Leonid Glazychev, Logrus IT CEO

Part I

Part II

It is a rule of thumb to build the rate structure around the actual cost structure. This is the only way to guarantee that both buyer and supplier approach project costs in a fair manner, and that quotes and budgets are transparent and reflect actual volumes of work to be completed.

Separate rates for each type of work and productivity

Generally, there should be a separate rate for each distinct combination of measurement unit, type of work and productivity.

• As discussed earlier, alternative types of work with varying productivities should not be combined into packaged rates. For example, one should not combine raw MT, (MT + editing) and human translation (with or without an existing TM) into a single packaged rate per a «universal translated word or page». These tasks have different productivities and can all apply to one and the same project simultaneously in varying proportions. For example, a project can include recycling of existing TM units combined with human editing of fuzzy matches, and MT applied for all new units and followed by human editing.
• Consecutive, «non-alternative» types of work, like translation and editing and/or proofing can be easily combined in a single rate for simplicity reasons, as far as they are based on one and the same unit, such as a word, piece or string.
• Even consecutive types of work should not be combined if they are measured in different units. For example, combining translation and page setting and/or art creation/localization in a single rate is generally a bad idea, because the number of words in the text is not directly related to page density, design complexity, and/or number and complexity of art pieces.

Project setup/management charges (PM fees), minimal fees and additional work

Do not forget to discuss project setup/management charges (PM fees), minimal fees, and additional charges explicitly.

The subject of PM fees and minimal fees is among the most contentious ones. Clients hate to pay minimal fees, especially when they send you a number of small or tiny projects, and these fees could easily increase the budget two- or threefold. They equally dislike PM fees, always telling suppliers that they want all-inclusive rates for everything, and these rates must include both PM and minimal charges.

Regrettably, there is no such thing as free cheese… Rates cannot be all-inclusive. Regular rates are based on certain assumptions, including quality source materials, availability of all required ancillary materials, «typical» client-side processes, file formats, expectations, requirements, etc. Average productivities are also calculated on medium to large volumes, where overhead is limited. No rates can cover all deviations from averages or standard expectations.

Let’s discuss the nature of PM fees, minimal fees, and additional charges in more detail.

Project setup charges

Project setup and management (PM) is a part of any project. It reflects effort spent on an average «zero-volume» or tiny project and varies depending on the project type and complexity. It is important to emphasize that project setup and PM costs are typically very moderate, mostly limited to 0.5 — 2 hours, and are negligible on any projects except for small or tiny ones.

Project setup/management charges cover the time required to process each request, provide communication, assign a team (or individual) to the project, replace/substitute resources as needed, check files prior to delivery, deliver the materials, process feedback or redeliver if necessary, invoice the project, issue POs to and pay all external suppliers, etc.

For small or tiny projects (effort below 1 hour) setup/management effort typically either exceeds or is comparable to the volume of work itself. For example, consider a 2-word translation project, or creation of a 5-second animation.

If you need a more tangible illustration, call a plumber to do a tiny job (5–10 minutes) and then try to pay precisely for 1/12 or 1/6 of an hour. You may learn a lot about yourself as well as enrich your vocabulary…

Minimal fees

Minimal Fee is NOT the same as PM charges. While PM charges cover non-production overhead [related to a zero-volume project], minimal fees cover very small production volumes (calculated effort < 1 hour).

The reason for charging minimal fee is obvious: if charged by the unit (word, art piece, etc.), the cost will be truly tiny. At the same time, freelancers and subcontractors are typically not ready to take a tiny job (including the overhead involved in switching to it from another task, responsibility, financials, etc.) and get paid close-to-nothing. They would rather charge their client (the supplier) the so-called fair minimal fee, typically between 0.5 and 1 hour or decline the job. The same applies to inhouse resources: assigning them to a tiny task typically takes considerably more of their time than the meagre amount covered by unit-based rates. The suppliers have no other choice but charge a similar minimal fee to the client.

