An Advanced Search Algorithm For Collaborative Marketplaces

Which transaction workflows are best for service marketplaces?

Collaborative marketplaces, from big beasts like Airbnb to a local site for river boat rentals, has at its core a search and filtering engine to present the potential customer with the most relevant results based on his or her needs. But how do we ensure that this potential customer sees the results most likely to lead him or her to make a transaction, especially when the search, based on the strictly-defined filtering criteria the user has set, does not yield enough results?

To help answer this question, Roobykon Software have some insights from the team at Cocolabs, who created the awesome Cocorico engine for on-demand and service marketplaces – and made it available open source.

The majority of startups that decide to base their implementation upon the Cocorico engine naturally want to maximize the number of search filters available to their users, in order to help them obtain the most accurate possible results. But this is a mistake. What has been true in the economy of the web as we’ve known it so far with ecommerce doesn’t always work in the sharing (or ‘on-demand’) economy. The consumption habits of typical visitors to ecommerce sites cannot be applied to the sharing economy.

Is the problem a severe one? Not at all – we just have to do a little bit of math.

Let’s take as example a search based on geographical criteria. The user will naturally select the locations that he or she feels will be most convenient – but these locations may not be served by any suitable suppliers. Here, the Cocorico engine also looks for results that are outside of the specified areas but which are likely still be to accessible to the user.

The more the options there are for the user to add lots of search criteria, the more likely it becomes that no single available result will match all specified requirements.

Of course, in an infinite universe, it would be great to offer the user a result that corresponded to their exact desires – that, after all, would give the greatest likelihood of a transaction workflow. But in the finite universe that actually exists, giving too many options leads to a high chance of having nothing to offer at all.

There is a risk that automatic extension of the scope of the search could lead to results that aren’t relevant, and so compromise the reputation of the platform in the eyes of the suppliers using it. But as long as users are made aware that their results are being augmented with others that could be useful to them, the likelihood is that they will be pleased to receive additional, helpful suggestions.

Cocosearch – an advanced search algorithm by Cocolabs which seeks to provide the potential customer with the results most likely to lead to a transaction – therefore proposes to look at the problem for the opposite direction, rather than trying to live with an unsuitable approach. In order to maximize the probability of finding a satisfactory result for the user, the engine offers a range of choices of varying quality, and arranges them into a ‘virtuous circle’, with the most relevant results near the middle and less relevant ones towards the edge.

Isn't it great, this sharing economy!?

Ensure Customer Satisfaction & Boost Conversions With Cocosearch

A reliable provider, a satisfied customer

Ecommerce sites set up their search engines so as to offer the buyer the result that is closest to what was searched for: I want a red balloon, the platform offers me red balloons. This outcome is the best-case scenario, as it gives the greatest probability of a successful transaction. Once the buyer finds what he’s been searching for and proceeds to payment, it is taken for granted that the merchandise will be delivered – because ecommerce sites have total control over the purchase process.

Service marketplaces, on the other hand, have to deal with suppliers of widely varying reliability. They are faced with a good deal of flexibility as to time and place of service delivery – i.e., the service provider can only deliver the service at a certain time and in a certain place. On top of this, there is the uncertain reliability brought about by the diversity of represented suppliers – micro businesses, solo entrepreneurs, individuals – of whose ability and/or desire to honor demand one cannot be sure.

In order to adapt to the constraints and meet the special requirements of service marketplaces, Cocolabs have tested a variety of search algorithms on different platforms. Below Roobykon Software team made a retrospective overview of how the Cocosearch algorithm evolved over time to deliver the most relevant search results to customers of the service marketplaces that use the Cocorico engine.

1st generation of the algorithm

The first generation of the algorithm did not address the offer reliability problem – the problem of how likely the person offering a service is to actually be willing/able to provide it. Instead it applied the traditional methodology of ecommerce sites directly to marketplaces. Accordingly, results were ranked by distance from the searched-for location, and according to the dates declared as available (the goal being to provide the closest result to what had been searched for). A few hundred transactions were processed before a major problem was identified concerning the rate at which suppliers were refusing to honor service requests. 

