AI-powered recommendations in the back office

They say that if you build a chair, people will use it to stand on. As an API-first company, we know this to be true: our customers continuously extend our Search API beyond what we had in mind while building it–sometimes in surprising and inspiring ways. It’s the same for Algolia Recommend – when we built it, we initially targeted product recommendations for ecommerce customers. However, our Recommend API applies to many other use cases. Customers have asked about providing recommendations for media streaming, news and blogs, and job postings. The list is long.

Netflix’s VP of Product Innovation Carlos A. Gomez-Uribe and Chief Product Officer Neil Hunt have written on the broad scope of recommendations:

“Humans are facing an increasing number of choices in every aspect of their lives—certainly around media such as videos, music, and books, [but also] vacation rentals, restaurants, … health insurance plans and treatments and tests, job searches, education and learning, dating …

We are convinced that the field of recommender systems will continue to play a pivotal role …

This article is more humble than that: we imagine how you can integrate recommendations into your back office systems and processes using our Recommend API. We describe the high-level benefits and functionalities for the end-user, addressing both technical and non-technical readers.

People never do just one thing at a time – How to leverage multi-tasking

We rarely buy only one item when we shop or do anything without combining it with another set of actions. Actions come in pairs or multiples. Businesses can leverage this with AI-powered recommendations.

Recommendations capture and model the regularity of our behaviors, providing insights into (a) what actions we consistently perform together and (b) how we can use these frequently combined actions to change the way we do business.

While this article is about recommendations in the back office systems, we’ll set the stage by first considering a more known use case: Shopping. We shop for milk and invariably buy too many items for us to carry in our hands. We buy three different fruit when we only need one. We go into IKEA for a few accessories and come out with a cart full of accessories for the house.

It’s called the IKEA effect. We shop for 1 and buy 10. 

The main question for IKEA and all supermarket chains is: Which of these additional 10 items get “frequently bought together”? 

Leveraging habits and patterns

AI-powered recommendations answer this question by detecting and leveraging these habits and patterns, for the benefit of customers (quickening their shopping experience) and businesses (inviting customers to buy more items). Product and content recommendations are now consistently proposed as “Frequently Bought Together”. Amazon makes product recommendations when we purchase or add items to a cart; Netflix makes content recommendations as we surf through their offerings; Spotify creates playlists based on frequent combinations of songs. 

(Note that we focus primarily here on frequency analysis. However, recommendations are also about finding related items. We’ll mention that as we go along, but our focus will remain on frequency.)

Recommendations in the back office

While recommendations have become pretty basic to ecommerce and popular media services, frequently performed actions can also apply to doing anything in regularity. Frequency analysis can function behind the scenes for any back-office systems (like Salesforce, SAP, ERPs, CMSs) by feeding analytics data into a recommendation-modeling engine and making knowledge management and common workplace tasks more efficient and uniform. Recommendations can lead to more coherent organizational knowledge, intranet searching, and AI inventory management. The same customer-facing B2C benefits can be applied to B2B, customized to the unique aspects of partner relations, procurement, and distribution.

That’s because, at base, recommendations are set up to capture and leverage any pattern of behavior that arises in any business domain – finance, customer relations, human resources, ecommerce, media, SaaS, services like hospitals and education, travel. 

An example: Recommendations and Salesforce

To illustrate this, we’ll use the example of Salesforce to manage accounts and customer relations. Note that we could have chosen any back-office system, such as software that manages inventory, legal or financial processes, HR, SAP, ERPs, CMSs, and so on.

What follows are summaries of what you can do. Some of these will be taken up in later blog posts, with more detail and the actual solution, including the code.

Advising employees about customers (B2C) and suppliers (B2B)

How can you blend any Salesforce analysis with a Salesforce B2B UI to AI-powered analysis and recommendations?

