10 Steps to Outsource Data Labelling

10 Steps to Outsource Data Labelling

 

What is data labelling?

Let us illustrate with an example.

A Special Intelligence Unit (SIU) of the armed forces of a particular nation has been given responsibility for handling counter-insurgency operations. The unit keeps tabs on various outfits suspected of carrying out insurgencies, and tracks their communications and movements to get advance warning about their intentions.

It started with monitoring their messages on mobile networks. As both sides became more sophisticated, operations expanded to cover monitoring voice communication, including live calls, their geographical position and propaganda videos circulated by the outfits. While the unit has been effective in controlling insurgency, the enhanced coverage has resulted in an expansion of the force consuming more and more public money, putting the government in a tight spot financially.

A leading software company has assured the government that they will develop an Artificial Intelligence (AI) engine that will take over the task of monitoring, substantially reducing the need for manpower and cost. They have requested the SIU for historical data based on which they have been reaching conclusions regarding the insurgents and their activities.

After satisfying themselves regarding the security of the data, the force shared files with them containing:

  • Text messages intercepted

  • Phone calls intercepted and recorded

  • Geographical locations and change in them

  • Promotional videos

The software company looked at a sample of the content and returned it saying it was of no use to them as it was raw data. While humans, with their intelligence, could interpret that data and make sense out of it, a machine cannot.

Members of the SIU were called upon to identify elements in each piece of data based on which they constructed their theories based on which they could take action. For example, in the text messages, they identified the words, or phrases, that held clues. Similarly, in the phone call recordings, the elements they could identify that held clues. The same with the other sets of data.

Once these elements were identified and linkages established, the data was again handed over to the company for ingestion by the AI engine. This time it was data that the machine understood and could draw inferences from.

What was done by the special forces was ‘data labelling.’ As more and more data was acquired, they also trained the staff of the provider so that they could identify and label the relevant pieces of data themselves.

And the AI engine that could track insurgencies was born. Enabled and trained through ‘data labelling.’

oWorkers provides date labelling services to clients from around the world operating in a variety of industries and enables them to create smarter Artificial Intelligence engines. This could include placing electronic markings like bounding boxes on image files, putting marks on significant areas on faces, tagging pictures with keywords, and many others.

 

Why outsource data labelling?

Though we have come a long way, there are still occasions when an outsourcing decision, a logical, properly evaluated one that will add value to the business, needs to be justified, only because it is an ‘outsourcing’ decision.

Let us first answer the question, “why outsource?”

How is an outsourcing decision taken? It is not a given. It is a choice. Right? Through an evaluation process which analyses the different variables in the equation and tries to reach a decision likely to be the most beneficial, generally expressed in financial terms.

But these are standard business rules. How is this any different from any normal business decision?

Exactly.

Outsourcing is like any other engagement between two parties. These parties could be individuals, businesses, governments, or any competent body permitted to take decisions.

Many people send their children to school. Is that not a form of outsourcing? We are outsourcing our children’s education to an organization known as a school instead of doing it ourselves.

We buy food from stores. Is that not outsourcing our food supplies to others instead of producing ourselves?

An outsourcing contract will materialise only if there is interest and mutual benefit of the two parties to the contract. Exactly like any other contract between two businesses or human beings.

With this background on outsourcing, we can move forward to answer the question, “why outsource data labelling?”

Data labelling can be one of the more monotonous jobs anywhere. Looking at similar sets of data again and again, to identify the same elements. And it has to be done by humans.

Every organisation employs skill sets that are core to their work. A restaurant employs people to cook and to serve. An insurance company employs people like actuaries to evaluate risk on events that enable them to offer insurance policies. Besides, the employed people have perhaps been interested in that profession and undergone education and training to become suitable for employment.  

Data labelling services calls for a skillset quite distinct from the skillsets employed by the business that has this requirement; of staff who are experts in data labelling. Moreover, since it is a repetitive task for which training is mostly provided on-the-job, the services of these experts are reasonably priced. 