2nd generation of the algorithm 

After discovering how many refusals there had been from providers to fulfill the services they’d advertized, the Cocorico team upgraded its search algorithm so as to promote results whose availability dates had been explicitly declared by the provider. Marketplaces based on the Cocorico engine generally consider the provider to be always available by default – but if the provider wants to improve its ranking in search results, it can explicitly declare its periods of availability and unavailability. This change delivered considerable improvements, but still left a lot of room for further advancements. 

3rd generation of the algorithm (Cocosearch)

We now clearly understand that the best result for the customer, the provider and the platform is whichever is most likely to lead to a successful transaction workflow. We also believe that the greatest obstacle to the conclusion of a successful transaction is a refusal or lack of response from the provider. With this in mind, the Cocolabs team reoriented its search algorithm to pay greater attention to the historical behaviour of providers.

The principles of the algorithm are based on the following elements:

Location

A search by location returns a series of results close to the location specified. This list is divided into two parts:

• Exact results - The exact results are those whose location (street, zip code, city, county, region or country) exactly correspond to the specified place.

• Close results - The second part of the list comprises all those results close to the specified place.

Availability

During a search by date, the two lists of results, divided by location (see above), are further divided, each again into two groups:

• Proven availability - Listings whose dates of proven availability match the dates of the search – these appear first.

• Undetermined availability - Below appear listings whose dates of availability are unknown.

Platform rating

To each of the listings within the four groups described above, which appear following a search by date and location, a rating is assigned based on the following rules:

• Listing completeness (weight X) - The degree of listing completeness takes into account the amount of information given by the provider about the listing: whether or not details are specified or a title has been entered, if the description contains more than 250 characters, whether a price is specified, whether the number of uploaded images is greater than the required minimum etc…

• Profile completeness (weight X) - The profile completeness takes into account the amount of information entered by the provider, including whether its description of itself contains more than 250 characters, and if there are more images uploaded than the required minimum.

• Profile rating (weight X) - Takes into account the ratings that the supplier has received. Results sorting is designed first of all to rank down poor providers, rather than to promote good ones. This approach is based on the idea that a good service is the expected standard, and so poor service has to be heavily penalized. The engine takes into account the average score of the provider, which is weighted by how many individual ratings it is made up of.

• Date of most recent calendar update (weight X) - The fact that a supplier’s calendar is regularly updated is correlated to a greater willingness to respond to requests and provide the service advertized – so, having a more recently-updated calendar will improve a user’s search rating.

• Number of completed services during the last 30 days (weight X) - A higher frequency of recent service completion suggest a willingness to respond to requests and to provide the services offered. Ratings are adjusted based on the number of bookings paid for and not cancelled within the past 30 days.

• Message response rate of the provider (weight X) - To calculate the rate of response, Cocosearch divides the number of messages sent by the provider by the total number he has received, improving the rating of suppliers with a higher response rate.

• Acceptance rate (weight X) - The acceptance rate of the provider is the percentage of claims which he has accepted, whether or not the booking was completed.

• Transactions success rate (weight X) - The number of successful transactions is calculated based on the number of bookings whose transfer has been authorized in relation to the total number of bookings.

• Response time (weight X) - Taking into account the response time of the provider allows Cocosearch to favor those who are the most responsive to customer requests. The response time rating is based on the time between the last message received and the response made to it for each of the supplier’s discussion threads. If no response has been made by the customer, this time is not taken into account. The rating is adjusted based on the average time of response.

• ‘Certified’ status (weight X) - This status can be assigned to listings by an administrator, for example after certain documents are provided by the supplier. Possession of ‘certified’ status helps the listings attain a better ranking in search results.

• Newcomer bonus (weight X) - To welcome new providers to the marketplace and to give them a chance, a newcomer bonus is assigned to all listings which were added less than 30 days ago.

• Random bonus (weight X) - In order to prevent listings’ immobility within search results, and to encourage providers which have been unable to obtain a favorable ranking, a random bonus is awarded to 5% of site listings. The receivers of this bonus are changed each day at random. The weight of the bonus also changes at random.