Even in that case the difference in the price paid to the freelancers or subcontractor (1 hour) and the price paid by the client (1 hour) is typically too small (it’s just the markup) to make the project profitable. That is why suppliers would normally charge both the PM fee and the minimal fee for tiny projects; otherwise such projects incur inevitable losses for them. One should realize that we are talking about a very moderate amount of money, typically limited to 1 workhour in total.

Additional work

Additional work is caused by special circumstances or project specifics that result in the need to perform additional tasks not included into standard rates. Additional work is strictly project- and process-specific. There is no additional work for standard projects that proceed normally.

Typical reasons causing additional work include the use of proprietary/challenging tools, processes or file formats, issues with handoff materials (incomplete, unclear or corrupt), unexpected complications, bulky extra guidelines or special requirements that need to be followed…

A typical example is a small, one-time design project where the supplier needs to study an extensive corporate brand book first, and this process requires a lot of time, often more time that the work itself.

Summary: PM fees, minimal fees and additional work

All service providers incur project setup/management costs. They also lose money on small and tiny projects, unless minimal fees are paid, and on extra tasks not directly included in rates, unless these tasks are paid separately. They need to get compensated one way or the other. If these expenses are not covered explicitly, they are simply hidden somewhere. Two most typical scenarios are explained below:

• The supplier has in-depth knowledge of your projects. They do not charge for PM, tiny projects and/or additional tasks separately because volumes are big enough, or there is a steady, intensive, easily automated flow of similar jobs.
  This same approach can be negotiated explicitly, which is beneficial for both parties. Clients get higher transparency and may also negotiate better rates or higher discounts. Suppliers will get the guarantee that they will get paid for additional expenses in cases when volumes temporarily decrease, or large projects disappear.
• The supplier sets higher profit margins to offset all PM/setup charges, minimal fees and extra tasks. You are not charged for these tasks separately, but… either pay higher rates that already include extra «padding», or get services of questionable quality, because the supplier was forced to select the cheapest resources to increase the margin. This is not an enviable scenario either. Apart from potential quality or scheduling issues, you always cover this extra «padding», with or without extra effort…

Do not try to skimp on PM fees, minimal fees, or additional work. The greedy pay twice…

Neither of the scenarios above is very appealing. It is much better and highly recommended to discuss all charges related to PM/setup fees, minimal fees and additional work openly and honestly in advance. Most suppliers are ready to provide discounts or waive these charges completely if/when volumes are high, at least some projects are big and bring a higher profit margin, or clients provide them with a steady flow of small jobs that can be automated. Also, clients often have simple ways to minimize the supplier effort on their end. This includes streamlining processes, using better tools, combining tiny jobs into bigger handoff batches, etc. But clients are often not implementing these optimizations without proper stimuli, i.e. unless they have to pay extra for handing off 10 tiny jobs instead of a single, bigger one. In real life trying to save on these charges often results in paying higher margins on everything, taking unnecessary and unknown quality risks, or losing a decent, quality supplier.

Best practices to minimize the number of rates and simplify their structure

1. Divide similar tasks with varying productivity by task complexity rather than by secondary factors, such as tools or processes applied, etc.
   For example, it is best to define 2–4 rates for page setting or art creation/localization by complexity (simple — medium — complex — ultra complex), not by the tool or particular process used. The same applies to hourly rates, where it makes sense to set as few distinct rates as possible. For example, one can apply a high consulting rate, a regular rate that covers PM and engineering, and another, lower rate for all other types of work. In video creation, where the most logical unit is 1 second of the final video, we differentiate between Simple 2D, Standard 2D, Cartoon 2D, Pseudo 3D, and Full 3D videos (not by tools or processes).
2. Apply multipliers for various levels of complexity rather than setting these rates independently. For example, a 95–99% fuzzy match is charged at 25% of the new word rate, or 1 second of a 2D cartoon is twice as expensive as 1 second of a standard 2D video. This makes the whole rate system considerably easier to maintain and update. When rates change, one only needs to change a very limited number of «cornerstone» rates. All others are recalculated automatically.
3. All tasks with a high level of unpredictability, for which one cannot come up with a reasonable, uniform productivity, should be paid by the hour.

