How much does it cost to replace an existing search application? This question inevitably closes every initial email or telephone call from a prospective client. As with so much of search application management, the response is the same: “It depends.”
In this column I will set out some of the dependencies, but if you are expecting a definitive prediction on cost, prepare to be disappointed. Also know: the following dependencies assume use of a cloud platform and a commercial product. Open-source search runs on a different model.
License proposals are based around some unit of performance, which is used to scale the license fee to the customer. Each vendor has its own approach, such as seat-based, page-based, repository-size based or query based. Vendors are trying to identify a unit that they can translate into index and query resource requirements. As a result, comparing license costs across a group of vendors is (as they intended!) difficult.
Search applications are computationally intensive for both index and query processing. It is not just about creating the initial version of the index but also the effort of updating the index, especially in cases where the timing from content creation to indexing is critical (e.g. within an hour rather than overnight), and the scale of the index rewrite.
Many organizations have inadequate knowledge about the scale of the repositories to be indexed, especially if the search application is much wider and diverse in content type than the existing application. Server utilization is a poor indication of the number of documents and the amount of processing therefore required.
Future demand and likely levels of use is another important consideration. You don’t want to find you have inadvertently broken through a glass ceiling into a higher (or worse!) price bracket. Search costs rarely scale just by index size.
Related Article: I Can’t Get No (Search) Satisfaction
Levels of Service
An important element of the cost equation with cloud environments is the acceptable uptime from the host service. While 99.9% may seem good, work out how many hours of downtime that might be. Maintenance downtime is also a factor and can grow challenging if your organization works globally and there are no convenient times for this work.
Going up to 99.99% can become significantly more expensive. Another factor is disaster recovery — how will the vendor support the required recovery point and recovery time requirements?
Processor charges add to costs as do support charges for monitoring and bug-fixing. Defining the need for and scope of the latter is a good topic for a day of round-table meetings, in particular to reach an agreement on the red flag response requirement. The response latency is another variable, which is a function of both the server and network. A decent starting point is 200 milliseconds on a query — anything over 500 milliseconds is unacceptable.
Type-Ahead, Spelling Correction and Auto-Suggestions
Every search application supports type-ahead, spelling correction and auto-suggestions. What is rarely appreciated is the number of calls the query API will make to the index to surface results. The number of characters an end user must enter to trigger these features can also impact the overall processing costs.
A familiar problem with these features, and indeed with levels of service, is that it is very difficult to extrapolate them from an existing on-prem solution and often the analytics are not good enough to translate activity into calls to the processor.
Related Article: Diagnosing Enterprise Search Failures
The key question here is if your existing search team has the skills and availability to implement the search application. The majority of search vendors offer little in the way of professional service support. You will either need to bring in a systems integrator or rely on your own team, with some lightweight support from the vendor. Remember you’ll still need to maintain the current application’s performance while waiting on the new application’s installation. The migration from one to the other could be either soft and phased or hard and total. The support around the organization also has to be defined and costed, especially if payment involves other business departments who assumed IT would foot the bill.
Situations where professional service support and maintenance support is provided partially or totally from a time zone more than three hours away from your team HQ call for extreme caution. Whose clock will count down the time to resolution?
Systems integration companies charge by the hour and day, so the concept of a fixed price integration is not on the table. Contracts will also include many caveats to cover unforeseen circumstances. In the case of bringing in a new application, the caveats will multiply. A particular challenge with search is it will take a minimum of everything being up and running for a few months before you can tick the “Successful Launch” box and receive the plaudits you think you deserve from your global workforce.
Leave ROI Out of the Business Case
You are not going to get a definitive indication of costs until you get to the pre-contract stage. So forget about making an ROI business case, as you will have no idea up front what the “I” will be.
Martin White is Managing Director of Intranet Focus, Ltd. and is based in Horsham, UK. An information scientist by profession, he has been involved in information retrieval and search for nearly four decades as a consultant, author and columnist.