Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.
The Case for Semantic-
Based Approaches to
Product Discovery
Guest Speakers:
Scott Compton, Senior Analyst, Forrester
Pete...
Guest Speakers
SCOTT COMPTON
SENIOR ANALYST
PETER CURRAN
GM, DIGITAL COMMERCE
January 2021
The Case for Semantic-Based
Approaches to Product Discovery
Scott Compton
Senior Analyst
© 2020 Forrester. Reproduction Prohibited.
While global retail sales decline, online sales will
continue to grow
Base: 435 to 765 US online adults; Source: Forrester’s August 2020 US COVID-19 Retail Consumer Survey
One-third of online ...
24% percent of US online adults who started using curbside
pickup during the pandemic expect to continue after the
pandemi...
Digital Commerce Search Best
Practices
Matching customer intent with product information.
8
© 2018 FORRESTER. REPRODUCTION PROHIBITED.
Visuals expedite the search and reduce incorrect selections.
Search and navig...
9
© 2018 FORRESTER. REPRODUCTION PROHIBITED.
Predictive search provides product images and a call-to-action – reducing ste...
© 2020 Forrester. Reproduction Prohibited.
Findability: Search filters, results
Search filters and results highlight price...
© 2020 Forrester. Reproduction Prohibited.
Findability: Filtering
Deeper filtering helps customers find products that meet...
© 2020 Forrester. Reproduction Prohibited.
Findability: Content search
+ Searchable
recipes
+ Searchable
ingredients
Conte...
© 2020 Forrester. Reproduction Prohibited.
Merchandising moments: Rich content
Deliver rich, meaningful content to help cu...
© 2020 Forrester. Reproduction Prohibited.
Merchandising moments: Comparison shopping content
Comparison functions as part...
© 2020 Forrester. Reproduction Prohibited.
Merchandising moments: Search “badging”
Use badging in navigation (including se...
© 2020 Forrester. Reproduction Prohibited.
Voice, Image and Scanning Search
New input expectations by the customer create ...
Behind the scenes of effective search
programs
Evolving the retailer’s search discipline to serve the best results.
© 2020 Forrester. Reproduction Prohibited.
What does ‘tuning’ search look like?
Start at global settings and work your way...
© 2020 Forrester. Reproduction Prohibited.
Cautions of using static rules
It is likely your tunings that are creating the ...
© 2020 Forrester. Reproduction Prohibited.
Using search insights throughout the organization
Progressive retailers are usi...
© 2020 Forrester. Reproduction Prohibited.
NLP And ML Enable Customer-Centric Search Architectures
For Future Conversation...
Thank You.
Scott Compton
Senior Analyst
2
Semantic Vector Search
A deep-learning dense-vector based solution to low-performing queries
Peter Curran
GM, Digital Comm...
24
COVID Date Night!
My real life search experience last Saturday night
25
Check out Palm Springs!
26
I love horror movies, and this is my fave!
The VVitch, 2015, Robert Eggers, A24
27
What to watch next?
28
You can’t always get what you want!
But sometimes …. (you know the rest)
29
Remember half marathons?!?!?
Suppose you had a great running top and you need a larger size a few years later.
30
What happens when we fix this?
Utilization of search grew
Search utilization surge from 14-16% baseline to 31-34%.
Sema...
31
Let’s visualize semantic vector space
32
A better visualization of vector space
33
Frequency
Head: skirt, laptop, toilet paper
Torso: black merino wool cardigan, 13” laptop on sale
Tail: warm carbon shi...
34
TRAINED ON POSITIVE RESULTS
What do people add to cart from search
results?
TRAINED ON ZERO RESULTS
What do people add ...
35
IN STOCK / OUT OF STOCK
Train on a combination of what people cart in both situations. Favor non-zero results encoder (...
36
Brands & collections not carried
37
Expanding to new business problems
Superseded parts
”Kit” results for projects
Improvement of the existing solution
Imp...
38
38
ORGANIC LEMONADE
A very precise result
Could we do more?
IMPROVE SEARCH RESULTS VALUE
39
39
ORGANIC LEMONADE
A less precise result
But a better experience
IMPROVE SEARCH RESULTS VALUE
40
40
ORGANIC LEMONADE
A more precise result
Higher potential order value
FUTURE: COMBINE LEXICAL &
VECTOR SEARCH
41
41
THE BALLAD OF
SONGBIRDS & SNAKES
A very precise query
An uninspiring result.
IMPROVE SEARCH RESULTS VALUE
42
42
THE BALLAD OF
SONGBIRDS & SNAKES
Less precise, but better inspiration
The user has a goal, but they may
not be as sa...
43
Fusion for Commerce & Digital
ML-based product & content discovery. Operable at massive scale for key use cases.
Search...
44
Questions?
GUEST SPEAKER:
SCOTT COMPTON
SENIOR ANALYST
PETER CURRAN
GM, DIGITAL COMMERCE
45
THANK YOU
Prochain SlideShare
Chargement dans…5
×
Prochain SlideShare
What to Upload to SlideShare
Suivant
Télécharger pour lire hors ligne et voir en mode plein écran

