#mices: Solving the E-Commerce Search Challenge
This is a modal window.
Das Video konnte nicht geladen werden, da entweder ein Server- oder Netzwerkfehler auftrat oder das Format nicht unterstützt wird.
Formale Metadaten
| Titel |
| |
| Alternativer Titel |
| |
| Serientitel | ||
| Anzahl der Teile | 48 | |
| Autor | ||
| Mitwirkende | ||
| Lizenz | CC-Namensnennung 3.0 Unported: Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen. | |
| Identifikatoren | 10.5446/68813 (DOI) | |
| Herausgeber | ||
| Erscheinungsjahr | ||
| Sprache |
Inhaltliche Metadaten
| Fachgebiet | ||
| Genre | ||
| Abstract |
|
Berlin Buzzwords 202016 / 48
14
19
26
43
47
00:00
ModelltheorieEDV-BeratungBenutzerbeteiligungComputeranimationXMLUMLBesprechung/Interview
00:35
ModelltheorieSichtenkonzeptEDV-BeratungSuchmaschineBitNetiquetteBenutzerbeteiligungSoundverarbeitungProjektive EbeneMessage-PassingKette <Mathematik>Coxeter-GruppeSelbst organisierendes SystemSichtenkonzeptBesprechung/InterviewComputeranimation
01:38
Dienst <Informatik>ExpertensystemOffice-Paketsinc-FunktionDreiecksfreier Graph
02:10
PolstelleFlächeninhaltDienst <Informatik>Domain <Netzwerk>InformationDatenanalyseDigitalisierungKette <Mathematik>EDV-BeratungComputeranimation
02:56
GEDCOMDatenverwaltungFokalpunktSelbst organisierendes SystemObjektverfolgungProzess <Informatik>ImplementierungKonstanteBewertungstheoriePhysikalisches SystemModelltheorieWeb SiteKraftEDV-BeratungProjektive EbenePhysikalisches SystemDifferenteFlächeninhaltBitDatenfeldPunktWeb logDienst <Informatik>Divergente ReiheRechenzentrumExpertensystemDatenverwaltung
04:04
ExplosionszeichnungPhysikalisches SystemModelltheorieSelbst organisierendes SystemGlobale OptimierungProzess <Informatik>Projektive EbeneEDV-BeratungWeg <Topologie>PlastikkarteAnalytische MengeBeweistheorieZweiXML
04:43
Gebäude <Mathematik>Globale OptimierungKonfigurationsraumBeschreibungskomplexitätProzess <Informatik>Prozess <Informatik>RückkopplungDienst <Informatik>VerkehrsinformationAbstraktionsebeneGraphische BenutzeroberflächeGewicht <Ausgleichsrechnung>Konfiguration <Informatik>WebcamWort <Informatik>KonfigurationsraumVollständiger VerbandSoundverarbeitungProdukt <Mathematik>EinfügungsdämpfungRadikal <Mathematik>Globale OptimierungBit
06:02
DatenmissbrauchAnalytische MengePhysikalisches SystemCoxeter-GruppeProdukt <Mathematik>DifferenteKundendatenbank
06:38
Physikalisches SystemSelbst organisierendes SystemGlobale OptimierungProzess <Informatik>ExplosionszeichnungModelltheorieSichtenkonzeptDigitalsignalAbfrageVolumenPerkolationstheorieTermDruckverlaufSteuerwerkGruppenkeimStandardabweichungDienst <Informatik>IndexberechnungAnalysisFrequenzTypentheorieStichprobenumfangRankingSoftwaretestPhasenumwandlungStrom <Mathematik>DatenanalyseMusterspracheZeitzoneLOLA <Programm>BitProdukt <Mathematik>Selbst organisierendes SystemVerkehrsinformationApp <Programm>GruppenoperationWeb-SeiteFlächeninhaltTermMailing-ListeDatenmodellAbfrageVisualisierungRankingPhysikalisches SystemDatenbankZahlenbereichKategorie <Mathematik>SchnittmengeTaskProzess <Informatik>Token-RingDifferenteGlobale OptimierungMusterspracheSelbstrepräsentationResultanteKonstruktor <Informatik>Software Development KitDatensichtgerätDigitaltechnikFächer <Mathematik>Dichte <Stochastik>EinsMAPSoftwareOffice-PaketDienst <Informatik>DigitalisierungQuaderObjekt <Kategorie>Einfach zusammenhängender RaumDifferenzkernWeb SiteLogistische VerteilungWald <Graphentheorie>Leistung <Physik>p-BlockGibbs-VerteilungVideokonferenzEINKAUF <Programm>Chirurgie <Mathematik>Extreme programmingAnalytische MengeSoundverarbeitungMultiplikationsoperatorBenutzerschnittstellenverwaltungssystemBitrateEgo-ShooterNewsletterBeweistheorieDatenverwaltungXMLComputeranimation
