Algos are automated tools for execution. The reason for using an algo as opposed to a risk transfer price is for the potential of performance enhancement. The economic reason behind this argument is very similar to car services. If you need a temporary car, you normally have two options:
take a taxi or hire a car.
In general taking a taxi is more expensive because you need to pay the labour of a driver plus the cost of using their car. This is equivalent to paying the risk-transfer price, in which the liquidity access fee (the cost of usage of car and petrol) and risk management fee (labour fee) are all packaged into a single price. Hiring means you only need to pay the rental fee, which is equivalent to the commissions in algo executions, but you also need drive yourself, i.e. manage your own risk.

Do you need an algo?
The answer is, as always, it depends. Renting is not necessarily cheaper especially for short and easy journeys. Another scenario when people prefer taxis is when the weather is bad or road condition are tricky. Essentially people are paying for the taxi in exchange for convenience and peace of mind. Both scenarios are exactly what we see in the FX market. People tend to use risk-transfer prices when the tickets are small, or the market is challenging, e.g. a large ticket in an EM pair.
Algos, in contrast, may function well for large tickets in normal market conditions assuming you are happy to spend the time to manage them.
In summary, the choice between algo and risk transfer price should be based on the factors including the size of ticket, market conditions, risk appetite and the amount of time you want to spend on a ticket. If you decide to use an algo, the next question to answer is what type of algo to use.
Types of algos
Broadly speaking, algos can be divided into two categories, rule-based vs optimisation-based. These algos can have simple rule-based trajectories (TWAP/VWAP) or behaviours (the so-called opportunistic algos). They cruise along themselves in the market, but their navigation ability is pretty basic. Essentially rule-based algos provide a tool to grab liquidity in an empirical way. Just like driving a conventional car, a lot of traders prefer to monitor the algo progress closely and steer it in flight, e.g. control the limits, pause or resume the algo, change gears (urgencies) or duration, etc.
In contrast, optimisation-based algos are like self-driving cars, which are more sophisticated and dynamic. The key component, which distinguishes them from the rule-based algos, is the self-steering system. It detects the change of market conditions and makes intelligent adjustments in an optimal way. An algo, which can dynamically to minimise the implementation shortfall, falls into this category. Although you can intervene with these algos “in flight”, they are designed to do their own job without interruptions.
Arguably the optimisation-based algos are a bit like blackboxes. As a result, a lot of traders feel more comfortable with simpler rule-based algos. However this is up to the preferences. The two types of algos suit two styles of trading:
hands-on or hands-off.
This in turn determines the different metrics in the performance evaluation.

Performance of algos
Generally speaking, algo performance boils down to two main factors:
its “mechanical capability” to grab liquidity, and its “steering ability” in different market conditions.
For the hands-on traders, who would like to steer the algos themselves, the “mechanical capability” should be the main focus. To measure it, it’s important to strip out the impacts from the general market movements and human interventions. This normally requires the scrutiny at the child slice level on the micro decisions of liquidity grabbing.
For the hands-off traders, they need to look at the overall performances of the whole package, i.e. take a macro view. The performance against the arrival price at the inception of the algo is the common metric for this purpose.
As we all know, it is crucial to use as much data as possible to get accurate performance measures. However, it is equally important to use the right metrics and take either a:
micro or macro approach
to suit a particular trading style.
Conclusions
Like most things nowadays, the FX market has become more and more data driven. In general the decision making processes from the execution desks have also become more robust and systematic. While the developments are rapid, there is still a long way to go for industry-wide usage of FX algos to become mature. At the current stage, human preference is naturally part of the decision making processes although more data means more statistical based decisions can be made. However, statistics can be sometimes confusing. While so many different benchmarks and metrics can be used, it is often a bit ad-hoc to cherry pick the preferred ones. Therefore it is very important to understand what a statistic really represents and choose the right ones to suit a particular types of algo and trading style.