My Quest for a Lie Detector in Price Negotiations

My Quest for a Lie Detector in Price Negotiations

Every one of us has to negotiate prices, whether it’s when buying a car, negotiating a salary, or selling goods and services to someone else. Now imagine you could tell when the other party is bluffing. For decades, I have been searching for those hints. Quick and dirty analyses of more than 3,500 mini-negotiations left me empty-handed. Only after analyzing more than a thousand verbal statements in detail, I was able to see the forest for the trees: The less bargaining power you have, the less likely you are to lie. More importantly, it seems I found a few telltale signs indicating that you are reaching someone’s real budget limit in a negotiation! 

A few weeks back, a student from my alma mater contacted me to conduct an interview for his bachelor thesis on the art of deception in negotiations. Having been a pricing consultant for decades, I have always been fascinated with this concept. What if I had a lie detector to use in price negotiations? I have been chasing this “holy grail of price negotiations” for years. 

First fail: Outsourcing

At one point, I read about an extensive negotiation experiment at a university where dozens and dozens of identical negotiations had been simulated via a chat software, i.e., all statements were captured in full. I called up the responsible professor and shared my idea with him: “Why don’t you analyze all those transcripts to look for patterns when negotiators approach their secret walk-away price?” The professor liked the idea - at least that’s what he said - but either it was too much work to analyze those thousands of statements or he did not find anything meaningful. In any case, I never heard from him again nor did I see any results published. 

Second fail: A simple experiment

Last spring, I had an epiphany under the morning shower - in fact, two epiphanies: 

  1. If I reduced an everyday negotiation down to the bare necessities and focused just on the one aspect that I cared about (the counter offer which would include the deception), I could field an online survey to thousands of people at a very low cost.
  2. If I kept the seller’s offer constant (i.e., the initial anchor) and exposed different groups of respondents to different budgets (i.e., the hidden walk-away or reservation price) in my experiment, perhaps the difference in responses of those groups would reveal systematic markers of deception. 

After much deliberation, I asked one thousand people just one question: “Suppose you want to buy a used car for yourself. The seller wants 20,000 €. But your budget is only 18,000 €. You want to try to lower the price further. What do you say?”

Then I asked another group of people for their reaction to a lower budget (22,000 €) and to round it off, I asked the third group and gave them a budget of 20,000 € (so exactly what the seller offered). That gave me three groups, let’s call them: 

  • “Budget < offer” (i.e., 18,000 < 20,000)
  • “Budget > offer” (i.e., 22,000 > 20,000)
  • “Budget = offer” (i.e., 20,000 = 20,000)

Only after I had fielded the questions online it occurred to me that coding 2,000 answers would be a horrific effort. So I said to myself: Let’s start with the numerical answers, they are easy to aggregate. About 25%-30% of the answers to that first question had a usable number. 

I expected the “Budget < offer” group to try hardest to bring the price down, since they were on the tightest budget, and consequently they would field the lowest counter offer, i.e., they would try to deceive the most. Conversely, the “Budget > offer” group would be the most lenient and have a much higher counter offer (i.e., deceive the least), and the third group would fall somewhere between the two.

Here’s what I got (on average):

  • “Budget < offer” (i.e., 18,000 < 20,000): 17,069 €
  • “Budget > offer” (i.e., 22,000 > 20,000): 17,308 €
  • “Budget = offer” (i.e., 20,000 = 20,000): 16,277 € 

The two first numbers are basically the same (i.e., the difference is not statistically significant). Damn! Not what I expected. At all. 

Third fail: Another round of even simpler surveys

Had I confused respondents with the budget figure that I had given them as a reference? Perhaps! So instead of doing the hard work and getting to the bottom of this by coding the open-ended answers (don’t ever tell this to my daughters…), I launched three additional surveys, this time with two amendments:

  1. Instead of an actual budget figure, I inserted only a statement on whether the budget is higher/equal/lower to the offer
  2. Instead of an open question, I included seven numerical counter offers to choose from (ranging from 14k to 20k): 

Analyzing the results of this second survey just took a few minutes (after fielding was complete):

  • “Budget < offer” (i.e., 18,000 < 20,000): 16,265 €
  • “Budget > offer” (i.e., 22,000 > 20,000): 16,454 €
  • “Budget = offer” (i.e., 20,000 = 20,000): 16,456 € 

Oh great, now all three numbers were basically the same! Aaaaargh! 