12:28
ExpertensystemSelbst organisierendes SystemStetige FunktionMehragentensystemSpeicherabzugProdukt <Mathematik>RückkopplungSmith-DiagrammAbelsche KategorieStrategisches SpielSpezialrechnerBitrateLesen <Datenverarbeitung>SichtenkonzeptSystemprogrammierungNotepad-ComputerPhysikalisches SystemProzess <Informatik>GruppoidPlot <Graphische Darstellung>Atomarität <Informatik>Arithmetisches MittelCloud ComputingSkriptspracheTaskPhysikalisches SystemBitPerspektiveSelbst organisierendes SystemTaskMAPInformationsmanagementProzess <Informatik>ExpertensystemStrömungsrichtungKartesische KoordinatenKomponente <Software>TermExogene VariableInformationsqualitätMultiplikationsoperatorResultanteDiagrammSichtenkonzeptGlobale OptimierungUmsetzung <Informatik>SpeicherabzugNichtlinearer OperatorInhalt <Mathematik>Generator <Informatik>CASE <Informatik>EDV-BeratungModelltheorieKategorie <Mathematik>Analytische FortsetzungRückkopplungBitrateZahlenbereichKonfigurationsraumEinsProdukt <Mathematik>GrenzschichtablösungAbfrageWirkungsgradRechter WinkelFokalpunktMinkowski-MetrikKonfiguration <Informatik>ProgrammfehlerOvalDifferenteOpen SourceMobiles InternetGamecontrollerServerWärmeleitfähigkeitEinfacher Ringt-TestATMOperations ResearchGruppenoperationGoogolWorkstation <Musikinstrument>Message-PassingVerknüpfungsgliedBildschirmfensterWidgetStellenringVerkehrsinformationInternetworkingGleitendes MittelGesetz <Physik>Formation <Mathematik>ComputerspielComputeranimation
18:18
Umsetzung <Informatik>BitrateGlobale OptimierungTermVolumenTaskMatrizenrechnungExpertensystemExplosionszeichnungModelltheorieSelbst organisierendes SystemProzess <Informatik>Physikalisches SystemElektronische PublikationKonfigurationsraumSystemprogrammierungElastische DeformationHypermediaInhalt <Mathematik>Attributierte GrammatikDigitalfilterIndexberechnungGanze FunktionTransaktionProdukt <Mathematik>DatensichtgerätMatchingRankingPhysikalisches SystemUmsetzung <Informatik>TermRankingInformationTwitter <Softwareplattform>KonfigurationsraumAttributierte GrammatikEinfach zusammenhängender RaumCASE <Informatik>Inhalt <Mathematik>SchlussregelAnalytische MengeProdukt <Mathematik>Prozess <Informatik>InformationsqualitätHypermediaPunktDifferenteBitrateSichtenkonzeptResultanteOrdnung <Mathematik>ZahlenbereichTabelleVerkehrsinformationDatenfeldTransaktionElektronische PublikationMAPRechter WinkelSpezifisches VolumenEinsAutomatische IndexierungBitVektorpotenzialNichtlinearer OperatorEinflussgrößeBenutzerbeteiligungFilter <Stochastik>SchnittmengeNetzbetriebssystemInformationsspeicherungEnergiedichteWasserdampftafelTransformation <Mathematik>DatenanalyseWidgetPRINCE2Weg <Topologie>VideokonferenzGrundsätze ordnungsmäßiger DatenverarbeitungDatenverwaltungQuaderNegative ZahlGewicht <Ausgleichsrechnung>AuswahlverfahrenDienst <Informatik>Turbo-CodeRechenbuchDatenaustauschSoftwaretestDomain <Netzwerk>TopostheorieWort <Informatik>DiagrammComputeranimationFlussdiagramm
25:15
