The one question that should run every AI-in-HR decision

A filter from Colgate-Palmolive's Michaela Schoberova for when AI is a multiplier and when it is a quiet trade-down

Most AI-in-HR roadmaps are built bottom up. Someone lists tasks the function does, ranks them by hours spent, and points AI at the top of the list. That is how you get to talent acquisition bots screening candidates who used AI to write their applications. Slop on both ends, not efficiency.

There is a better filter. I picked it up from Michaela Schoberova, SVP HR for Global Commercial Transformation at Colgate-Palmolive, on the latest Workestration episode. Plainly stated:

For any AI move, ask what relational, cultural, or trust-based asset you are about to convert into a synthetic shortcut, and whether that conversion is a trade you actually want to make.

Use it on every line item in your roadmap. Some might stay and others will quietly come off (good riddance!)

Where the filter came from

I asked Michaela whether Colgate, having built proprietary digital consumer twins for product innovation, would build the equivalent for employees. Skip the surveys. Get continuous signal. Faster, cheaper, never grumpy at 4 p.m. on a Friday.

Her answer was a careful no. The reasoning is very useful.

She named three reasons. There is less data on the employee side than on the consumer side, and twins need data. Privacy is in a different ethical category at work than it is in a search trail. And, the part that runs everything: Colgate has over time invested deliberately in relationship-driven HR. The company considers that investment as competitive advantage. Replacing the relationships with synthetic proxies trades a long-term asset for a short-term gain. The technology could probably do it. The value case however will be short-lived.

This is the perfect example of the filter in action - holding a useful tool against a question of what it costs you in the assets you actually compete on.

The worked example: where digital twins do pass the filter

The case for synthetic consumers is documented in MIT Sloan Management Review's piece by Thomas H. Davenport and Randy Bean. Colgate's system blends internal research, partner data, public datasets, purchase histories, search behavior, and physiological responses into virtual consumer personas. The twins compress the early stages of product innovation. Real humans still test what gets shipped to consumers.

Run the filter on this and the answer is clear. Most consumers do not want to sit in focus groups (ever.) There is no long-term relationship asset being converted. There is no trust contract being eroded. The twins are doing work that nobody in the system was doing well anyway, and the company still validates with humans before anything reaches a shelf.

The bonus is unglamorous. Michaela's team partnered with academic researchers to quantify the qualitative output of the twins, so two synthetic respondents can be compared as cleanly as two real ones. That is the kind of investment that makes the difference between a flashy demo and a system that holds up (and makes the team proud in the process.)

The same filter pointed at the employee twin idea gets a different answer. Less data, more privacy risk, and a relationship culture they are not willing to convert. Same technology, different verdict, because the assets are different.

What the filter tells you about adoption

The standard AI rollout looks like this: pick tools, write a policy, mandate training, expect adoption. The standard outcome is a chasm between a handful of enthusiasts and a long tail of people who never log in.

Colgate's approach has two parts. The AI Hub is a single front door to Gemini, ChatGPT, and agent-building tools, all behind a firewall. Access requires a short mandatory training. The training mostly works as reassurance for cautious users that this is sanctioned.

Adoption then runs through voluntary peer cohorts called "Super You." Groups of people doing similar jobs trade concrete applications, including the ones that failed. Voluntary, role-specific, optional, built around what people actually do.

Run the filter here too. Top-down prompt training would convert peer trust, the relationships in a team, the safety to admit you don't know how to use a tool, into a synthetic shortcut: the LMS module. That conversion costs you exactly the substrate that makes adoption work. The peer cohort design protects the asset and rides on top of it.

What the filter tells you about HR's AI use cases

Michaela was specific about where HR points AI inside Colgate. Coaching at scale, because coaching used to be reserved for executives and the better tools can offer it to middle managers now. Knowledge management, so a new hire can find the right policy in seconds. Learning content built at speed. Skill-based career and talent management, which she described as much harder to get right but worth the effort.

Did you notice what is missing? The talent acquisition arms race. The "always-on" engagement bot that promises to replace the engagement survey. The synthetic skip-level.

Apply the filter to each.

AI coaching at scale: the tools that ask questions, set goals, and remind you of them are doing work most managers were not doing consistently anyway. There is no relational asset being converted. The thing being scaled is a service that, in most organizations, only senior leaders ever got. Anna Gallotti, who chairs the ICF Task Force on AI in Coaching, has been arguing that the right design augments human coaches rather than replacing them. The filter agrees with her.