It is highly unlikely the business would like to deploy their cooks and actuaries to label the data that needs to be fed to their under-development AI engine. They will probably make a hash of it, while at the same time ignoring their core responsibility on which rests the success of their employer. The business is likely to prefer an outsourced service with cheaper labour and perhaps with tools and applications meant to facilitate the task while exercising process control?

This, then, in simple terms, is the case for outsourcing Data Labelling Services.

Of course, at the start of any exercise, experts in the particular field, cooks or actuaries, may be required to ‘show the way’ and train the resources who are going to be doing the major part of the exercise.

To summarise, the following are the main reasons why many organisations find it preferable to outsource data labelling:

  • Enables them to focus on core activities of the business

  • Allows specialised attention to be given to the task by specialists, reducing errors and enhancing efficiency of the process

And even more preferable to leverage data labelling solutions offered by oWorkers, across industry segments like Autonomous Vehicles, Medical AI, Satellite & Aerial imagery, Sports, Retail, Augmented Reality, Insurance, CCTV & Security, Robotics, Agriculture and several others. 

 

 

The 10 steps to success when you outsource data labelling

Selecting the right vendor is often the focus of effort for most outsourcers, but there is a lot more that needs to be done, both before and after the vendor is identified. Without adequate attention to each of these steps, the end result could be sub-optimal. As an outsourcer, you wouldn’t want that to happen.

Outlined here is a recommended process that will guide you from the point where you have started thinking that you would like to data labelling solutions, through to the end point where the vendor has started delivering.

 

1. Identify Requirement

Many business deals do not succeed because the client (buyer) is not clear on the requirements. What are they buying? Why are they buying? What problem will it solve? What value will it add? Identification of your need is a good starting point to outsource data labelling. Additionally, some clarity on success (or failure) criteria will be even better. A recommended checklist of items that this phase should cover:

  • Objective of outsourcing

  • Type/s of data

  • File formats

  • Annotations required

  • Defining rules

  • Domain knowledge requirement

  • Timelines

  • Volume

 

2. Advertise requirement/ Seek participation

Once it is clear that you would like to consider outsourcing as an option, you would need to make known to the prospective vendor community about it, with information from relevant items of the checklist created in the earlier phase. This can be done in many different ways:

  • Issuing a Request for Proposal (RFP) is the process preferred by large enterprises. An RFP is a standardised proposal form where you specify the information you seek.

  • Reaching out directly to vendors who may be providing similar services to competitors.

  • Advertising in trade circles, if you are a part of them.

  • Look up prospective vendors online or through directories like Yellow Pages and inform them one by one of your requirement. They will respond if interested.

 

3. Shortlist vendors

By now you would have hopefully received some interest from prospective vendors. If you have been flooded with responses, at this stage, based on information received, you should shortlist down to a few, perhaps two or three, with whom you can engage in a more detailed manner. A B2B engagement is a time-consuming affair. The larger the shortlist the more of your time it will require. If responses are inadequate or absent, you may need to review the terms and conditions you have set out. Perhaps they are too strict for vendors to be interested in.

 

4. Detailed discussion and evaluation

This is where the detailed discussions will happen and both parties will be required to share information. In most B2B cases, this will be preceded by the execution of a Non Disclosure Agreement (NDA) which binds both parties to treating the information received as confidential and enjoins them to ensure it is handled with the utmost care. This is the stage where the potential vendor will make a case for being selected, scope of services will be discussed, including indicative pricing. This is one of the key phases of the process.

A recommended list of capabilities and attributes to look for in a partner:

Prior experience

Facilitates a quick start instead of a slow ramp-up.

Over the last 8 years, oWorkers has successfully executed several data annotation projects for global clients. oWorkers provides proficiency in a wide variety of annotations, like Bounding boxes annotation, Keypoints annotation, Polygons annotation, 3D cuboids annotation, LIDAR segmentation, Text categorization, Linked entities recognition, Grammatical & discourse analysis, Review & sentiment analysis, Moving bounding boxes and Objects tracking (SOT,MOT).

Quality and accuracy

Consistently providing superior quality and accuracy is the best way to advertise your capability. Testimonials from existing clients are also helpful.