Konfiguration <Informatik>ImplementierungUmsetzung <Informatik>DifferenteSichtenkonzeptProdukt <Mathematik>Online-KatalogRückkopplungAbfrageRechter WinkelAnalytische MengeBildschirmmaskeProzess <Informatik>VerschlingungCASE <Informatik>Physikalisches SystemGlobale OptimierungAnalysisTaskKategorie <Mathematik>Domain <Netzwerk>Attributierte GrammatikAutomatische IndexierungUmwandlungsenthalpieTermKonfigurationsraumSchlussregelAbstimmung <Frequenz>Open SourceKeller <Informatik>BitInhalt <Mathematik>Web logOffice-PaketDivergente ReiheAusdruck <Logik>WellenpaketDatensichtgerätDateiformatVideokonferenzDienst <Informatik>DatenverwaltungInternetworkingFormation <Mathematik>BootenOrtszeitMultiplikationsoperatorSystemsoftwareBenutzerschnittstellenverwaltungssystemTransportproblemElement <Gruppentheorie>WhiteboardSystemidentifikationBesprechung/Interview
33:51
Vollständiger VerbandTorusKeller <Informatik>QuellcodeComputeranimationBesprechung/InterviewXMLUML
Transkript: Englisch(automatisch erzeugt)
00:09
Hi there. Welcome to continue the day. I'm happy to welcome Fabien Engel. Fabien is a Senior Search Consultant at For You and Your Customers,
00:22
and he will tell us about solving the e-commerce search challenge, an effective and lasting approach based on the For You and Your Customers exploded view model. Yes, thank you for the introduction.
00:41
Welcome to my talk. So as you already mentioned, I want to highlight that e-commerce search can be a challenge and I will present these effective and lasting methodologies based on a bit more like a holistic methodology on how you can look at a digital organization.
01:05
So let's dive into it. This is the agenda. I'll quickly give an introduction, present some typical search challenges, which I saw in my experience as a search consultant,
01:23
then present the exploded view model, which we use quite often in our consulting projects, and then I will go through each of the different layers of the model and explain these methods which I just mentioned. So quickly about the company For You and Your Customers.
01:45
We have around 160 experts all over the world. We are in Europe, have different offices here in different countries. And also, since around two years, we are in Australia, in Melbourne, in Sydney,
02:04
where we opened our offices and offer our consulting services there as well. Now, our services are ranging in different areas. So we do customer experience consulting, one of our biggest domains is actually e-commerce,
02:21
where we do search merchandising, but also digital information supply chain, BIM, MUM, MDM, and data analytics. Some of our customers, so as you can see already, we are working with some really well-known brands together in different industries.
02:44
So both B2C and B2B, for example, just to mention some, I guess you have heard of some of those already. Now, quickly about myself. I started at For You and Your Customers in 2016 as a search merchandising consultant.
03:06
So I did many projects in this area, also worked with different search solutions, just to mention some like Elasticsearch, Solr, but also a bit more business-oriented solutions like Attract Fertilper and Bloomridge.
03:22
Last year, I went to Australia for half a year to also conduct a search project in this area. And since last year in October, I'm in Berlin, where I'm doing consulting projects in search merchandising area, and also developed this field here for For You and Your Customers.
