The more time I spend running experiments with #AI to generate qualitative customer and consumer research, the more I’ve begun to realize that adhering to the rules and consistency of output should be the standard. Certainly, there are times when rules might be broken to accommodate the whims and biases of stakeholders. Sometimes this applies to the practitioner as well.
Phases of a Job
We’re told that a Job to be Done has eight (8) to nine (9) phases - where resolution applies as an addition to service innovation framework. Here’s a refresher:
Define: in the define phase, we want to know what aspects of getting the job done need to be defined, planned, or assessed by the {{end user}} upfront in order to proceed.
Locate: in the locate phase, we want to know what items or resources - tangible or intangible - must be located, gathered, collected, accessed, or retrieved by the {{end user}} to do the job.
Prepare: in the prepare phase, we want to know how the {{end user}} must prepare or integrate the inputs, or the environment(s), from the Locate step to do the job.
Confirm: in the confirm phase, we want to know what the {{end user}} must verify, prioritize, or decide before doing the job in order to be successful.
Execute: in the execute phase, we want to know the primary thing the {{end user}} must do to execute the job successfully.
Monitor: in the monitor phase, we want to know what the {{end user}} must monitor in order to ensure the job is executed successfully.
Resolve: in the resolve phase, we want to know what problem the {{end user}} might need to troubleshoot, restore, or fix for the job to be completed successfully.
Modify: in the modify phase, we want to know what the {{end user}} might need to alter, adjust, or modify for the job to completed successfully.
Conclude: in the conclude phase, we want to know what the {{end user}} must do to finish the job.
If this looks familiar, it comes straight from the Jobs-to-be-Done Canvas I created a number of years ago. That, of course, comes straight from the approach taught by Strategyn for their Outcome-Driven Innovation methodology. Here’s a link.
The other day I was a challenged by a reader to duplicate the perfection of a Job Map developed by an expert. I don’t have the entire map, but the beginning is the most important part.
Job: Remove an anatomical structure surgically
Executor: Surgeon
Here’s a snapshot:
Step 1: Establish Initial Access?
This is clearly an Execute step. What happened to Define, Locate, Prepare, and Confirm? They seem to have been completely disregarded. Is that what makes the professional more perfect than a rules-based algorithm? If so, does that mean the rules are actually irrelevant and the processing of a human mind - with all of the subjectivity and bias therein - is what defines perfection? 🤔
If you’re a client stakeholder who is trying to validate your current assumptions then that may seem like perfection.
How many of you feel the need to validate an executive’s idea as a part of your work…versus the opposite? 🦛
But, does the best value-targeting process for innovation come from the bias of humans? I thought we were trying to rid ourselves of that. Is there a difference between procedural jobs vs. core functional jobs or even consumption jobs? I don’t have the answer to that. This simply seems subjective to me.
How can I emulate this?
My prompts are continually evolving. First goal was getting the rules I was trained in down pat. Beyond that, I’ve had to deal with other issues I’ve stumbled across. The starting and ending points of a job map are one of those issues. While this scoping would break the rules, I’ve decided to incorporate it - a beginning point and an ending point. If I don’t enter those variables, my prompt simply follows the standard phasing rules.
The other thing I’ve noticed is fidelity. For example, between the starting and ending point, how many steps (details) do I want? So far, I have not taken into account that quite often a practitioner is forced to conflate multiple jobs into one. So, many of you may have seen examples of jobs in Webinars that had well over 20 steps; sometimes approaching 30.
If you’re following the rules, there are no jobs - unless you are reciting procedures (which are not jobs) that require that many steps. If you are conflating multiple jobs together - such as multiple customer journeys - I would fully expect to see that. In fact, I’ve done that!
I would need to dramatically change my prompts, and the rules that I’ve been trained on to get AI to output a conflated job. The only reason to do that would be to reduce research costs. But how good would the research be?
Emulating the rule-breaking
I used the exact same job statement and executor as above. I also gave it a starting point of Make the Incision. This is because I generated a job map with no scope imposed and that was the first execute step. Since that’s how the perfect map started, I figure I’d need to do the same. I generated 20 success metrics, and here’s how they came out:
Make Initial Incision - The ability to accurately make the first surgical cut at the pre-determined site, following the surgical plan and adhering to best practices to minimize risk and complications.
Minimize the time it takes to position the scalpel for optimal incision, e.g., angle, depth, etc.
Minimize the time it takes to disinfect the incision site, e.g., using antiseptics, sterilizing equipment, etc.
Minimize the time it takes to achieve optimal visibility at the incision site, e.g., lighting, positioning, etc.
Minimize the time it takes to administer local anesthesia, e.g., injection time, dose calculation, etc.
Minimize the time it takes to confirm the location for the incision, e.g., anatomical landmarks, preoperative markings, etc.
Minimize the time it takes to calibrate any surgical technology used during incision, e.g., electrosurgical units, laser systems, etc.
Minimize the time it takes to verify that surgical instruments are ready, e.g., scalpel sharpness, scissor action, etc.