Example: Cost structure for written translation and localization

As far as we have extensively discussed ways to do the rate structure right, at least one comprehensive example of a viable rate structure is due. I have selected the field I know better than some others, written translation.

Nowadays translation extends far beyond the traditional, purely human, typewriter-era effort. Rates associated with most popular approaches or technologies are listed below, along with some basic recommendations.

1. Human translation + Review. The only proper way to measure human translation and editing is in source words (glyphs in cases when the source language is hieroglyphic, like Chinese or Japanese).

• Under special circumstances, when words cannot be easily counted, one can work with rates per string or other unit, calculated based on the average number of words per this unit.
• TM recycling needs to be counted separately at all times. All wordcounts need to comprise the complete spectrum that includes new words as well as all types of TM fuzzy or complete matches and repetitions. If units other than words are used, one needs to calculate recycling based on these same units.
• Rates for TM fuzzy matches and repetitions are best expressed as fractions of the new word rate and depend on the recycling effort. Lower fuzzy match levels result in higher effort (lower discounts). For example, 95–99% matches can be charged at 30% of the new word rate (weighted effort = 30%), which is equivalent to a 70% discount.
• When all TM-related rates are agreed upon, one can measure volumes/effort in so-called weighted (or effective) words. Under this model the wordcount for each fuzzy match category is multiplied by their relative weighted effort. Resulting weighted wordcounts are summarized across all categories, from new words to all types of fuzzy matches to complete matches and repetitions.

IMPORTANT. The «unweighted» total wordcount does not represent TR effort in cases when TM is applied (which is common practice). It could be 10% new words and 90% repetitions, or the other way round. The actual effort (and cost) may vary significantly depending on the level of TM recycling and quality in each case. That is exactly the reason why packaged rates, such as rate-per-page, do not work.

WHAT TO EXPECT. The times of the so-called TEP (translation + editing + proofing, three pairs of eyes) are long gone. Given the current level of pricing pressure from clients and global competition, the best that decent suppliers can provide under the circumstances is translation + review (TR, two pairs of eyes). I know of numerous cases where suppliers do not provide even that… (We are not talking about special cases of medical or other highly specialized translations, where subject matter experts need to be involved, and the number of verifications/checks increases). Sometimes clients even request third parties to back-translate the content for QA purposes, to make sure everything is translated correctly…

IMPORTANT. While clients sometimes request rates for «raw» human translation (without editing), I would highly recommend to never apply this approach. This puts the supplier in a situation where they cannot provide a reasonable level of quality assurance. As a result, either the supplier feels no responsibility for the quality of work, doing only a part of it, or the client is always unhappy and tries to make the supplier add some extra quality control, which means… squeezing review into the translation-only rates. The same applies to standalone editing after an anonymous/unknown source, because quality is once again unpredictable…

2. MT editing is typically charged as a fraction of the human translation rate. The discount is defined by the editing speed, which varies significantly depending on the language pair, MT engine, subject matter area, MT training (if applicable), and other things. It is hard to predict this productivity upfront. The most reliable ways to estimate editing productivity are as follows:

• Stage an actual benchmarking experiment on a limited subset of the volume that needs to be edited, or
• Run a limited holistic LQA on a representative subset of raw MT as described in detail in a separate article, How to Efficiently & Affordably Evaluate MT and Crowdsourcing Translation Quality). Evaluate editing productivity based on holistic adequacy and intelligibility scores.

IMPORTANT. As discussed earlier, n-gram-based automated evaluation scores or editing distance do not provide reliable estimates for the MT editing speed. I strongly advise not to rely on these methods when negotiating MT editing discounts.

IMPORTANT. As emphasized earlier, MT introduces a much higher risk of serious adequacy errors «slipping through the cracks». It necessitates a totally different approach to editing. One needs a bilingual editor who carefully compares each translation to the source. Given that very often the editor has no way to clearly distinguish between units with MT origin and others, all units need to be reviewed very carefully. This results in a significantly higher editing/review effort compared to the review carried out after a human translator.