03:44
So the projects, working with data scientists, BIM experts, UX designers, system architects, data analysts, they are to deliver these search projects together and to create consistent concepts in the e-commerce search area.
04:05
So what are typical search challenges? I just mentioned three here, which I saw in some consulting projects from my experience. So for example, customers have a small amount of search analytics data,
04:24
or they use a very, very small amount of those which they are tracking actually and collecting. So they actually do not know which data to look at, which data to track, or actually how to analyze the data and use it to improve search.
04:43
Second issue, which I often saw is, for example, search technilization, which I call here a bit abstract words, basically means that search is very often regarded as a technical topic. So the ownership is in IT departments,
05:04
and that leads to complicating processes when it comes to the feedback from the business. Basically, when business sees our search is not working well, we need to improve something here. And these feedback loops, how to then improve the search,
05:22
how to configure search properly, this is done by creating tickets very often and leads to very long processes which are not very efficient. So business stakeholders very often do not have like a graphical user interface where they can configure certain business rules,
05:41
and they have little involvement in the optimization processes. And also often, it occurs that companies are just building their own search, they are underestimating the complexity, and they're forgetting about many business requirements or customer requirements.
06:02
And last but not least, kind of like data confusion, which I call it here. So not all of the available data which is actually available in different systems, like analytics data, for example, or customer data in CRM systems, they're not being used in the search system to improve the user experience,
06:26
just as an example for personalization purposes, or generally to make product rankings, for example, more relevant. So how to overcome these challenges? Now, we have this model, we call it the exploded view,
06:44
which we also use in the search area. This is a quick introduction. So if you look at these six different layers here, we have, starting from the top, we have the customer layer, which basically reflects the customers,
07:01
which, what can we actually understand or know about the customers? How can we use insights about customers? This is covered in this layer. The general layer covers all channels, which are actually connected to the e-commerce search. Typically, of course, the online shop, the website, but also other channels like, for example, an app can be analyzed on this layer.
07:28
Then we have the organization layer, where you have all of the roles and people, which are conducting certain tasks to improve experience in the channels for the customers.
07:41
We have the process layer, where efficiency is being generated, and the system layer, where you have the system landscape, the search system, of course, and all other systems which are connected to search, and the data layer, which includes all of the databases and the data models and the data you're actually using in your search system.
08:05
Based on that model, we now can look at our search and improve the experience and overcome the search challenge. Let's start with the customer layer.
08:22
What can be done here? We can actually use analytics data. For example, we just look at one year, the last year, data of search queries and the number of searches per query. We can generate certain reports out of the data, our talk.
08:44
For example, depicting the categories, search categories. What are actually people searching for? Are they searching for brands, categories, product numbers, and so on? We can then generate a report, which actually gives us insight about how our customers are searching and make this more transparent.
09:04
We can also analyze the average token length, the number of tokens are being searched for. One very important other thing is here, this ABC analysis, which is a very handy method.
09:21
You basically divide the search terms into three different groups. First of all, the A queries, which consists of normally a very low number of searches and a very high number of searches per search term. These are basically the most important searches for customers.
09:43
These should be corrected individually, of course. Then we have the B group where we have already a bigger number of terms and less searches per term. Then we have the long tail, the C group, with many, many different kinds of searches, but very low number of searches per search term.
10:04
Of course, these three different groups, what I observed actually is that many companies are just looking at the A group. They're just looking at the top terms. The long tail is very often neglected, at least in a bit more individual optimization processes.
10:22
What can we do in these three different groups to actually look at the data more holistically? A group would actually, as I just already mentioned, individually look at the terms. We would create curated product lists with certain ranking rules, which can improve the search experience.
10:43
We do maybe visual merchandising, create synonyms and redirects also for content sites, and maybe add banners to the search results pages to make it a bit more appealing. Then we have method two for the B groups. These actually can be optimized through reports.