Minimize the time it takes to inform team members of the upcoming incision, e.g., anesthesia, surgical assistants, etc.
Minimize the time it takes to adjust any draping or shields to expose the incision site, e.g., retractors, towels, etc.
Minimize the time it takes to measure the length and depth of the planned incision, e.g., using a ruler, depth gauge, etc.
Minimize the likelihood that surgical drapes impede incision, e.g., slipping, bunching, etc.
Minimize the likelihood of cutting in the wrong location, e.g., misidentification, shifting, etc.
Minimize the likelihood that anesthesia fails to numb the incision site effectively, e.g., incomplete injection, incorrect dosage, etc.
Minimize the likelihood of uncontrolled bleeding during the incision, e.g., arterial spurt, extensive capillary bleeding, etc.
Minimize the likelihood that poor visibility compromises the incision, e.g., inadequate lighting, obscured sight lines, etc.
Minimize the likelihood that surgical instruments fail during the incision, e.g., scalpel breaking, equipment malfunction, etc.
Minimize the likelihood that external factors disturb the incision process, e.g., surgical team movement, equipment interference, etc.
Minimize the time it takes to confirm depth of incision, e.g., visual inspection, depth gauge, etc.
Minimize the time it takes to evaluate the quality of the initial incision, e.g., straightness, depth consistency, etc.
Minimize the time it takes to prepare for the next surgical step post-incision, e.g., switching instruments, informing the team, etc.
When you compare the two, the AI version seems more complete, and less directed. You should also note that 20 is not arbitrary, it was an instruction. I could have generated 50. While I can never guarantee the same verbiage and structure as a human would identify, I can guarantee you that no two humans would create the same map. That’s a fact..
Note: there were also some references to cancer terms in outcomes across the steps, which makes the generic Job Statement useless as an input to AI. Humans can simply state a job, then create a loosely correlated map, and then outcomes related to a more specific job and context. AI can’t. More on that below. 👇
Inputs to a Job Map
It’s easy to state jobs like Remove an anatomical structure surgically and Get to a Destination on Time or Extracting oil from an unconventional reservoir. But I’d like to point out that when you look it any of these job maps, you’ll see specific steps and outcomes that clearly suggest a much tight scope and sometimes directional information about how it is to be accomplished.
We can consider some of these how’s as context, I suppose. But it’s clear from these presentations that no context was supplied. There a few examples that imply a context that is not defined as a part of the job:
Extract Oil from an Unconventional Reservoir - the first three steps in this job map suggest that job has a context that was not disclosed:
Determine how to free hydrocarbons from the formation - the terms have suddenly been changed
Define the fracture plane - this seems somewhat directional. Why do you need a fracture plane? It’s possible that there is scope that was not disclosed.
Determine how to create the fractures - ditto
Removing an Anatomical Structure Surgically - as was pointed out to me the job may have been closer to surgeons who need to remove a suspected cancerous structure surgically “because their outcomes literally mention metastases, cancerous tissue, etc. And the whole job map follows the logic of a tumor excision procedure.”
Get to a Destination on Time - there are several with this one. First is that it conflates multiple jobs together. But more importantly the job statement never mentions the use of an automobile, yet we see the automobile throughout.
Determine how much time to allow for vehicle preparation - what vehicle? Who said anything about a vehicle?
Decide whether or not to make the drive - who said anything about driving?
Walk to the vehicle - again, what vehicle? But more importantly this implies implies a destination.
Prepare the vehicle for the drive - ditto
DRIVE TO THE DESTINATION - I thought the job was get to the destination 🚨🚨🚨
Park the vehicle - This clearly is in the context of driving an automobile, and could even be considered a separate job.
Walk to the destination - this is the second destination, excluding any stops along the way. It also implies how. What if your circumstance was as a disabled person who needed to transport a wheelchair?
I’m not suggesting that these maps cannot point to new innovations, or innovative approaches. I’m suggesting that they are not following the rules, are not clear in their scope, and they may have imperfections at some level. There’s clearly a lot of bias. Therefore I have to ask, how do you define perfection?
Also: what innovations did these studies lead to? I’ll concede that “Get to a destination on time” - which was done before Google purchased Waze generated data that demonstrated quite clearly why purchasing Waze made a lot of sense. But this was not Google research, they made that purchase decision on their own.
In Closing
Who gets to decide what’s perfect? Who get’s to decide what’s good enough?
Perfection often leads to overserving, and sometimes self-serving. For those of you who are students of Jobs-to-be-Done Theory, you should know that disruption comes from below. In this case, good enough is better on several dimensions.
Scope and context do matter. But those customers / clients are solving very specific problems that the rest of the market - companies who are competing for growth - probably don’t care about. In fact, I would contend that all they need is simple directional insights that ensure they are less likely to follow the ideas of their founders - who are notorious for being dead wrong.
“It’s better to be approximately right, than precisely wrong.” h/t Eric Eskey
Execute
Locate
Define
The first three steps in the surgical job. I had to go back and double check
Awesome piece! Thanks!