In today's rapidly evolving technological landscape, businesses are increasingly turning to Artificial Intelligence (AI) to gain a competitive edge. Companies are strongly preferring candidates with "AI" written anywhere on their resume. However, it is crucial for companies to decide their hiring strategy beforehand. Where to hire from, and who to hire, are two important questions companies must answer before the hiring happens. In this blog, I explore the strategies of hiring an in-house versus outsourced team, as well as hiring a cheaper, lower productivity candidate versus an expensive high productivity candidate across 4 scenarios.
The Objective of Hiring People: Driving Business Metrics
The essence of hiring, whether it's bringing on board a new employee or engaging a service provider, is to make a tangible impact on business metrics. It's crucial to have a well-defined set of output metrics that are expected to shift positively as a result of the hiring decision. These metrics might not always be directly quantifiable in terms of cost savings or revenue generation, as the relationship between different metrics can be complex.
Take Amazon, for example, where Customer Satisfaction is a key metric. A customer who rates a product 3 out of 5 and another who rates it 5 out of 5 may contribute equally to short-term revenue. However, over time, higher customer satisfaction levels are assured to result in an exponentially higher revenue, as witnessed from the rise of Amazon as a Tech giant. However, linking a higher customer satisfaction score to higher revenue is not always possible on a project basis. Hence, organizations often use intermediary metrics like Customer Satisfaction score as the success metrics of a project. This practice is healthy, as long as the business leaders have enough conviction in the correlation of the project success metric and profitability.
Therefore, the primary objective of hiring in any role is to meet a success metric that in turn, has a positive effect on the company's financial health. The shift in the success metric is viewed as the "return" on the investment made in hiring.
The Key Metric for Measuring Project Success: ROI
It's a common mistake for enterprises to focus solely on metrics like Total Money Saved or Total Products Sold as indicators of a project's success. This can lead to situations where a company might spend $1 million to save $100,000 and still consider the project a success because they're measuring success in terms of Dollars Saved. This approach can be misleading, especially when the time utilization of non-contractual employees is not accurately tracked, making it challenging to ascertain the actual costs of a project.
In line with Milton Friedman's principle that a company's primary objective is to generate profits, the most pertinent metric for evaluating any business action is the Return on Investment (ROI), which is calculated as the ratio of money earned to money spent. A higher ROI indicates more efficient growth for the company.
Therefore, when deciding whether to outsource or insource a project, ROI should be the guiding metric. This can be approached by assuming either a fixed investment with variable returns or vice versa. For the sake of simplicity, let's consider the "returns" as fixed, where the predefined metric shifts from point X to point Y. The variable in this equation is the investment or the costs associated with achieving the desired movement in the metric.
Costs involved in Insourcing
Fixed Costs
(I) Sourcing - Sourcing a good candidate with required skillsets who also aligns with the culture of your company is not only a direct expense, but also an indirect expense in terms of the time required to close a candidate. Typically, closing a good AI Engineer takes 3-4 months, and requires at least 2 months worth of salaries of that position.
(II) Onboarding and Training - An AI Engineer is typically trained for 2 months before she is put into production. Additionally, a month also goes into Knowledge Transfer.
(III) Compensation - Joining Bonuses typically range from 1/5th to 1/3rd of an AI Engineer’s salary. Additionally, companies assume at least a 2 year commitment from an Engineer when they hire, and hence, 2 years worth of salary is also a fixed cost, just that it needs to be paid in installments. Annual cost of other benefits like insurance and other reimbursements is close to a month’s salary.
(IV) Infrastructure and Tools - An AI Engineer will be provided with with her own laptop, office space, and office tools.
(V) Termination Cost - Cost of terminating a full time employee is extremely high in terms of bad PR of the company [optional].
Variable Costs
(I) Salary and Performance Bonus- Salary after 2 years, and the bonus paid out upon exceptional performance of the company or the individual are variable costs.
Costs involved in Outsourcing
Fixed Costs
(I) Vendor Selection Costs - Similar to candidate sourcing, vendor selection is a time consuming process. It is important to choose a service provider that has the required talent in case of an unprecedented project, or prior work to show for, in case of a project that has been done before in the industry. Most importantly, the service provider should have leadership who you can put your complete trust in. The vendor selection process often involves a PoC that lasts a month. Assuming 3 vendors to select from, it equals 3 months salary of an outsourced AI Engineer.
(II) Onboarding & Training Costs - Onboarding costs being the same, the training costs are much lower for an outsourced entity, as the mandatory company trainings are not required, Moreover, since most vendors like Tailored AI complete a Proof of Concept, or PoC, before being onboarded, they already undergo half the training required to take on the project. Hence, onboarding and training costs are merely 2-3 weeks.
Variable Costs
(I) Vendor Fees - Service providers often charge an hourly fee for AI projects, since the outcome can be uncertain. However, some also charge project based fees that will be paid once a particular outcome is achieved. Depending on the number, complexity, and the size of projects, the fees will vary.