The initial conclusion was obvious: in price negotiations, your counter offer (i.e., the degree of deception) is not influenced by your budget. Or framed as a lie detector insight: The type of counter offer you receive from your opponent does not tell you if you are approaching her reservation price. While this may be a great insight in and of itself, it failed as a lie detector. And so I forgot about those surveys for months and months...until that student contacted me.

Fourth attempt: A closer look at the numbers

The most obvious next step in analyzing the survey results in more detail was - of course - to look at the distribution of numerical responses. Even though that just takes a few more clicks in the pivot table, I simply did not do it last year, don’t ask me why. 

First, I looked at the surveys with the closed answers (again, no data to clean):

No alt text provided for this image

Unfortunately, the one blip at 14,000 Euro, where the “Budget < offer” group seems to be slightly above average, is not quite statistically significant...though it fit my story beautifully. For a moment, I have to admit that I was considering to ditch the significance concerns (which most people don’t really understand anyway, as most Covid-19 analyses show) and hail victory nonetheless. Then professional pride got the better of me and I moved on…

...to the first set of surveys, the ones with the two thousand statements. (In the graph below, you see more price points since answers were in writing and not restricted.) Let’s look at this graph step by step:

No alt text provided for this image

The first result is that groups “Budget = offer” (grey) and “Budget > offer” (green) have a very similar distribution, consistent with the averages reported above. So far, so bad. 

The result that stands out (marked by the magnifying lens) is the pretty significant difference at 18,000 Euros! This means that the “Budget < offer” group (that has to fight the most to bring down the seller’s price) is much more likely to come up with a high counter offer of 18,000 Euro than the other two groups. Conversely, this group is less likely to counter-offer 15,000 Euro (the bottom of the price range given).  

Therefore, the group that has the weakest bargaining power (i.e., the lowest budget with this highest distance to the seller’s offer) is far less likely to deceive the seller with an aggressively low counter offer. This is counter-intuitive because both, science and conventional wisdom suggest that a lower counter offer is more likely to bring down the price of the seller. Or to put it bluntly: The less you have, the less you lie in a price negotiation

But let’s not get too excited just yet: This isolated insight is difficult to operationalize in a negotiation. For good reasons, the science of detecting lies is based on polygraph devices that track multiple indicators. So I knew I had to dig deeper.

Fifth attempt: Finally doing all the legwork

When I revisited the raw data of my old surveys a few days ago, I discovered that I had started to code the qualitative responses but had stopped after the first few dozen. If you have ever coded survey responses, you will probably agree that this is not only more difficult than it looks, but also mind-numbing toil. (By the way, I am not referring to the occasional insult that respondents sneak into the survey - for example, let me leave this little gem here untranslated, I am sure you’ll get the gist: “Ahhh f*ck deine muddaaa” - oh well, the joys of online surveys...).

Most answers fell into the following categories (ranked from the more sophisticated to more brute force):

  1. Highlighting budget restriction (“I only have 15,000 Euros!”)
  2. Highlighting competition (“Give it to me for 18k Euros or I will buy it somewhere else”)
  3. Highlighting defects of the car to lower the price (“This car has several defects.”)
  4. Offering cash payment in return for a discount (“I give you 17,500 in cash.”)
  5. Simply stating that the seller’s demand is too high (“That’s too much!”)
  6. Asking the seller for her final offer (“What is your last price?”)
  7. Simply walking away from the deal (“Sorry, I will buy the car somewhere else.”)

Here are the highlights of what I found when I analyzed the frequency of these categories between the three groups. For clarity, I will focus on comparing just two groups with each other, “Budget < offer” and “Budget = offer”:

No alt text provided for this image

For the “Budget < offer” group (with a weak bargaining position), highlighting the (true) budget restriction is by far the most frequent answer (37%). More importantly, a respondent from the “Budget < offer” group is 58% more likely to use the budget restriction argument than someone from the “Budget = offer” group! This group is also much more likely to just walk away from the deal. On the other hand, people from the “Budget = offer” test group are 110% more likely to ask the seller for her “last price” and 186% more likely to lie to the seller and say that her offer is just too much, even though it is not. 

The online survey provider also captures the time it took the respondent to produce an answer. The average answer time across all three groups was about 

  • 22 seconds for the closed answer surveys (where the respondents just chose one out of seven numbers)
  • 41 seconds for all open text answers (incl. purely numerical answers)
  • 49 seconds for open text answers that I could code as one of the categories above 

All values are quite reasonable. For the closed answer surveys, there was no difference between the three groups. The only thing that stood out was that if people caved (i.e., chose the counter offer in line with the seller’s offer, 20,000), it took them significantly less time to respond: about 14 seconds instead of 22 seconds, with no difference between the three groups. 