With QA (Quality Assurance) and QC (Quality Control) processes that represent the client and aim to detect and resolve errors, oWorkers consistently delivers over 98% accuracy. It leverages bast in class technologies in this effort.

Speed and turnaround

In a fast-paced world, speed is of the essence. The sooner the AI engine can ingest requisite data, the faster it can be brought to market.

In most cases, oWorkers will turn around today’s work even before you begin work the next morning.

Pricing

In any commercial engagement, this is a given. One party deliver goods or services, and the other pays a price for it in money terms. In a B2B, the basket of services and products provided is unique to the engagement, as is the price for it. While low is desirable, the outsourcer needs to ensure that the pricing terms offered will add value to their business instead of opting for the lowest number.

Our existing clients save upto 80% when they use our data labelling solutions. The same opportunity is available to all. Our pricing is transparent, with a choice of per hour or per output unit pricing.

Multi lingual

This is an important consideration for global businesses as well as for ones with aspirations, instead of having to take on a vendor for every new language.

Supporting over 22 languages across three global centers, oWorkers is uniquely positioned when you look for partners to outsource data labelling.

Technology and Data Security

Technology forms the backbone of every business. Along with data security. Data labelling services are themselves an input to Artificial Intelligence, a developing technology. Harnessing the right technologies is an important consideration for success.

ISO (27001 and 9001) certified, oWorkers leverages the best technology for data labelling solutions through its partnership with leading providers.

Scalability and access to human resources

Since BPO processing is a human activity, suitable resources, at the right price, in adequate numbers, should be available to the partner to enable them to carry out this task. Also enables business calling up and down as per requirement.

oWorkers possesses the flexibility to ramp up and ramp down, by a hundred headcount in 48 hours, particularly for Computer Vision projects.

Financial health and management support

Thought not relevant for operational delivery, these factors are important for ensuring consistency. Financial headwinds for the partner could lead to cutting corners in all projects.

oWorkers has been a consistently profitable enterprise. It pays local and social taxes for its employees and is deeply rooted in the communities it operates in.

 

5. Shortlist down to 1

Eventually you will decide on the most suitable partner and issue a Letter of Intent. This will be after detailed discussions, site visits, interacting with staff members, exploring combinations, negotiating on price and service levels. Contractual terms, which would have been already discussed during the previous stage, would now be formalised. The others will be retained a backups if this partnership falls through for some reason.

 

6. Finalise terms and sign contract

The formal agreement is executed based on mutually acceptable terms and conditions.

 

7. Trial Run

Initiate a Trial Run if agreed upon. The contract terms would shed light on the success criteria for this step and logical next steps based on various outcomes possible.

 

8. Implementation Project Plan

If this has not been done at the contracting stage, the parties will develop and agree on an Implementation Plan which defines the steps each of them must take to reach a steady state. In other words, reach a point at which the activities envisaged in the contract are running at the expected level. The timelines are also defined in the Implementation Plan.

 

9. Team identification, training and ramp support

The vendor will need to identify the team that will support activities under this contract while the outsourcer will arrange the initial training. Vendor support teams will start playing their regular roles for this project as well. Technical handshakes required will also be made.

 

10. Begin work, test and then go full steam

Work begins. If volumes are large, there is a ramp-up generally provided for in the Project Plan. Staring slowly, the work gradually ramps up to handle the agreed volumes.

Note: This is an indicative sequence of steps and not mandatory. Not all steps will be needed in all cases. In some cases, the sequence could also change, like a Trial Run could happen before a contract is signed.

 

Conclusion

oWorkers, as a pure-play data entry BPO company, has a unique position in the industry, and has been ranked in the Top 3 globally for data entry services. The solutions it offers are scalable, offering flexibility to ramp by 100 resources in 48 hours, a demonstration of their deep connect with the communities they work in. Its resources are employees, not contractors.

Our carefully selected and trained staff, are equipped to handle data labelling services for a variety of uses like Sentiment Analysis, Named Entity Recognition, Geo Labelling, etc. With support for 22+ languages, ISO 27001:2013 and 9001:2015 certifications, 24×7 operations and centers in Europe and Africa, oWorkers should be your first choice to outsource data labelling.