Performance review season, whether it’s once, twice, or four times a year, is almost a universally dreaded time for employees and managers. Toronto-based Hypercontext is looking to change that with its recently released AI-powered feature designed to help managers generate employee performance reviews.
The tool, which produces reviews based on one-on-one meeting notes and assessment data stored in Hypercontext’s existing software platform, is aimed at helping managers generate higher-quality reviews quickly and streamlining what can often be a weeks-long review cycle.
For CEO and co-founder Brennan McEachran, the tool represents a key step into the human resources (HR) sector for Hypercontext. But the recent explosion of AI tools has also sparked discussions regarding the role of AI-driven decision-making across departments, including HR. With the technology still in its infancy, experts have raised questions about the ethical implications of relying on machines to make determinations that could directly impact employee livelihoods.
“We saw an opportunity not just to defend the space, but also be helpful.”
– Jocelyn Brown, Hypercontext
“Bringing AI into the performance review process is inevitable,” McEachran argued. “Whether or not we’re going to do it, or someone else is going to do it, it’s inevitable that this is going to happen.”
“We thought, not only can we do it better than most, but also we care,” he added. “We’re in it to make good managers and try to make employees and organizations better.”
With claims to apply AI to performance management responsibly, McEachran spoke to BetaKit about how he believes Hypercontext’s tool addresses the inherent risks.
Entering the HR arena
Hypercontext, formerly SoapBox, was founded by McEachran and CTO Graham McCarthy over a decade ago. Initially, SoapBox offered a cloud-based employee engagement app that allowed businesses to crowdsource employee ideas and feedback. In 2021, the startup rebranded to Hypercontext and shifted its focus to providing a multi-use software tool aimed to help people managers run better one-on-one meetings.
The startup’s core offering lets managers create and collaborate on meeting agendas, record notes, create action items, collect employee feedback, and receive insights to improve meetings. Over time, the startup made various improvements to the meeting app, such as integrating company objectives and key results (OKRs) into meeting flows. By spring 2022, the CEO said the “missing piece” of its app, according to the startup’s customers, was the review process.
Performance reviews can occur on an annual, bi-annual, or quarterly basis for most companies, McEachran said, but for managers, they may involve months of one-on-one meetings, feedback sessions, and multiple reviews for each employee, and can require a good deal of time, analysis, and creative energy to complete.
Jocelyn Brown, head of customer success and sales at Hypercontext, said the team had noticed most HR professionals were unsatisfied with their performance management tools, noting they were often not well integrated with managers’ lifestyles and work.
“We saw an opportunity not just to defend the space, but also be helpful,” Brown added.
Up to this point, Hypercontext’s primary focus had been on managers, but the company began contemplating the significance of HR departments as the primary entities responsible for equipping managers with the tools to become more effective leaders. McEachran also noticed that as more managers adopted the startup’s app, HR departments would interpret it as a cue to invest in performance management software suites, which ultimately kicked Hypercontext out of the process.
“Instead of running from that fight, or avoiding that fight with the well-financed American companies who are out there selling to HR, we [chose to] enter that battle,” he said.
Showing the math
Hypercontext began building its review tool in the summer of 2022, months before the explosion of generative AI tools like ChatGPT, and McEachran found there was some skepticism about whether the technology was ready. “November, December, January rolled around, ChatGPT blew up, and everyone was like, ‘I take it back. You are right. The tech is ready,’” he said.
According to Stephanie Bergman, a senior consultant in HR and diversity, equity and inclusion (DEI), and team lead at Toronto-based Bright + Early, AI is not completely new in the HR world. She noted while it has traditionally found more widespread usage in recruitment processes, such as resume screening, its applications have been expanding in recent years. But with the rapid proliferation of tools like ChatGPT, discussions surrounding these tools have shifted toward their ethical implications and potential influence on our work dynamics.
“I think that AI can be really well leveraged to give [HR teams] a head start, simplify things, review significant amounts of data or content, as long as it’s leveraged as a tool, and not as the end-all and be-all,” Bergman said.