0

Partager

Télécharger pour lire hors ligne

The Case for Semantic-Based Approaches to Product Discovery

Télécharger pour lire hors ligne

Search is a key component of companies’ product discovery and customer service strategy. Much more than simply avoiding the dreaded “null results” problem, sophisticated search programs provide contextually relevant products and content in real time, based on deriving intent from the customer and using that information deep into the customer interaction to serve their needs explicitly.

Guest speaker and Forrester senior analyst, Scott Compton, is an expert in retail ecommerce and direct to consumer marketing. Scott advises companies on the strategies and technologies needed to serve the customer within today’s quickly – changing ecommerce environment. Lucidworks Digital Commerce General Manager, Peter Curran, will host a webinar on January 21 discussing:

-The many ways companies can apply digital commerce search to drive onsite conversions, grow revenue and increase customer loyalty
-How to extend the insights from your company’s search program to omnichannel customer service use cases
-From Lucidworks, how Vector Search helped some of the world’s largest ecommerce companies address consumer goals over the 2020 holiday shopping season
-How AI-powered product discovery enables retailers to understand and deliver on customer goals for higher customer satisfaction and engagement amplifying the impact on AOV and conversion

Featuring
-Scott Compton Senior Analyst Serving eBusiness & Channel Strategy Professionals, Forrester
-Peter Curran General Manager, Digital Commerce, Lucidworks