Knowledge management: pure efficiency play, no relational cost. Easy pass.

The engagement-survey replacement: this is where most HR teams trip. The synthetic shortcut is appealing because surveys are slow and people don't act on them. The asset being converted is the contract that asking employees what they think is a real signal you intend to listen to. If your survey program is already broken, fixing the listening loop matters more than building a better predictor of what people would have said.

Talent acquisition bots at the top of the funnel: probably a bad trade. The asset is the early signal of how your company treats people, which candidates use to decide whether to keep paying attention. Converting that to a chatbot saves recruiter hours and burns a much larger pool of slow-build trust.

The point of the filter is not to be conservative but rather to make the trade explicit.

A deeper version of the filter

Michaela's lens on this is shaped by her training in applied positive psychology at Penn. The deeper version of the filter is character strengths research, the taxonomy of traits that societies across history have consistently valued. Courage, kindness, wisdom, fairness. The argument: any AI adoption decision should be evaluated against those traits, alongside productivity gains. Agency without values is just velocity.

She made the case practically. The single biggest predictor of wellbeing and longevity in the research is human relationships. Anything that erodes them, including useful technology, deserves scrutiny. We may already be losing ground we cannot easily get back.

The clearest version of this idea came not from a research citation but from her daughter's first-grade classroom. The school sends seven-year-olds to the local bakery, bookstore, pharmacy, firefighters, and EMTs, to ask questions and build community. The skill being taught is the one Michaela thinks adults are losing. If schools do not teach it, employers will have to.

Colgate has a grown-up version of this called Footsteps, where employees collect points for going out and talking to actual consumers (not the digital twins.) The program runs in parallel with the AI investment, on purpose. The filter at work: the most contrarian move you can make during an AI rollout is to protect the human contact the technology cannot replicate.

Leadership for a BANI world

The filter has a leadership counterpart. Toward the end of the conversation, Michaela referenced Bob Johansen at the Institute for the Future and his recent book Navigating the Age of Chaos, co-authored with Jamais Cascio (who coined BANI) and Angela F. Williams, published October 2025.

BANI is the post-VUCA frame: brittle, anxious, non-linear, incomprehensible. Michaela's three antidotes:

Empathy, to lower the ambient anxiety around you. Clarity, to choose what to focus on even when you don't know everything, and to name explicitly what is not changing. Agency, to move forward with imperfect information, set boundaries inside which others can experiment, and tie the whole thing to values so speed has a direction.

Each maps onto a part of the filter. Empathy is the awareness that relational assets exist in the first place. Clarity is the discipline to name the trade you are making. Agency is the willingness to make the trade once you have named it.

What HR will regret in three years

I asked Michaela what HR will regret not acting on now. Her answer:

The skills people will need. Critical thinking, because AI is always positive and reinforces what you already believe. Relationship building, because the super-variable for wellbeing does not change. The ability to ask better questions, which is a learnable skill we are not currently teaching well.

The way HR delivers learning and development. People are using these tools at home for personal coaching, basic research, and reflection. The L&D function that still ships traditional courseware will, in three years, look oh so outdated.

HR's seat at the strategy table. Not because HR is owed it, but because the work the function does, on capability building, organizational and work design, and adoption, is exactly the work that determines whether AI investment pays off.

What to consider (and may be take action on Monday)

Three moves, all of which run through the filter.

Audit your top ten AI use cases. For each, write the one-sentence answer to: what relational, cultural, or trust-based asset are we converting, and is the conversion worth it? Anything that fails the test does not need to be killed, but it does need to be redesigned.

Build a peer cohort. Voluntary, role-specific, structured around concrete examples including failures. The skill being built is judgment about when to use AI, which is only learnable by doing in the company of peers.

Run a Footsteps program. Send people out to talk to actual customers, frontline staff, or each other, on purpose, during the AI rollout. The investment compounds in exactly the direction the technology cannot.

The most quotable thing Colgate has done is build synthetic consumers. The thing worth copying is the discipline to ask, every time, what asset the synthetic version is replacing, and whether the company can afford to lose it.

Listen to the full conversation at Workestration.ai or on [Apple Podcasts], [Spotify], [YouTube].

References and further reading:

Get the latest episodes directly in your inbox