  • Imagine that you’ve built a customized Salesforce UI that manages your B2B contacts, and it contains the latest transactions of individual suppliers. If an employee wants to improve a particular supplier’s buying experience, they can use recommendations based on the sales history of other suppliers (using anonymous data). Though this is often done in B2C, it can work just as well in B2B, where the consumers are suppliers of goods or services. 
  • A recommendation engine can capture the details of every transaction, whether from online transactions by customers or suppliers, in physical stores, or from the back end B2B (between suppliers). Thousands of transactions from these sources form a robust and very accurate recommendations model of your business. 
  • Next, by displaying these recommendations on the Salesforce UI, employees can reach out to suppliers to bring potentially missed opportunities to their attention. This avoids large amounts of time and money on manual analysis and simplifies your business model with a more intelligent and targeted inventory and informed clientele.
  • Finally, you can apply recommendations for process automation, knowledge management systems, inventory changes, and distribution channels by having the system feed recommendations into the back office purchasing software. Employees in charge of inventory can accept, reject, or modify these recommendations to rethink the catalog.

Recommendations for employee UIs and back office workflows

Let’s say that your customer relations staff (Sales, Customer Support, Account Managers, and other such customer-facing employees) perform a set of actions – the same set of actions every day. If you feed these actions into the recommender engine, the system will be able to group the actions frequently done together. There are several benefits to this:

  • Creating a better UI. The UI can be re-configured to place these actions closer together on the screens or in the workflow. The UI can integrate search and recommendations with a federated search experience. 
  • User testing. You can automate user testing to create the best UI by determining that certain frequently combined actions are difficult to accomplish.
  • Personalized experience. By combining personalization with recommendations, individual employees can follow a personalized workflow that makes sense for their role in the larger business context.
  • Workflow efficiency. A recommender system can recommend workflows that follow a guided UI procedure or wizard focusing on the most frequently combined actions.
  • Task automation. You can automate some of these steps by analyzing which parts of these actions can be done more efficiently by computers. Using recommendations to help break down each step can help rethink the whole process from end to end.

Adding search and AI-powered recommendations to integrate back office systems and knowledge management

If you’ve decided to “searchify” Salesforce–that is, you’ve made search integral to your salesforce UI and user workflow–then you can leverage Algolia’s AI as it relates to search results. You can combine and integrate your back office systems with a multi-display interface, called federated search, that incorporates recommendations along with other information on the same screen. 

Recommendations mean you can start displaying what employees regularly consult while they search for related items. 

  • For example, by allowing users to search a financial database from a customer screen in Salesforce, you can start to combine Salesforce with your own internal financial accounting system. Think about the scenario where several employees regularly consult contractual and other information in and out of Salesforce. If you have not integrated search, your employees would have to change applications to find what they need. By integrating distinct systems into the same platform (Salesforce or other), you can start seeing patterns across systems to see what employees are doing in tandem. 

The benefits could be:

  • A more efficient and accurate way of doing business.
  • Redoing the interface so that the information from different systems appears on the same screen. For example, showing the contracts from the legal database on unrelated screens like a customer’s call history.
  • Rethink the platform itself – that is, the set of screens that make up the larger application. Make it easier for employees to consult a customer’s accounts while looking at the customer’s profile on another screen at a single click.

Conclusion – curiosity and grit, and a call to action

There’s a reason we say “burger and fries” without thinking twice. Or “wine and cheese”. And not so often “burger and wine”. But we don’t always know why, except that, well … “these are words that go together well”. Patterns exist. The jackpot is to uncover lesser-known habits and profit from that newfound knowledge. It just takes some curiosity and grit. 

For example, an often unmentioned factor in any advanced use case is the developer. If given the time, motivation, and curiosity, any developer can do practically anything with a general-purpose recommendations API. Companies often focus on the commerce-driven use cases – product and movie recommendations, measured by increased sales and conversions. While there’s plenty of opportunities there, there’s also an untapped universe in the back office, where the benefits may appear less monetary or less obvious, but with some curiosity and grit, it doesn’t take a whole lot of work from a good developer to integrate AI-powered recommendations into just about any back-office data or workflow – with enormous benefits.

About the authorPeter Villani

Peter Villani

Sr. Tech & Business Writer

Recommended Articles

Powered by Algolia AI Recommendations

The anatomy of high-performance recommender systems – Part V

The anatomy of high-performance recommender systems – Part V

Paul-Louis Nech

Paul-Louis Nech

Senior ML Engineer
How any business can benefit from personalization and recommendations

How any business can benefit from personalization and recommendations

Marie-Laure Thuret

Marie-Laure Thuret

Technical Product Manager
What is a product recommender (or product recommendation engine)?

What is a product recommender (or product recommendation engine)?

Catherine Dee

Catherine Dee

Search and Discovery writer