Stay ahead of the curve
Get exclusive insights delivered directly to your inbox. Join thousands of tech leaders.
TL;DR
- Expert bias in R&D is judging a new approach by the standards of the approach you know best.
- It is strongest when the alternative is worse today but improving faster: exactly the case that decides the next decade.
- In 2007 GE launched a high-efficiency incandescent programme. LED was improving more than five times faster. GE suspended the programme 20 months later.
- Put a measured improvement rate next to expert opinion and treat disagreement as the signal.
The moment every R&D leader recognises
GetFocus co-founder Jard van Ingen has sat in the room many times. The leadership team calls in its experts, and "the experts can calculate to six decimals why it will never work. Everything is black and white."
They are usually right about the calculation. The calculation is about today.
What is expert bias?
Expert bias is the tendency of specialists to judge unfamiliar approaches by the standards of the familiar one. In R&D it shows up as confident dismissal of an alternative that underperforms today. Experts measure the gap. They rarely measure how fast it is closing.
| Experts are strong at | Experts struggle with |
|---|---|
| How the current approach works and fails | Approaches from outside their field |
| What is feasible today | How fast a weaker alternative is improving |
| Practical constraints: cost, scale, safety | Change that looks flat, then compounds |
| Depth in one field | The cross-field picture a strategy needs |
Expertise is built on the problems a company already has. The next solution often comes from a field nobody in the room specialises in.
The experts built a better filament
Backtest · Lighting · 1990–2026
The experts built a better filament
In 2007 GE launched a high-efficiency incandescent programme. LED was improving more than five times faster.
Improvement rate per year, computed from patents public at each date · GetFocus backtest: incandescent vs LED
In February 2007 GE, the company Thomas Edison founded, announced a High-Efficiency Incandescent programme targeting twice the efficiency by 2010. It was the expert answer: decades of filament know-how, a clear engineering target.
The GetFocus lighting backtest shows what the patent record said that year: LED improving at 47% a year, incandescent at 8.8%. The rate had favoured LED since 1990, three years before the efficient blue LED existed. In December 2008 GE suspended the programme "to place greater focus on LED and OLED". In 2020 it sold the lighting business.
Treat disagreement as the signal
When outside evidence and expert opinion disagree, most teams treat it as a problem. The best leaders treat it as the point. Christophe Perthuisot, former SVP of R&D at Moët Hennessy and Danone, on what happened when improvement-rate data contradicted his experts: "That is great. What you want is conflicting perspectives. You want the system to challenge existing thinking and the prevailing paradigms within the company." (Read his story.)
He also knew how his own teams behaved under pressure: people heading programmes will fight for them. Structure beats goodwill. Put a number in the room that nobody owns.
How to put evidence next to your experts
- Ask a different question. Not "will it work?" but "how fast is it improving, and when does it catch up?"
- Bring a rate, not a snapshot. The Technology Improvement Rate is computed from the patent citation network, the same record your experts' own field writes into. It is explainable maths, not a black box, so experts can interrogate it.
- Backtest on a call they remember. Show how the method would have judged a past decision in their own industry, using only the patents public at the time. Sceptics move on evidence they can check.
- Keep both voices in the decision. Experts own feasibility, cost and constraints. The rate owns direction and pace. Christophe: "The actual decision to invest remains with the head of R&D."
Backtests · no hindsight
Test the method on a call your experts remember
Each case is replayed using only the patents public at each date.
| Case | Rate called it | Filings agreed | Lead |
|---|---|---|---|
| Data storage HDD vs SSD | 2001 | 2010 | 9 years |
| Lighting Incandescent vs LED | 1990 | 2002 | 12 years |
| Automotive Hydrogen vs batteries | 2001 | Never | 25 years and counting |
| Engines Rotary vs piston | 1972 | Never | 54 years |
Source: GetFocus backtest library
FAQ
What is expert bias?
The tendency of specialists to overweight what they know and judge unfamiliar approaches by the standards of familiar ones. In R&D it shows up as confident dismissal of alternatives that are weaker today but improving faster.
Should R&D leaders trust data or experts?
Both, side by side. Experts are best at constraints and feasibility today. A measured improvement rate is best at direction and pace across fields. Where they disagree is where the most valuable questions are.
Why is technology forecasting difficult for R&D teams?
Because most teams forecast from current performance, and exponential improvement looks flat until it doesn't. Forecasting from the improvement rate removes that blind spot: MIT research shows each technology's rate is stable over decades, and the faster improver has historically won.
How do you convince sceptical technical experts?
Show the method on a decision they already know the outcome of. A backtest using only the data available at the time is harder to dismiss than any forecast.