  • Soyez le premier à aimer ceci

The Case for Semantic-Based Approaches to Product Discovery

  1. 1. The Case for Semantic- Based Approaches to Product Discovery Guest Speakers: Scott Compton, Senior Analyst, Forrester Peter Curran, GM of Digital Commerce Lucidworks
  2. 2. Guest Speakers SCOTT COMPTON SENIOR ANALYST PETER CURRAN GM, DIGITAL COMMERCE
  3. 3. January 2021 The Case for Semantic-Based Approaches to Product Discovery Scott Compton Senior Analyst
  4. 4. © 2020 Forrester. Reproduction Prohibited. While global retail sales decline, online sales will continue to grow
  5. 5. Base: 435 to 765 US online adults; Source: Forrester’s August 2020 US COVID-19 Retail Consumer Survey One-third of online US adults note that items were out of stock and that shipping times were longer than expected.
  6. 6. 24% percent of US online adults who started using curbside pickup during the pandemic expect to continue after the pandemic is over. Base: 435-765 US Online Adults Source: Forrester’s Retail COVID-19 August 2020 Consumer Ad Hoc Survey
  7. 7. Digital Commerce Search Best Practices Matching customer intent with product information.
  8. 8. 8 © 2018 FORRESTER. REPRODUCTION PROHIBITED. Visuals expedite the search and reduce incorrect selections. Search and navigation: Visual product cues Source: https://www.costco.com + Costco’s predictive search results provide product information and images in the search result. This may help with product recognition, allowing the customer to bypass the search result page with confidence.
  9. 9. 9 © 2018 FORRESTER. REPRODUCTION PROHIBITED. Predictive search provides product images and a call-to-action – reducing steps, speeding checkout. Search and navigation: Shopping functionality Source: https://www.ralphs.com + Ralphs provides a quick-link to add to cart directly from the search menu where the customer is even able to adjust the quantity. For the product- focused consumer, this reduces steps and expedites check-out.
  10. 10. © 2020 Forrester. Reproduction Prohibited. Findability: Search filters, results Search filters and results highlight price, product ratings, fulfillment options. + Fulfilment options + Detailed product filters + Ratings and reviews + Strong product details including price per unit
  11. 11. © 2020 Forrester. Reproduction Prohibited. Findability: Filtering Deeper filtering helps customers find products that meet dietary and nutrition preferences, needs. + Dietary and lifestyle filters
  12. 12. © 2020 Forrester. Reproduction Prohibited. Findability: Content search + Searchable recipes + Searchable ingredients Content search helps customers find ideas and products. Don’t forget customer service content!
  13. 13. © 2020 Forrester. Reproduction Prohibited. Merchandising moments: Rich content Deliver rich, meaningful content to help customers more deeply explore products. + Amazon has extensive content on the product detail pages, including product benefits and video + Recipe/event content helps to expose multiple product uses
  14. 14. © 2020 Forrester. Reproduction Prohibited. Merchandising moments: Comparison shopping content Comparison functions as part of the eCom suite can be employed as well as comparison content in the product descriptions. + Amazon uses comparison content on the product detail page. Convenience-focused customers need comparison shopping content to find the right product.
  15. 15. © 2020 Forrester. Reproduction Prohibited. Merchandising moments: Search “badging” Use badging in navigation (including search filters), product results pages, and product detail pages. + Visual badges highlight the items in weekly circular + Content support for badging definitions + Lifestyle badging used in nav and search filters
  16. 16. © 2020 Forrester. Reproduction Prohibited. Voice, Image and Scanning Search New input expectations by the customer create new challenges for retailers + Voice, image, and scanning search
  17. 17. Behind the scenes of effective search programs Evolving the retailer’s search discipline to serve the best results.
  18. 18. © 2020 Forrester. Reproduction Prohibited. What does ‘tuning’ search look like? Start at global settings and work your way down to keyword / product settings Field Mapping & Weightings Synonyms Boosting Products Redirects
  19. 19. © 2020 Forrester. Reproduction Prohibited. Cautions of using static rules It is likely your tunings that are creating the customer experience hiccups Static rules in a dynamic environment have their limitations. Hard-coding products into the results doesn’t work very well with OOS situations. Static rules can sometimes overlap creating search traps that are difficult to unravel. Beware of multiple synonyms meanings – Example Khaki. Beware of two-way synonyms – Red / Ruby.
  20. 20. © 2020 Forrester. Reproduction Prohibited. Using search insights throughout the organization Progressive retailers are using search data in new ways Use keyword insights to drive SEO pages for converting products. Customer Service B2B Customer Retargeting SEO Historical searches can be attached to the account records. Audiences can be built from search activity. If you have discovered intent why not empower your CS reps?
  21. 21. © 2020 Forrester. Reproduction Prohibited. NLP And ML Enable Customer-Centric Search Architectures For Future Conversational Engagement Revamp site search to jumpstart AI, chat and personalization
  22. 22. Thank You. Scott Compton Senior Analyst 2
  23. 23. Semantic Vector Search A deep-learning dense-vector based solution to low-performing queries Peter Curran GM, Digital Commerce Lucidworks Inc. January 21, 2021
  24. 24. 24 COVID Date Night! My real life search experience last Saturday night
  25. 25. 25 Check out Palm Springs!
  26. 26. 26 I love horror movies, and this is my fave! The VVitch, 2015, Robert Eggers, A24
  27. 27. 27 What to watch next?
  28. 28. 28 You can’t always get what you want! But sometimes …. (you know the rest)
  29. 29. 29 Remember half marathons?!?!? Suppose you had a great running top and you need a larger size a few years later.
  30. 30. 30 What happens when we fix this? Utilization of search grew Search utilization surge from 14-16% baseline to 31-34%. Semantic vector search utilization dropped from 16% to 14%. Semantic results convert at 2x to 3x true null results. A few findings from a top-20 US retailer running Fusion & semantic vector search over Cyber 5 2020 Massive growth over cyber 5 last year Clickthrough rates up 20% Search-influenced orders up 30% Reduction of nulls by 91%
  31. 31. 31 Let’s visualize semantic vector space
  32. 32. 32 A better visualization of vector space
  33. 33. 33 Frequency Head: skirt, laptop, toilet paper Torso: black merino wool cardigan, 13” laptop on sale Tail: warm carbon shirt dress, white gold rose cut ruby ring Lexical Search Prevails Vector Search Prevails Lexical vs. semantic vector search Vector search isn’t a panacea. But it solves your hardest problem at scale. Percentage of Total Query Volume 20-40% 20-40% 20-60%
  34. 34. 34 TRAINED ON POSITIVE RESULTS What do people add to cart from search results? TRAINED ON ZERO RESULTS What do people add to cart after zero results? Two encoders, two opinions Two separate vector spaces give us different opinions on the same query at runtime. In this case, we trained purely on Adobe Analytics Clickstream data.
  35. 35. 35 IN STOCK / OUT OF STOCK Train on a combination of what people cart in both situations. Favor non-zero results encoder (opinion 2). PRODUCTS NOT CARRIED Rely on the persistence of loyalists. What do they cart after zero results? Favor zero results encoder (opinion 1). VOCABULARY Rely on the persistence of loyalists. What do they cart after zero results? Favor zero results encoder (opinion 1). MISSPELLING Semantic similarity. Favor non-zero results encoder (opinion 2). Four use case categories Solved with absolutely zero curation (no synonyms) and existing training data from Analytics platform.
  36. 36. 36 Brands & collections not carried
  37. 37. 37 Expanding to new business problems Superseded parts ”Kit” results for projects Improvement of the existing solution Improved embedding management Training “forward” That’s cool! What else can it do? Solution areas we’re working on now
  38. 38. 38 38 ORGANIC LEMONADE A very precise result Could we do more? IMPROVE SEARCH RESULTS VALUE
  39. 39. 39 39 ORGANIC LEMONADE A less precise result But a better experience IMPROVE SEARCH RESULTS VALUE
  40. 40. 40 40 ORGANIC LEMONADE A more precise result Higher potential order value FUTURE: COMBINE LEXICAL & VECTOR SEARCH
  41. 41. 41 41 THE BALLAD OF SONGBIRDS & SNAKES A very precise query An uninspiring result. IMPROVE SEARCH RESULTS VALUE
  42. 42. 42 42 THE BALLAD OF SONGBIRDS & SNAKES Less precise, but better inspiration The user has a goal, but they may not be as satisfied with a literal match as a semantically & thematically useful match. MERCHANDISING SEARCH RESULTS
  43. 43. 43 Fusion for Commerce & Digital ML-based product & content discovery. Operable at massive scale for key use cases. Search Type Ahead Guided Navigation Semantic LPQ/ZRP Browse Listing Landing Pages Classify & Enrich Finders & Configurators Personalization Recommendations Chatbots Self-Service Traffic SEO: Sitemap Expansion SEM: Auction Buys
  44. 44. 44 Questions? GUEST SPEAKER: SCOTT COMPTON SENIOR ANALYST PETER CURRAN GM, DIGITAL COMMERCE
  45. 45. 45 THANK YOU