Hypercontext’s performance review tool uses information gathered from one-on-one meetings—such as answers to scaled assessment questions, goal progression insights, and one-on-one notes—to generate an initial draft of a review. The tool then aids managers in making edits and revisions to that initial draft. McEachran said the objective is to provide managers with a solid starting point and more time to refine the review, rather than starting from scratch.
“It’s almost like having an HR business partner sitting beside you as you do this, helping you come up with better words or better phrases or making something sound more positive or more negative, without the need of actually having an HR business partner,” McEachran added.
RELATED: Thoughtworks’ Rebecca Parsons on using threat models to prevent AI peril
Another concern raised by experts is explainability, meaning, the ability to understand and interpret the reasoning behind the responses generated by an AI model. Natural language processing tools like ChatGPT have come under fire for often generating inaccurate or outdated information with little justification for those outputs. McEachran said that these tools tend to function like “auto-complete on steroids,” whereby the model’s outputs are meant to be cohesive, but not necessarily trustworthy.
“I always think of it like a Grade-A bullshitter, they’re just making up BS for their essay. And, yeah, it sort of looks like there’s an essay there, but there’s no substance,” he said.
McEachran said the “word-smithing” behind Hypercontext’s tool is built on GPT-4, the same model that powers ChatGPT’s paid version, but noted there are “significant differences in privacy, security, and training data policies between ChatGPT and the models and APIs” used by Hypercontext.
Hypercontext’s tool parses real qualitative and quantitative data derived from the notes, goals, and agendas from one-on-one meetings managers hosted through its platform. Managers can write their own notes as part of those meetings, and answer various assessment questions, such as whether an employee lives up to the company’s values, to influence how the AI ultimately generates a first draft.
To prevent the AI from hallucinating, which occurs when a model makes up a confident response that is unjustified by its training data, Brown said Hypercontext gives the AI background knowledge on best HR practices and relevant employee and company information, so it can generate accurate content. The AI also uses statistical models to minimize any text that goes against given instructions and breaks reviews down into “chunks,” rather than producing a whole draft at once.
Finally, Hypercontext also sets boundaries for the AI and trains it to use what it knows about its own role and challenges without going on “wild tangents,” Brown said.
The growing use of AI in HR has also stirred a debate about bias and fairness. In 2018, for example, Amazon scrapped its own AI recruiting tool that showed bias against women. Bergman argued that bias exists in HR whether AI is being used or not, and while AI can be used to tackle certain types of biases, such as recency bias, these models can potentially exacerbate other forms of bias.
“We know that AI tools are generally kind of biased towards the identities of the folks that built them,” Bergman said, adding that organizations should always conduct a thorough audit process of performance reviews to check for biases against certain identities before the reviews impact employees’ salaries or enjoyment of work.
McEachran said Hypercontext’s tool shows users the goals, notes, and calibration questions that were used to generate reviews, and allows managers to investigate the data points behind each draft. Comparing the review tool to a calculator, he said, “Instead of just spitting out the answer, we do the long division, step by step. We show the math.”
A tool only as good as its user
Hypercontext launched its performance review tool in late April, and according to Brown, will write roughly 267,000 words for approximately 500 reviews this month. It is certainly not the first startup to bring AI into HR, and Brown said this latest tool puts Hypercontext up against performance management players like Lattice, 15Five, and Leapsome. This is in addition to the startup’s competitors in the meeting management space, which includes Canadian companies like Fellow.app.
Now that Hypercontext has entered the HR sector, competition will undoubtedly heat up. McEachran believes as AI continues to permeate HR departments, the true differentiator will be the data quality going into each system. For that data quality to remain high, he believes the need for human managers isn’t going anywhere.
“If you show up and are trying your best to be a great leader for your team, you get reviews so easy, they write themselves,” McEachran said. “If you’re a terrible manager, you’re going to have to put in all that manual work.”
“Any tool is only as good as its user,” Bergman commented. “With Hypercontext’s [tool] specifically, it relies on the one-on-ones. It’s relying on having an excellent manager who’s properly running one-on-ones and having regular feedback conversations with the team.”
Feature image courtesy of Hypercontext.