In other words, even if editing after MT increases productivity twofold compared to human translation from scratch, the supplier cannot provide a 50% discount. This is because the proper process also involves review effort that cannot be skipped. Moreover, the volume of this review effort, if done properly, increases significantly compared to the review of human translation.

Let’s consider an example above in more detail. Under normal conditions one editor typically covers after three good translators. Thus, the overall translation + review (TR) effort consists of 75% translation and 25% editing. Let’s assume that editing after MT is twice as fast.

• This means that MT editing accounts for 37.5% of the original human TR effort.
• As mentioned earlier, adequate MT review is approximately 1.5 times slower due to the necessity to conduct more in-depth, bilingual checks on all units. This is equivalent to another 37.5% of the original human TR effort.
• The total (MT editing + in-depth review) amounts to 75% of the human TR effort.

This reduces the maximal discount to 25% as far as the supplier keeps the quality at the same level and preserves the same profit margin. Ideally, both parties, the client and the supplier, need to benefit from a faster, more advanced workflow. They need to share the savings, so the actual, viable discount will be even less than 25%…

SUGGESTION. It is best and easiest to define 3 to 4 different MT editing discounts covering certain productivity gain ranges, similar to TM fuzzy match rates (see above). For example, you can define four MT editing categories and discounts associated with them as follows: meagre productivity gain (less than 10%; no discount), moderate productivity gain (10–25%), good productivity gain (25–40%), or excellent productivity gain (above 40%; discount up to 25–30%).

3. Raw MT. Raw MT does not require direct human translation effort, but there can be a small cost associated with using private MT engine subscriptions and other resources. Raw MT is typically either provided for free (when volumes are limited and/or human effort is involved at later stages) or charged by word at rates that are lower than human translation rates by one or two orders of magnitude. Human effort is involved in various other ways.

• Preparation and setup, typically charged by the hour.
• MT training only makes sense for truly big projects. It includes TM selection for the corpus, preparation/cleaning of these TMs, and training itself. Both human effort and direct cost vary significantly depending on the type(s) of MT engine, other systems used, cleaning approach and requirements, and multiple other factors.

IMPORTANT. When you skimp on proper MT selection and cleaning procedures during MT training, very often you get worse results compared to stock MT. It is the «garbage in — garbage out» principle in action.

4. All other tasks, such as glossary term mining, glossary translation and management, TM management or cleaning, let alone page setting or art localization, etc. need to be charged separately either on an hourly basis, or using the most suitable rates (pieces, pages, art pieces).

Quality Goals and Metrics

We are often ready to sacrifice quality to a certain extent given tight budget limitations and/or challenging project completion dates. In all cases this should be a conscious choice; all participants need to explicitly agree on minimal quality requirements.

Quality expectations depend on a number of factors including the project budget, timeframe, subject matter area, content importance, visibility and level of exposure. The strictness of these expectations directly affects the production process and rates/cost associated with it, often quite considerably. For example, relatively lax requirements for knowledge bases or user support communication make it possible to apply MT with minimal or no editing, reducing the cost significantly. On the other hand, very high expectations for medical translations call for a more complex, multistage process with heavy human involvement, which results in much higher costs.

To define quality expectations, we need to:

1. Select or create a quality metric most suitable for the case, and
2. Set tolerance levels («pass» or «fail») for each quality factor within this metric.

One doesn’t need to create quality metrics and set goals from scratch; it is not a very simple task. One of the most universal approaches that works under almost all conditions, from MT and crowdsourcing to polished marketing texts is the Quality Triangle approach. It is described in detail in multiple publications, including best practices and advice for multiple scenarios, and is also completely scalable and flexible.

The methodology is free to use as far as you provide explicit references to the author. You can start with existing, preconfigured «building blocks», including quality scales, error typology, severity scale, and tolerance levels, and later create your own, custom metrics based on these blocks. You can create your own 3D, hybrid quality metrics from scratch in minutes using ready building blocks provided in this article. A complete 3D quality metric comprises the following components:

• Three tolerance thresholds that divide acceptable («pass») quality from unacceptable («fail») and hereby define expectations;
• Two holistic scales for measuring selected holistic factors, such as intelligibility and adequacy (most relevant for measuring translation quality), or engagement and relevance (or informativeness) for original monolingual content, etc.;
• An atomistic error typology;
• A severity scale for assigning weights to atomistic-level issues.