(II) Communication Costs - Since the service provider often hosts their team off-site, there is a cost of communication involved. However, customer obsessed service providers like Tailored AI work in multiple time zones and are always attentive to the customers’ needs. Coupling that with a talented group of people who get exactly what the customer wants This helps in drastically reducing the communication costs.
(III) IP and Data Security Costs - Service providers scale by replicating case studies. This means that even though the customers get the code, the same problem statements can be pitched and solved for competing companies by the service providers. If this happens, there are significant legal costs that a company needs to bear. To ensure this does not happen, contracts should be airtight in terms of IP rights, and the reputation of the service provider in terms of data handling should be considered. Additionally, compartmentalizing the project also helps in maintaining the IP, as only the sourcing company is aware of the end to end project details.
(IV) Quality Assurance - Companies with strict code quality policies will need to spend on ensuring that the code delivered by the service provider is up to the company’s delivery standards.
Measuring the ROI
Although the output metric may not directly be inflow of money, the decision maker should be aware of the approximate impact on the inflow of money for the output metric that he has decided. In absence of this understanding, it would be impossible to ascertain whether completing the project makes sense at all. Typically, companies look at 5X estimated ROI in terms of inflow of money / money spent to take on a project.
If the project makes sense, then to choose between insourcing and outsourcing, the only deciding factor is the total cost. Other than the costs mentioned above, an important factor is the opportunity cost of not completing the project. A project completed 2 months early delivers impact for 2 extra months than a project completed 2 months later. We will consider this while calculating the total costs of the project.
Scenario Analysis
To model ROI calculation, we will consider two parameters. (1) Income of the country the company is based in (2) Complexity of the Project to be done. Each parameter can take values High or Low, giving us a total of 4 scenarios.
There are two decisions to be taken for each of the scenarios (1) Whether to insource or outsource and (2) Whether to select a High talent or Low Talent candidate. Ideally, the metric we should look at is Developer productivity, however, since productivity cannot be known beforehand, we use talent, or on-paper skill of a developer as a proxy. This is merely for modelling purposes, and does not indicate lack of talent in a person in general.
Here is some data, and some assumptions where data is not readily available.
Annual Salaries of high and low talent AI Engineers in high[1] and low income countries[2]
Outsourcing Rates for High and Low Talented Engineers in High and Low income countries[3]
Assumptions
I’ll assume a revenue of 2X the salary of an in house AI Engineer upon successful completion of a project.
Project Completion times for high complexity and low complexity projects for high and low talent engineers[4]
Cost Analysis
The independent variable being project completion time (X), we will now calculate the cost function for each scenario for variable talent and complexity of project. For our model, since it doesn't make sense for any company to outsource projects in high income countries, we will skip that scenario.
Scenario 1 - A company in a high income country doing a complex project
(I) High Talent - In House → 360 (2 years of salary) + 30 (recruiting cost) + 60 (bonus) + 30 (infrastructure cost) + max(0,x-2) * 15 (salary after 2 years) = 480 + max(0,x-24) * 15
(II) High Talent - Outsourced → 28.8 (onboarding cost) + 7.2 ( KT ) + 9.6 * x (vendor fees) + 1* x (communication costs) = 36+ 10.6 x
(III) Low Talent - In House → 240 (salary) + 20 (recruiting) + 40 (bonus) + 30 (infra) + max(0,1.5*x-24) * 10 (salary after 2 years) + (1.5x-x) * 2 * 10 (opportunity cost of late completion) = 330 + max (0,1.5x-24) * 10 + 10x
In this scenario, In-House low talent is the clear winner
To Summarize
For 4 different scenarios, we get 4 different answers
What are the limitations of the mathematical model?
The results reported above are generated for the assumptions mentioned in the post. The assumptions will fail towards the tails of the bell curve, where extremely complex projects can't be completed by anyone except the topmost scientists even if they are provided with an infinite duration of time, while the easiest projects can be completed successfully by even interns at virtually no cost.
The next challenge is identifying a talented candidate, or a service provider with talented candidates. It is impossible to merely ascertain this from interviews or assignments. The best signal of talent is often the alma mater of the employee / service provider. Service providers like Tailored AI exclusively hire from world renowned Engineering colleges like IITs, IIITs, and BITS to tap into the country’s top talent pool.
The cost structure in the proposed model heavily discounts the benefits of having an in-house developer if the company’s core focus is on technology. Having an in-house team of talented developers who understand the Tech well is a moat that can’t be easily broken into.
Next, companies often weigh the convenience of being able to communicate with the service provider or employee in their native tongue and accent heavily. It is frustrating not being able to get the point across, and it can't simply be put in as a cost in terms of communication cost. In these cases, it is important to choose candidates or service providers who have previously worked with people based in the native country.
Last but not the least, compliances imposed by the law often dissuade companies from doing things outside the boundary of their organization. Here, it again helps to choose service providers having previously worked with companies under the same jurisdiction, as they are well aware of the compliances.
