Federated search is a critical first step for information retrieval and knowledge management. But federated search alone likely won’t be enough for most organizations’ enterprise search needs. Let us break it down.

Search once and search it all. Traditional search required your employees to log into different disparate systems to find those nuggets of knowledge they needed. That meant knowing which silo to look at, not to mention passwords and logins. Then along came federated search, which gave users a single interface in which to query across all these silos.

This seemed like the perfect solution as organizations grappled with content across different cloud-based and on-prem sources and ecosystems of record. Yet what the Covid-19 “pause” revealed (and what many IT leaders have known for a while) is that federated search is not enough.

Here’s what you need to know to truly meet your users’ expectations for a true digital workplace.

What Is Federated Search?

First, the basics: Federated search retrieves information from a variety of sources via a search application built on top of search engine(s). One user makes a single query and the federated search engine simultaneously searches multiple, usually disparate databases and ecosystems of record, returning content from all sources for presentation in one user interface to the user.

In complex organizations with thousands of sources in the cloud and on-premise, this is invaluable.

In addition, most federated search engines can also pass along the users’ credentials in the search query. For example, a manager at a tax advisory firm may want to search through previous years’ tax returns – but that manager should not have credentials to access other firms’ forms and information resources. By enabling the search engine to pass the user credentials in an authenticated user setting, the security model of the content is still respected.

Independent Research Firm ReportThe Forrester Wave: Cognitive Search, Q2 2021

Types of Federated Search

There are two distinct  approaches to federated search: search-time merging and index-time merging.

Search-time merging (also known as query-time merging) requires a query federator that sends out a query  to  separate search engines of each data source. The query federator in turn, produces an aggregate list of search results. This solution’s advantages in simplicity need to be weighed carefully with the costs to speed. Because each index is maintained and searched separately, the response time can be significantly slower than the real-time expectations of users.

Index-time merging is a type of federated search system that creates a single, unified index of all content from all data sources, that is searched at one time. The effort to implement an index-time merging solution is high — but worth it for the search user experience when compared to query-time merging. Most tech-savvy organizations end up choosing an index-time merging solution solely for that reason.

What are the advantages and disadvantages of search-time merging?


#1 Speed of Implementation

Search-time merging solutions are simple and quick to implement. Without having to set up a central index, you can quickly get this method set up.

#2 Difference in data sources

One of the reasons why it is so quick to implement is that the indices do not have to be standardized. If your dataset for one index is in one format, and the other is another, it is not an issue for a search-time merging solution.

#3 Always up-to-date

Because each individual index is always up-to-date, the search-time merging federated search engine does not have to consistently go back and index the source to make sure the content and results are up to date.


#1 Poor search relevance

Since each search engine creates its own relevance, ranking results across all content sources is near impossible, so  most organizations end up returning results according to a deterministic attribute such as date, price, alphabetical, etc. .

For example, in some complex manufacturing settings, users  may search by any number of product synonyms or the SKU itself. Catalog maintenance and taxonomy management is much more complex with federated search engines. And with the sheer amount of tuning required, most digital experiences with federated search will still fall short.

#2 Poor search user experience

Most people assume that the search experience only involves the search box and results page – but users today expect much more than that. Taking the prior example a step further, think of the best search experience you’ve had and all of the different capabilities that requires: autocomplete, query suggestions, filtering and faceting … the list goes on. In a federated search scenario, unless all “federates” support each of those capabilities your federated search box won’t be able to either.

The search UX is incredibly important to users; search is what makes our lives easier and often many users’ first reflex (thanks, Google!). In a web search capacity, conversion rates rise by up to 5X when search is optimized. The bottom line: the search user experience is too important to overlook.

#3 Slow response times

Response times become an issue when the response for the results automatically defaults to the slowest search source. This is primarily an issue with search-time merging, as each index needs to be searched and maintained separately.

What are the advantages and disadvantages of index-time merging?


#1 Better relevance

If you are able to bring everything under one roof, you can at least manually (which will be quite painful) tune results for relevance. This is slightly less painful with an index-time merging solution.

#2 Modern search user experience

In a query-time merging solution, the weakest link of your search experience becomes the search experience. So if one data source lacks the ability to use facets or filtering, filters and facets won’t be there in your federated search solution. Index-time merging only takes the content. and then filtering, faceting and other capabilities of the search user experience are built on top, ensuring a best-in-class search experience.

