Thunderhead envisions and works on an autonomous customer journey orchestration engine. This engine essentially discovers the actual journeys customers follow and optimizes them in real time to help the customers achieve their goals. This way, it also drives engagement and creates value for business. It bases on a constant analysis of customer journeys. With this capability, journeys that are offered to the individual customer journeys are (and stay) optimal – considering the business goals and objectives.
These are, admittedly, quite sophisticated solutions, but this is exactly the point. The needed technologies are often there and can be adapted and used.
The same principles can be applied to process mining and -automation. Process mining tools identify choke points in business processes. They also identify repetitive tasks in processes. They can see the data that is used, so they can start to auto-create the necessary workflows that take away tedious tasks from users.
The last example that I want to showcase, are chat bots. The existing bots with their capabilities are known. It is also known where these bots need to hand over the conversation to human agents because they reach the limits of their skills. Mostly, these human-to-human conversations are known, too, as they still happen in a chat environment. With that, solutions to further problems and the way to get there, are available. Even the conversations.
So, a system that builds the corresponding bot, or at least a prototype, is conceivable and should be delivered. Agents will be grateful for yet another simple thing that is taken off their plates.
Is all this simple? No. A silver bullet? Of course not. Possible? Yes, with some constraints. Helpful for users and customers? Sure as!
So, I am calling on software companies to embrace what I would call autonomous automation and to go for it.