Search is a key component of companies’ product discovery and customer service strategy. Much more than simply avoiding the dreaded “null results” problem, sophisticated search programs provide contextually relevant products and content in real time, based on deriving intent from the customer and using that information deep into the customer interaction to serve their needs explicitly. Guest speaker and Forrester senior analyst, Scott Compton, is an expert in retail ecommerce and direct to consumer marketing. Scott advises companies on the strategies and technologies needed to serve the customer within today’s quickly – changing ecommerce environment. Lucidworks Digital Commerce General Manager, Peter Curran, will host a webinar on January 21 discussing: -The many ways companies can apply digital commerce search to drive onsite conversions, grow revenue and increase customer loyalty -How to extend the insights from your company’s search program to omnichannel customer service use cases -From Lucidworks, how Vector Search helped some of the world’s largest ecommerce companies address consumer goals over the 2020 holiday shopping season -How AI-powered product discovery enables retailers to understand and deliver on customer goals for higher customer satisfaction and engagement amplifying the impact on AOV and conversion Featuring -Scott Compton Senior Analyst Serving eBusiness & Channel Strategy Professionals, Forrester -Peter Curran General Manager, Digital Commerce, Lucidworks

Vues

Nombre de vues

415

Sur Slideshare

0

À partir des intégrations

0

Nombre d'intégrations

1

Actions

Téléchargements

7

Partages

0

Commentaires

0

Mentions J'aime

0

×