There was one marked difference in timing between the three groups, and that was with the people who gave an answer that fell into the categories mentioned above: Both, the “Budget < offer” and the “Budget > offer” group took on average about 56 seconds to respond, whereas the “Budget = offer” group only took 35 seconds! In other words, you are more likely to answer much quicker if the seller’s offer hits your budget, regardless of what your answer will be!

Frank’s Lie Detector For Price Negotiations

What do we make of all this? It’s fair to say that we have penetrated the “fog of war” in price negotiations at least a little bit with the largest-scale quantitative study that I am aware of. Let me summarize the core findings in this simple overview: 

No alt text provided for this image

If your opponent...

  1. ...refuses your offer without giving any other reasons (“just too much”)
  2. ...answers quicker than usual with a response (regardless of what that response is)
  3. ...asks you for your “last price”

your opponent is more likely to deceive you and your offer is actually close to your opponent’s budget target!

If your opponent...

  1. ...refers to a credible budget restriction 
  2. ...credibly threatens to walk away from the deal

your opponent is more likely to be truthful and the gap may still be substantial before you can close the deal.

The data shows a lot more rich differences - and very interesting ones, including gender differences and age differences, but for most of these, the sample size is still too small to be significant. Stay tuned for more!

Appendix, aka the Nerd Section

Thinking of Ronald Coase? “If you torture the data long enough, it will confess”?

Well, the good news is that there was pretty much zero fancy analytical methods used in the analysis of these survey results. That is the beauty of the monadic approach, it is elegantly simple. 

Here is the exact phrasing of all six surveys (including my corresponding internal survey title and number):

Nego_10plus_1: Angenommen, Sie wollen privat einen Gebrauchtwagen kaufen. Der Verk?ufer will 20.000 €. Ihr Budget ist zwar 22.000 €, aber Sie wollen versuchen, den Preis weiter zu drücken. Was sagen Sie?

Nego_10minus_2: Angenommen, Sie wollen privat einen Gebrauchtwagen kaufen. Der Verk?ufer will 20.000 €. Ihr Budget ist aber nur 18.000 €. Sie wollen versuchen, den Preis weiter zu drücken. Was sagen Sie?

Nego_at_3: Angenommen, Sie wollen privat einen Gebrauchtwagen kaufen. Der Verk?ufer will 20.000 €. Ihr Budget ist tats?chlich etwa 20.000 €, aber Sie wollen versuchen, den Preis weiter zu drücken. Was sagen Sie?

Nego_h?her_4: Angenommen, Sie wollen privat einen Gebrauchtwagen kaufen. Der Verk?ufer will 20.000 €. Ihr Budget ist zwar h?her, aber Sie wollen versuchen, den Preis zu drücken. Was w?re Ihr Gegengebot?

Nego_tiefer_5: Angenommen, Sie wollen privat einen Gebrauchtwagen kaufen. Der Verk?ufer will 20.000 €. Ihr Budget ist aber geringer. Sie wollen versuchen, den Preis zu drücken. Was w?re Ihr Gegengebot?

Nego_gleich_6: Angenommen, Sie wollen privat einen Gebrauchtwagen kaufen. Der Verk?ufer will 20.000 €. Ihr Budget ist tats?chlich etwa so hoch, aber Sie wollen versuchen, den Preis zu drücken. Was w?re Ihr Gegengebot?

In surveys no. 1-3, respondents had a maximum of 66 characters to respond. In surveys no. 4-6, respondents had 7 different numbers to choose one answer from (14k-20k). 

Respondents always could skip the question altogether. Fielding was not through an online panel (with the typical panel challenges), but primarily through various online publications where respondents could access articles in return for answering questions. The resulting sample bias (see table below for RMSE Score) was corrected using a weighting for each respondent. 

No alt text provided for this image

As with any monadic survey design, to compare results between the groups, all samples have to have a similar structure, which they did, as all were weighted to be representative for Germany’s online users, i.e., 95% of the population. To check how robust the answers are, I also looked at the unweighted results and they are very much in line with the weighted results. To keep things as simple and as transparent as possible, I then used the unweighted, raw data for all the analyses above.

I cleaned the respondent data in two simple respects: For the timing measurements, I replaced the top and bottom ten outliers with the average time. For the numerical counter offers in surveys 1-3, I excluded all data points below 10k (31 responses out of 2003) and above 20k (9 responses out of 2003).