11:06
Here we go less on the individual terms and look a bit at it more holistically, but it also involves synonyms and modifications, redirects and ranking rules, for example. Then method three is the long tail optimization. Here we would not look at the terms individually,
11:23
but we actually would create a set of representative search terms. That is what it could look like. Then from these representative search terms for each of the categories of searches,
11:40
we can actually test the search, how search is performing for these different search patterns or search categories. Then we can, based on that, for example, if we see how this is not performing well, we can deduct general optimizations for these pattern of searches.
12:00
This actually then will lead to the fact that we can optimize the whole long tail and not only the individual queries. This would be the customer layer. Channel layer, very often, of course, the search is only used on the online shop,
12:22
but also, of course, other channels like an app or a new set of systems, for example, where a search system can be queried, should be analyzed. There can be done some expert reviews, for example, for user experience improvements. Then typically, it's good to collect those in the backlog
12:42
and prioritize those by effort and severity. Then, of course, work on those optimizations in the different channels, which are a bit more UX components to improve search. Now, the next level is the organization layer.
13:04
Looking at the organization layer, we actually should look at certain search roles. Which current search roles are actually in the organization? Are there people looking at search currently or is it maybe distributed in different teams?
13:21
Maybe there's no responsible person for search in an organization. Then what are the responsibilities of those roles which are involved in search? Which are the people? How much time do they actually use here to improve the search?
13:40
What are the requirements for the organization per role? To also recruit certain people which have the skills to optimize search continuously. There can be a search term. Of course, the search goes into many other teams also. For example, here I have a typical setup
14:03
how a search organization looks. It's a bit small. I hope you can see it. We have, for example, as core roles, there could be a search application consultant which understands the search system in depth who can also fix bugs, for example.
14:23
And who is a bit more the owner or the consultant for the search application. And then we have the capability owner who is a bit more the person from the business perspective who is doing the optimizations from a strategic level.
14:41
And then the further roles, as I just mentioned, they're going into the different teams. So into the content team, for example, we have content specialists, product data specialists, we have category specialists, and maybe also external consulting firms. And all of these roles work together to generate good search
15:02
and ensure this efficiency. So now when we talk about search optimization processes, so what are these people doing, which I just mentioned? First off, what can be done is a generation of a use case diagram
15:24
to create a view, a current view of the current search organization and which people are actually involved in using search. So here it comes to getting a view on what is the overall view,
15:41
what use cases are covered in which system. So what use cases are people actually, search use cases are conducting people in which systems, which people are conducting these use cases, and then what dependencies and relationships are here. So when we have that view, the status quo view, basically,
16:04
this can help to then, as the next step, go and create a search operation model for search. So we work with the terms of task processes and support items. Some other companies, this operation model can have other names probably.
16:23
Here we define it like this. And what are tasks? So I just go into this in more detail now. Tasks would now be here, for example, 16 tasks, which we identified as best practice tasks to create a consistent and efficient search operation model.
16:44
These could be, for example, the optimization of zero result queries, one result queries, internal search searches, for example. But also, yeah, for example, the optimization of queries is a high number of requirements.
17:00
And then we have the processes, which are actually per task depicting and exactly describing which steps are needed to conduct those tasks. So the search roles, which I mentioned before, should then sit down and frequently conduct those tasks, attach with those processes to actually optimize search.
17:24
And the support items can be certain dashboards here. We created those for one customer in Google Data Studio. Of course, it can be other dashboarding tools as well. And these ones can then improve the efficiency of conducting those tasks there.
17:41
So for example, if we have here on the top right, you can see a high frequency, low conversion term dashboard. So basically someone would go into that report and from top to bottom, analyze our search terms with a low conversion rate but a high number of searches and optimize those ones
18:02
with certain configuration rules, for example, or feedbacks for the business, for the product information management system to improve search data quality and so on and so on. So this would be one example. We would have that report with the search terms,
18:24
the low conversion but high number of search volumes. Of course, these ones, if we look at these four quadrants, these ones would be in the top left corner. And of course, it would be the best to get these into the upper right corner here
18:42
to make these highly potential search terms to improve the conversion rate here and by that, improve the online sales. And here we have another tool, it's the RayC matrix, which then in the end can be used to map the roles
19:02
which have been defined and the processes together and define who is responsible, who is accountable, who should be consulted and who should be informed when conducting those tasks. And this basically makes up the whole operating order for search, which is a bit of manual process
19:24
but ensures the continuous efficiency with the knowledge of the business people involved. So the system landscape level or layer. This one now looks at the system landscape.