#3 Faster response times

Is it easier to find a needle in one haystack or multiple haystacks? Obviously, the one. Therefore, because everything is centralized in an index, it can return results faster searching one index as opposed to searching multiple with a query-time merging solution.

#4 Inclusion of data that does not have a search engine

Even in today’s day and age, not all content has a search engine – and that makes it even more important to bring into a central index.


#1 Speed of implementation

Index-time merging solutions take a longer time to implement. Creating the index, navigating connectors, ensuring the single data model is set up – it’s a time and resource investment.

If you try to build this yourself, this becomes a much bigger problem. If you can buy an established product with connectors and a platform that will normalize data across sources, you can avoid this disadvantage.

#2 (Potentially) slower update times

There may be a delay in updating the content in your central index from the content sources.

This is why you need a solution with connectors. Connectors can have an incremental update feature so every five to ten minutes, you can see incremental updates of changes. Think quick updates all of the time – but this is very difficult to build yourself.

Challenges With Federated Search

Users today expect more than federated search. They expect relevance.

Why? As knowledge workers and consumers, we have been trained by digital first-companies, from the most engaging ecommerce site to our favorite streaming sites, to have the most relevant information queued up front and center.

But, as stated above, relevance is near impossible with query-time searches, and while possible, highly labor intensive with index-time searches. There is a solution to reduce the labor in index-time searches — combining index-time merging with artificial intelligence — more specifically, machine learning.

Machine learning analyzes user behaviors signals and the content types themselves, to better understand user intent and context.

To understand how Coveo approaches this, request a demo.

Contact UsRequest a Demo

How to Go Beyond Federated Search

The Coveo Relevance Maturity Model lays out a clear framework for understanding the maturity of your organization’s digital relevance and enterprise search. While index-time merging is a step above siloed search where each source is searched separately, it still fails to meet user expectations for predictive, intelligent digital experiences. As you move up the CRMM, the value to the business multiples.

Coveo Relevance Maturity Model

Level 1 – Secure, Unified Ranking

Unified ranking applies a single rank profile across all of the results, eliminating the distinct ranking and relevance models from disparate search engines. Many approaches to federated search through index-time merging can make it to this stage at least. Any relevance tuning is through manually-imposed rules.

Level 2 – Content Navigation

While unified ranking ensures the most relevant results are at or near the top of the list, users typically want to be able to filter and further dissect results. Configurable rich facets provide the very first step of personalization for users, so that they have natural classifications against which to narrow down the initial search results to navigate through content to more tailored, personalized results.

Level 3 – Tunable Relevance

Organizations at this level are providing information to users based upon systematic tuning of results through data models. Relevance is tuned automatically, through applying query tuning models, rather than a complex set of individual rules that quickly become unmanageable, and impossible to scale.

Level 4 – Contextual Relevance

Tunable relevance models are further refined by receiving additional contextual inputs at the point of the search query or related behavior.  Examples would be, what users click on, what they don’t, query rewrites, etc.

Organizations at this level are providing highly relevant information to users that is auto-tuned based on models and user-specific cues, to rank results uniquely to each individual, for a higher level of personalization.

Level 5 – Contextual Suggestions

Further related information is provided proactively to users based on an understanding of what they’re trying to achieve. This information is not directly being sought, but is likely to help them.

Best-in-class organizations at this level are also able to recommend names of de facto expert individuals within their organization or community who are authoritative on the given topic.

Level 6 – Self-Learning Predictive Recommendations

At the highest level of relevance maturity, organizations are able to discern their users’ likely intent, by matching usage analytics and behavioral data, such as web click stream data, third-party purchase preference data, and making recommendations as to what content they are most likely to want or need. Machine learning algorithms auto-tune such recommendations and rankings, to maximize specific business outcomes such as shopping cart conversions, or support case deflection, enabling true one-to-one user engagement and upskilling.

Ready to take the next step? Discover how you can have relevant search on day one.

DiscoverCoveo AI-Powered Search


Share this story:

About Rachel Schultz

Rachel Schultz is the Content Marketing Manager at Coveo. She blends her background in journalism seamlessly with her B2B marketing expertise and obsession with data to create compelling content for the Coveo community. When she’s not working, you’ll find her reading (strictly non-fiction), hanging out with her puppy or taking in all NYC has to offer with her husband.

Read more from this author