All results reported above have a p < 0.01, so they are highly significant.

You may have noticed that - oddly enough - the 186% difference occurs twice. I triple checked the numbers and it is what it is (5.6% vs. 16.1% and 9.2% vs. 3.2%)

Selected caveats:

  1. These findings were generated in an experimental setting where there was nothing really at stake for the participants. 
  2. Participants were not screened to have any relevant experience in negotiating prices, but rather to be representative of the overall population aged 18 and above.
  3. As with all negotiation advice, it can become self-defeating if both parties act accordingly. However, in my 25+ years of shadowing and conducting price negotiations, I have seen many smart people (and myself!) fail miserably at applying even simple principles everyone is well aware of. 

A few relevant articles, some with my key insight, for further reading: 

Galinsky, Adam D., and Thomas Mussweiler. "First offers as anchors: the role of perspective-taking and negotiator focus." Journal of personality and social psychology 81.4 (2001): 657.

Kristensen, Henrik, and Tommy G?rling. "Anchor points, reference points, and counteroffers in negotiations." Group decision and negotiation 9.6 (2000): 493-505: “The results showed as expected that the counteroffers were higher for a high than for a low anchor point, and higher for a high reference point when the anchor point was perceived as a gain than for a low reference point when the anchor point was perceived as a loss.” 

Moosmayer, Dirk C., et al. "A neural network approach to predicting price negotiation outcomes in business-to-business contexts." Expert Systems with Applications 40.8 (2013): 3028-3035: “... it showed that target price played a more important role in B2B price negotiations. “

Van Poucke, Dirk, and Marc Buelens. "Predicting the outcome of a two-party price negotiation: Contribution of reservation price, aspiration price and opening offer." Journal of Economic Psychology 23.1 (2002): 67-76: “A total of 384 experienced managers participated in 192 simulated seller–buyer negotiations. More than 57% of the variance in negotiation outcome can be explained by two reference points, namely buyer's and seller's intended initial offer. We introduce the notion of `offer zone', which is the difference between aspiration price and initial offer. Offer zone has a significant and consistent influence on the negotiated outcome. Results from 106 participants in a replication study with a different `no deal' structure confirm our findings.”

White, Sally Blount, et al. "Alternative models of price behavior in dyadic negotiations: Market prices, reservation prices, and negotiator aspirations." Organizational Behavior and Human Decision Processes 57.3 (1994): 430-447: “In two studies varying the levels of these three factors, only reservation prices, not prevailing market prices or negotiator aspirations, account for significant variance in negotiated outcomes.“

Srivastava, Joydeep, Dipankar Chakravarti, and Amnon Rapoport. "Price and margin negotiations in marketing channels: An experimental study of sequential bargaining under one-sided uncertainty and opportunity cost of delay." Marketing Science 19.2 (2000): 163-184: “Rather, these data suggest that the bargainers created simplified representations of the price negotiation and used heuristics to develop their offers and counteroffers. We observe two systematic patterns of deviations from the SE model. Some manufacturers may have used the counteroffer levels to infer the distributors' competitive stance and factored this into their responses. Thus, even though the distributor counteroffers carried signals of the consumer reservation price, the manufacturers delayed agreement because they either did not recognize the signal or thought it was unreliable. In other cases, the data are consistent with a simple, nonstrategic model (EMP) in which the manufacturer and the distributor divide the monetary payoff (surplus) equally. The results show that the effectiveness of signaling mechanisms depends not only on the economic characteristics of the bargaining situation, but also on shared individual and social contexts that influence how signals are transmitted and interpreted.”

Michal Stulgis

Head of CX Team | PKO Bank Polski

4 年

Awesome article, Frank Bilstein! Does the phrase analysis confirm some "I" vs. "you" differentiation? Looks like if you have enough money, you tend to influence or refer to the seller ("your" price is too high, is this "your" last offer). And if you don't have enough funds, you focus on yourself ("my" budget can't stand it, "I'm" gonna walk away). Could it be, or it's oversimplifying?

Fantastic read, thank for all the details!

Ram Subramanian

Shaping, Supporting, and Sustaining Value for the Galderma Therapeutic Dermatology portfolio | Global Vice President | Market Access Executive

4 年

Thanks for sharing Frank Bilstein!

回复
Alessandro. Monti

Professor | Advisor | Coach for Pricing, Sales and Marketing

4 年

Very thoughtful article - and definitely agree with the joy of online surveys..

要查看或添加评论,请登录

社区洞察

其他会员也浏览了