19:42
So basically which systems are currently being used, which systems are being connected to search currently, but also which configuration files are there per system. So which syndromes are there, which redirects, for example.
20:01
So configurations and also analyzers in the search system. And then generally looking at also how often data is being exchanged between the systems to analyze this and get a better view on that. And then later, of course, it is another step to define,
20:22
okay, which systems do we actually want to connect in future to our search merchandising solution with the idea in mind to, of course, improve the user experience for the customer again. Here you can see some typical search merchandising solutions. We are also working with those.
20:40
And you can, very often companies do not know what systems they can actually connect. So even would be possible to connect certain likes from Instagram to the search system or from Twitter or for even reviews from Basavo to then calculate a score,
21:01
for example, for certain ranking rules. In the search system. So this is even possible. And of course, analytics data, we have the PIM data, the product information, media assets, also contents can be, of course, indexed in search. And we have the e-commerce systems where sometimes also product information
21:24
and other collected information is coming from. And now when talking about the last layer, we want to answer these questions. So we look at actually which data
21:42
we want to connect to the search system. So for example, product data can be transaction data, analytics data, customer data, all of these different data points can be connected to search, to the search system. And then we would look at, okay, which attributes can be identified,
22:00
which maybe still need to be transformed in order to use it in search. And also where do we want to use that data in search, of course, in the search index and features and so on. And along with that, of course, the data quality is very important. This is what I just mentioned regarding the transformations,
22:21
which may be still needed to ensure the quality in the customer facing channel. So looking again through the layers, basically to the customer, what does the customer see in the channel? Okay, so we're looking at from the lowest layer,
22:42
basically the data layer. What does the customer actually, what data does the customer need? So again, which data should be made searchable, should be in search. How does the data needs to be cleaned? For example, that's what I just said already,
23:03
which data should be used in filters, but also in rankings, for example, and what data do we actually want to show to the customer in search in the results list. And since we very often call this field also, not only search, but search and merchandising, this goes hand in hand very often,
23:23
both in product lists, but also in these additional campaigns and recommended products, basically, which you can see on the bottom, which attributes should be used to display campaigns, to filter down those products, to rank those products as well.
23:44
And this is basically one method to get an overview of that, which we call the data definition sheet. So this is just basically a table where we can get a thorough overview
24:00
and high-level overview of the attributes we want to use in search. So again, all what I just said, just to collect this data, the attributes, certain example values, not to see what examples can we actually see here.
24:21
If we can use or want to use that data for the customer, and then where it should be used in the system as well, which can then be defined here. This is basically already the talk. Thank you for listening. If you want, if you're more interested in this topic,
24:42
it would also be possible to plan like a one-to-one session where I can give some more insights how we do that with some customers and tell more about some certain cases from customers where we did that.
25:01
Other than that, I would be happy to discuss this further in breakout rooms, room session after this. And thank you very much for listening. Cool. Thank you, Fabien. We have one question that had a lot of votes
25:22
in the Slack channel. It was about this high-volume low-convention conversion queries you talked about. The question was, can they always be transformed in high-conversion ones? Are there cases in which such queries are just exploratory and you can't do anything?
25:44
Very often, the person analyzing these queries comes from business, of course, and normally has knowledge about these search terms and should be able to, with those processes which I also just showed,
26:01
should be able to find certain rules to optimize search. There can be many things. I mean, it can be that you create synonyms, redirects, but sometimes it's not that easy. Sometimes you cannot just create a synonym and then it's done because sometimes, of course, you may not have the products
26:22
just in the catalog. So people, for example, cannot find anything that is a low-conversion query. Or the content is just some products. They're just not very popular.