Depending on the situation you can apply simpler metrics with fewer components. For example, for quicker, perfunctory reviews it is sufficient to apply a 2D, fully holistic metric. To review standalone strings or other non-contiguous content we apply «traditional», 1D, purely atomistic reviews…

Summary. Negotiating rates/prices does not make much sense unless you have clearly defined quality expectations and communicated them to all project participants upfront. Quality expectations can directly affect the production process itself, and all rates associated with it.

Understanding the SLAs, and why they are so important

SLAs represent the most overlooked and underestimated area in numerous industries, including translation. For example, quality requirements and discussions attract considerably more attention and are the subject of numerous heated discussions, even though they represent just a fraction of the overall set of expectations and requirements covered by the SLA.

SLAs summarize all things that clients consider mandatory based on their processes, needs and priorities, as well as primary risks and procedures to address failures. SLAs describe primary service aspects, including communications, production capacity, volumes and turnaround times, emergency contacts, technical and process-related requirements, metrics applied in all areas, services and/or languages covered, project booking and management, monitoring and reporting, custom project performance or quality assurance metrics, and many other things.

Clients often imply things or consider them common knowledge but fail to understand that these same expectations or requirements are not as obvious to suppliers. On the other end, suppliers often only have rather vague ideas of what is most important to their clients, what drives their decision-making or causes their ire.

How do SLAs affect rates? Mostly in three ways:

• Specific requirements often automatically disqualify certain groups of suppliers. For instance:
  • Aggressive deadlines, big volumes, challenging milestones, expected significant production peaks and valleys, complex processes, high technical support expectations, 24×7 support, etc. automatically narrow the selection down to bigger, more mature and scalable companies. Expect a higher level of service and higher rates…
  • Small volumes, agreeable timelines, lack of technically challenging expectations, and dedicated inhouse project management on the client end make a viable case for smaller suppliers or individuals. In this case overall pricing may prove lower, and bigger companies may not be as interested or competitive…
• Complex processes, custom tools, or other technical requirements, round the clock support, and similar things mentioned above not only narrow the supplier selection, but also inevitably create additional project risks or tasks that often result in increased rates or higher project cost. Typical examples of such requirements include guaranteed replies from the supplier within two hours, same-day updates or fixes for smaller volumes, etc.
• Specific CAT or MT tool requirements can often result in hidden «technological surcharges» or supplier inability to work on the project. This is true even for relatively popular tools. The naive dream of perfect tool/format compatibility and the possibility for the company to standardize on a single CAT (or any other) tool proved to be just a dream. Various clients use different tools, and suppliers have to maintain a whole tool «zoo» to comply.
  Tools cost money, and their popularity varies, while free tools have severe limitations and/or compatibility issues.
  • Suppliers may not support a certain tool, which automatically disqualifies them if a ramp-up period is not an option.
    o The required number of licenses may prove too expensive, and the purchase is not justified. In this case the supplier may need to discuss potential technology surcharges or ask the client to provide the licenses, which is not always met with understanding.
  • Freelancer licenses are not free for most tools. Freelancers may either decline a job that needs to be done in a tool they do not have or ask the client to provide a license.

SLAs are so important not just because they influence rates, but mostly because they outline the full spectrum of expectations. Leave these things out of the original RFP or agreement, and you are almost guaranteed to run into serious misunderstanding or disagreements.

• Leave special requirements or process details out of the RFP, and unsuspecting suppliers will be seriously misled. They may seriously over- or underestimate the effort and propose unreasonable rates that are either way too high or too low.
• Omit the requirement to provide same-day updates from the contract, and both the client and suppliers are in for a nasty surprise…

Summary. Treat SLAs with respect and make them a part of any agreement. Start with formulating the most basic requirements in major areas, such as communications, production capacity, volumes and turnaround times, emergency contacts, technical and process-related requirements, quality control measures, project monitoring and reporting, metrics applied in all areas, required services and/or languages covered, project booking and management, etc. SLAs can affect both supplier selection and rates.