TO SUM UP

When managing a platform focused on the sharing economy, the major goal is to ensure customer satisfaction and maximize the conversion rate – and further, to make the enterprise profitable. The main difficulty of this market is in the fact that no collaborative platform is able to guarantee the reliability of every listing placed on it – because the providers are usually individuals, or professionals acting in a private capacity. The latter, who do their main business outside of the platform, do not show either the dedication or the involvement of a company whose ecommerce business is its sole or main activity. The major risk is that the customer faces repeated refusals, or even worse, an absence of responses.

The listings of unreliable providers must be identified and ranked down in search results. Such unreliability appears according to the provider’s situation – the availability of the things he is offering as part of his service, and his need to make money from them. For example, if I am planning to take a trip in four months’ time, I might offer my car for rental (and therefore also do without it) during the period between now and then, in order to save money for my travels. But there’s a strong chance that a potential customer contacting me in six months’ time about rental of this same car will get a refusal or just no response, either because I have not updated my listing, or I simply because I’ve stopped visiting the site. This experience happens again and again for a great number of customers of collaborative marketplaces; it should be avoided at all costs, no matter what the stage of the platform’s lifecycle.

For platforms to succeed it is critical that they don’t look at the sharing economy as we have, to date, been accustomed to looking at the rest of the web. The participatory economy breaks the rules. It revolutionizes distribution networks and reshapes consumer psychology. Now, the customer is no longer looking for a cheaper place to buy for the product which he noticed in the store but found too expensive; rather, he’s after something that ‘feels right’. Customer satisfaction is subjective, yet it will be the holy grail of the search engines used in future service marketplaces.

The essence of Cocoseach can be found in the reasoning behind the following conundrum: consumption habits have changed, customer expectations have changed; the way of providing services must change so as to favour the results that will lead to a successful transaction.

Building A Transaction Workflow - Cases For Service Marketplaces

Which transaction workflows are best for service marketplaces?

When it comes to service marketplaces, classical transaction workflows just don’t work. In e-commerce marketplaces, managing stock is relatively easy, since it’s controlled by a single management team. In contrast, service marketplaces require availability dates and times to be dynamically maintained by service providers often made up of many individuals. As a result, those tasked with maintaining such marketplaces face uncertain provider reliability – and this calls for a new approach to the transactional workflow.

The models first proposed by service marketplaces presented the following workflow:

1. Customer makes a request to the provider.

2. Provider accepts the request from the customer.

3. Customer receives an acceptance message.

4. Customer pays for the services.

5. Provider receives the payment.

6. Provider fulfills the request.

A workflow along these lines is used by the majority of service marketplace solutions.

But this workflow has many flaws, not least the lack of a trusted third party to help prevent fraud and create an atmosphere of confidence. It also has the inconvenient aspect of requiring the customer to be involved twice: first creating a request, then, only later once acceptance has been received, making payment (steps 1 and 4).

Roobykon Software has discovered the optimal solution earlier found by Cocolabs, who tested a variety of different transaction rearrangements that are in use on different platforms, serving different types of customer group.

The transaction workflow which delivered the best results in terms of conversion rate, acceptance by the provider and fraud prevention was the following:

1. Customer makes a request to the provider by entering his credit card details, with only pre-authorization performed at this stage.

2. Provider accepts the customer’s request and triggers completion of the payment transaction.

3. Platform receives the payment to escrow.

4. Provider fulfills the request.

5. Platform releases the payment to the provider – after the service completion date and provided the customer hasn’t filed a dispute.

The Launch of Cloud Application for Distance Learning

Breaking the silent note, Roobykon Software is proud to share the results of its dedication. Guys, we definitely haven't been sitting on our hands since the last month! Quite to the contrary, we’ve integrated the cloud learning application for our great Chinese partners - http://home.yincaiyun.cn/.

All our efforts were based on the open Canvas LMS made by Instructure. We achieved the back-end part through the Ruby, and Ruby on Rails framework. Also, used NodeJS as a runtime environment, Webpack as a module bundler, and Gulp.js as the wide toolkit. As for the front-end part - it’s clearly React and CoffeeScript. Eventually, we’ve gathered all these pieces together, and the hardly messed-up puzzle was finished way before the deadline.