26:41
People just don't want to buy them or they want to search for something, but they don't expect those products which are not so interesting, which may be correct results, but don't lead to a conversion. So that can, of course, happen. And in this case, it can also be that the outcome
27:01
of such an optimization is just the feedback to the business, to the product management, to procure other products, to add more products for that search query to the catalog, and that might also help to improve search.
27:21
So there is normally always a way to improve, but you, of course, need to know the steps or the options you have to do that. And very often it already stops just creating synonyms or redirects or something, but it can also be that these business processes and feedback for the business
27:43
needs to be involved. Okay, cool. Yeah, so this was a question by Matteo, and then James answered also that the aim is not, well, there is a discussion starting there. I hope you can follow up then in the breakout room.
28:04
But the aim is not that all queries are getting into the top right quadrants, but applying this view helps to identify, this is what you say, where to concentrate on and identify potential to improve. And of course, it is just one option. I guess experiences will be shared in the breakout room.
28:23
I see someone else typing. Meanwhile, I had also one question for you personally, which was, well, here you show how do you design some process and identify what has to be done.
28:40
How do you follow up afterwards once this is done? How do you go until the implementation of technical solutions for that? Or where do you stop? So yes, there are different kinds of products we also do.
29:01
So we have those initial reviews where we do just a review in these different layers and then give insights on what can be improved. There are these optimization potentials, for example. So these are the outcomes of this analysis would be then certain backlogs with optimization ideas.
29:23
And then after that, it is of course also a task to implement those ideas. And that is also something where we support our clients. So we do trainings, for example, on processes, also on how to use these systems. This is of course when the system is already there
29:42
and should also be used further. But when it comes to new implementations and we can already consider all of these things right from the beginning. So we can already do an analysis. We can then consider all of these processes already the new search solution after we implemented it.
30:04
So then in this process, basically when a new solution is implemented, it's going live and we already have the processes prepared. And this is basically the idea set up. OK, great. Another question is coming from JP.
30:21
What is the best way to handle search relevancy where the data in catalog is different? Sorry, it's from different domains. For example, someone like amazon.com where people can search for books and clothes. And then, sorry, yeah, that's it from JP.
30:43
Can you say the question again? What is the best way to handle search relevancy where you have different domains in your catalog? For example, books and clothes, like in Amazon. Yeah, of course, you would have different attributes for different product categories.
31:04
You will find out based on what people search. So basically when we do this search analysis, which I showed initially in the customer layer, you will find out what people are looking for. And then based on that, you can actually identify
31:20
all of these different categories. And you can see where search maybe currently is not performing well. And it might be that maybe you don't have a specific attribute in place for a specific product category. So then, of course, it would be good to add that attribute to the search index to make it searchable.
31:42
But this would already be when you have a search system which you have implemented where you already have a basic setup and you just identify those improvements. So that's what you can do with the analytics data. Of course, if you want to implement a new search system, it's of course good to set up a basic configuration
32:02
for your search index based on your attributes, on all of your different products you have. So this is a bit like the work to be done before that and the implementation, what products do we have, what basic attributes do we need, and later by looking at the analytics data
32:21
to then identify issues with that. And then later, even with different categories, it should be no problem there. Okay, I think we're done with the questions. Do not hesitate if you have any more.
32:41
We have some minutes left for Fabien to respond you. I also would like to remind you that the next session who is called Towards an Open Source Tool Stack for e-commerce search will not be umbrella. It will be with the link that was provided to you by Renée.
33:05
So you will, if you are willing to join, you have to turn off your cameras when joining in and enter your name in the form. And I think, meanwhile, I had another question. No, that's it.
33:22
That's it. Well, no, there is a lot of conversations coming. I think it would be great for you, Fabien, to join the conversation. So let's meet on Jitsi. I think you all have the link.
33:40
I will put it back in the conversation so that we can end the discussion there and continue it. Cool. Thank you, Fabien. Have a nice... No, no, that was...