Summary — Rates in the Ideal World

It is essential to consider rates only within the context of the overall picture that also comprises detailed rate definitions, assessing risks and setting risk tolerance levels, and formalizing all requirements and expectations (SLA). Isolating rates from this context results in miscalculations, mutual dissatisfaction or misunderstanding, underbidding, etc. When other components are taken out of the picture, one cannot even set rate expectations or compare rates from various suppliers objectively; in each case assumptions on which rates are based are unknown and may vary dramatically.

1. Finalize the list of potential risks. Concentrate on the most damaging ones that you need to address first and separate them from the risks you are ready to take. Both lists directly influence the content of the SLA (see below). For example, if deadlines are crucial for global software builds, the SLA needs to contain an explicit set of priorities (time over quality procedures) and procedures to rectify potential deviations from the process. Otherwise, the supplier may have a different order of priorities in mind, and will act accordingly, delaying the delivery by several hours to complete all quality checks… On the other hand, if quality expectations are relatively moderate but time is of essence, one can set more relaxed acceptance thresholds and explicitly allow or insists on MT use.
2. Make sure you have a clearly written and comprehensive SLA in place. The SLA formalizes the rules of the game part of the client-supplier relationship. You may need multiple SLAs when multiple scenarios are involved, for example one regular SLA and another one for urgent projects, with different delivery, communication, or other requirements. Time on writing and polishing the SLA is time well spent. It helps suppliers to understand all client requirements and expectations in their entirety and minimizes both future misunderstanding and underbidding.
   The two most conspicuous parts of the SLA that should be described in detail are as follows:
   • Major DOs and DON’Ts, including mandatory tools and processes, MT use (must be done, is allowed, or is prohibited), communication and reporting requirements, etc.
   • Quality expectations and metrics need to be agreed between the client and supplier, and then conveyed in their entirety to all parties involved, including individuals working on the project. Unless there is a common understanding in this area, mutual dissatisfaction is almost guaranteed.
3. Reconcile the SLA with the available budget, as these two are tightly interconnected. Process, tool, delivery, quality, communication, scalability, and other requirements from the SLA can seriously limit supplier selection and define the expected reasonable price range. Very often you may need to reduce either expectations or the scope, or increase the budget. Make sure to find the right balance between the two, preferably before starting an RFP or talking to suppliers. Be prepared to redefine this balance when new circumstances arise.
4. As said earlier, the ideal rate structure is strictly cost- and effort-centered, transparent, and scalable. The advice below applies to both clients and suppliers:
   • You can combine various consecutive tasks with the same unit of measurement within a single, unified rate. (Translation + Review, Page Setting + PDF Generation, etc.)
   • You should not combine alternative tasks with varying productivities into a single rate. (Human translation, recycling from the TM, and MT).
   • Explicitly discuss minimal fees, as well as PM and setup fees for all types of work. Make sure that this area is not overlooked, because service providers incur most overhead on small and tiny projects. Discuss ways to reduce this overhead (and accompanying minimal and PM fees) and related discount models.
   • Make rates as scalable as possible by applying multipliers and/or complexity factors to productivities for more labor-intensive tasks (such as localizing complex art pieces vs. regular ones, translating highly specialized text, etc.).
   • Try to minimize the number of similar rates. For example, one can apply universal translation rates for groups of similarly priced languages. Also, it is a good idea to limit the number of various hourly rates to just two or three at most.
   • Do NOT try to set rates for tasks with seriously varying or unpredictable productivity. Just agree on paying them by the hour as far as the supplier provides a sufficient, reasonable explanation.
5. Provide a comprehensive, detailed set of rate definitions for all rates within the selected structure. These definitions will help both parties to clearly understand what tasks are included in each rate and which ones are paid separately, what productivities are expected for various tasks, what TM fuzzy match grid(s) and MT usability/discount categories are used, etc. Rates without definitions, such as «just translation» are subject to a whole variety of interpretations; they are almost guaranteed to cause disagreement later on.