This cloud system gives an opportunity for teachers & students to manage their studies at one place. It provides all the necessary digital tools they need and tangibly increases communication. Online courses, tests, instructions, grades - everything on one platform, on any device, at any time.

We truly hope our experience will serve an example for other creative teams!

Top 5 Digital Analytics Tools for Marketing

Like most of the consumers, we in Roobykon Software prefer to try first and then purchase, especially when we’re talking about solid analytic platforms that would strongly influence the business decisions. And it seems like a horrible dream to stuck with the prepaid tangled platform and keep on trying to push out everything it can provide. At the same time, with the numerous proposals, it’s so easy to get into this trap. That’s why day by day, bit by bit we’re collecting advices out of our friends and influencers to make a final decision.

The following review of the marketing analytic tools is not pretending to be a deep research, but more likely the last push for those who have doubts.

So, here it goes:

1. Google Analytics

Now, this is an awesome platform to start with, and for most of us, it’s truly a way out because it’s free. And what is more important it gives quite relevant information! So here are the common pros:

• you can check out where your customers are coming, what site’s sections they are surfing, how often they return, and lots of other useful information;

• the more you will be involved in your site's analytics, the more details you’ll get, and at the same time it wouldn’t be difficult at any stage;

• incredibly robust in terms of what it offers, absolutely enough for a small business, and moreover it’s completely sharpened by them.

Overall, it’s a must-have platform, and if you are not running this on the site, you obviously should - in cohesion with other tools.

2. Woopra

Woopra is another online platform that allows you track page views and provides data about your conversion funnels. Plus, it could help you to learn more about your customers’ actions to improve website conversions. Biggest pros:

• allows tracking data on your website, email, apps and more;

• allows monitoring more than one website at the same time;

• an ability to see visitors on your website in real time in addition to aggregate stats;

• provides a live chat feature, so you can actually talk with your customers while they are on your website.

An awesome real-time customer analytics platform, period. There are both free and paid options depending on the number of functions you plan to use, but in general, this solution is a little pricey.

3. Crazy Egg

Crazy Egg is an excellent additional tool that’ll help to figure out your visitors’ engagement and raise your conversions. In short, Crazy Egg:

• allows building ‘heat maps’ and, literally, tracking every single click of your visitor, so you can easily correct roughnesses in the website usability;

• allows you to see what specific site blocks are most clickable and investigate your user’s interests;

• helps to improve website design and boost conversions.

Moreover, there are another two solid arguments for Crazy Egg - absolutely easy to setup and has a 30-day money back guarantee on all accounts.

4. KissMetrics

KissMetrics offers a cool system that helps to find out more about your visitors, but it rather would be more practically useful for eCommerce than for blogs:

• the data funnel tool helps to find website’s strongest and weakest features;

• allows making necessary adjustments and increase your chances of getting conversions;

• tells you, in great detail, what visitors are doing on your website, before, during, and after they make purchases.

When you need to implement your CRO strategy and make sure that your sales funnels are working effectively, Kissmetrics would come in handy.

5. Clicky

If you are starting to take your online business a bit more seriously, then Clicky might be a better option.

• you get real-time analytics, including SpyView, which lets you observe what current visitors are doing on your site;

• simple to use and presents all the data you want to see clearly;

• one of the coolest heat maps out of numerous.

Also, a great plus is that Clicky offers a free service if you have only one website and a Pro account for a monthly fee.

INSTEAD OF CONCLUSION

All in all, before you’ll make a final decision, you should consider three critical futures - channels to process, complexity, the number of integrations.

Whether you’re doing SEO, social media marketing, content marketing, or a mix, it’s essential to have a chance to track everything from one single platform, so be aware of choosing the platform flexible enough to grow along with your appetites and able to accommodate your future needs. But at the same time, when your team is constantly struggling to use a particular platform, it can quickly become more trouble than it’s worth, and the benefits will be negated by the complexity.

And finally, be sure to look into integration capabilities. In future, it could cause serious financial losses and ridiculous downtime.