Running a pre-mortem on your research synthesis
A synthesis is persuasive by construction. Before it becomes a roadmap, spend an hour assuming it was wrong and asking which sentence did the damage.
The short answer: pick the three or four conclusions your plan depends on, trace each to the sources behind it, write the most plausible story of how each could be wrong, and make the weakest one the first thing you test with real people. It takes about an hour and it is the cheapest risk reduction available to a founder.
A good synthesis is a dangerous document. It takes thirty hours of messy footage and returns a page of confident statements: this audience has this problem, they currently solve it this way, they would value that. Every sentence is fluent. None of them shows how much evidence stands behind it. A claim supported by twenty independent practitioners and one supported by two sponsored videos look identical on the page.
The pre-mortem is the habit that restores the difference.
Find the load-bearing claims
Not every conclusion in a synthesis matters equally. Most are detail. A few are structural: if they are wrong, the product is wrong. The first step is to find those few.
A useful test is to ask, for each conclusion, “if this were false, would I still build the same thing?” If the answer is yes, set it aside. If the answer is no, it is load-bearing. In most plans there are three or four, and they tend to be of the same kinds: that the problem is frequent, that the current solution is painful, that a particular group has it, and that the group can and will pay.
Trace each one to its support
For every load-bearing claim, pull up the sources. This is where citations earn their keep, and where a summary without them becomes impossible to audit.
| Question to ask of each claim | Healthy answer | Warning sign |
|---|---|---|
| How many independent sources support it? | Many, unrelated to one another | One or two, or several from the same channel |
| Who are those sources? | People who do the work | Vendors, affiliates, or commentators |
| Did they show it or say it? | Visible in their actual workflow | Asserted as an opinion |
| How recent is the evidence? | Within the period the market still resembles | Old enough that tools have changed |
| Did any source contradict it? | Contradictions found and explained | None looked for |
The second row catches the most common failure. A claim can be widely repeated because it is true or because several people are paid to repeat it, and the difference is not visible until you look at who is speaking — the sorting method is in vendor content versus independent creators. The fifth row matters because a synthesis that reports no disagreement has usually smoothed it over; handling real disagreement is covered in resolving contradictions in video research.
Write the failure stories
Now assume it is a year from now and the product has failed. For each load-bearing claim, write two or three sentences explaining how that claim turned out to be wrong. The stories should be specific and plausible, not catastrophic.
A few that recur across many projects:
The problem was real but rare. Practitioners complained vividly, and the complaint made good video, but it happened twice a year. Nobody subscribes to a monthly product for a twice-yearly annoyance.
The workaround was good enough. The spreadsheet was ugly and people grumbled, and they had also spent years tuning it. The pain of switching exceeded the pain of staying — a pattern explored in switching costs and why people stay.
The people on camera were not the buyers. Creators were sophisticated freelancers. The paying market was employees at larger firms whose tools are chosen by someone else.
The gap closed while you were building. An existing tool that everyone already uses shipped the missing feature. Judging whether an idea is a product or a feature someone else will add is the test in the feature-or-product test.
The past tense matters. “It might turn out that…” invites reassurance. “It failed because…” invites explanation, and people are much better at explaining an outcome than at predicting one. That shift is the whole trick of the exercise.
Ask what the corpus could not have shown
Some weaknesses are not in the claims but in the sample. A corpus of public video has known blind spots, and a pre-mortem should name the ones that apply.
- ✗People who have the problem and never post about it
- ✗Segments whose work is confidential or regulated
- ✗Purchasing decisions made by someone off camera
- ✗Problems too dull to make a watchable video
- ✓A note on which segments the corpus under-represents
- ✓A few conversations aimed at the missing group
- ✓An explicit open question about who signs the cheque
- ✓A second source type to cover what video skips
The purchasing question is the one video answers worst, and it is best closed by talking to people directly. How the two methods divide the work is set out in video research versus customer interviews, and the wider list of ways an AI reading of video can mislead is in failure modes of AI video research.
Rank the risks and pick one
The output of the exercise is a short ranked list. For each failure story, judge two things: how likely it is given the evidence you just reviewed, and how much damage it would do. A story that is both likely and fatal goes to the top.
Then choose the single top risk and design the smallest test that would tell you whether it is real. If the risk is that the problem is rare, the test is asking ten practitioners when it last happened. If the risk is that the buyer is someone else, the test is finding out who approved the last comparable purchase. If the risk is that the workaround is good enough, the test is asking for a commitment before the product exists.
This is the practical value of the pre-mortem. It converts a general unease about whether the idea is right into one specific question with a specific way to answer it.
Record what would change your mind
Before moving on, write down the result that would cause you to stop or change direction. Deciding this in advance protects against the later temptation to explain away bad news. It also gives a future review something concrete to compare against, which is the basis of checking whether the research held up after launch.
The pre-mortem and the stopping rule fit together. Once the load-bearing claims have been checked and the top risk has a test attached, further broad research adds little, and the right move is to start — the judgment described in when to stop researching and start building.
Using the synthesis against itself
If the research lives in a project you can question, much of the tracing can be done by asking. Useful prompts include: which conclusions rest on the fewest sources, which sources disagreed with the main finding, what proportion of the corpus came from vendors, and what the strongest argument against the recommended direction would be. Each answer should come back with the videos and timestamps that support it, so that you are checking evidence and not taking a second summary on trust.
What this costs
A pre-mortem uses the corpus you already have. As of October 2026, Hobby is $19 a month with 25 videos, 2 projects and 30 chat messages, Pro is $59 with 80 videos, 8 projects and 100 chat messages, and Studio is $199 with 250 videos, 20 projects, 500 chat messages and 3 seats, each with a 7-day free trial — see the pricing page.
Open a synthesised project and ask which conclusions rest on the fewest independent sources. Every answer comes with citations. 7-day free trial.
Closing thought
Confidence is cheap to produce and expensive to be wrong about. An hour spent arguing against your own conclusions will not make the idea safe. It will tell you which part to worry about, and that is usually enough.
Frequently asked
What is a pre-mortem?
An exercise where you assume the project has already failed and work backwards to explain why. It is the reverse of a post-mortem: instead of asking what went wrong afterwards, you ask what is most likely to go wrong before you commit, while changing course is still cheap.
Why run one on a research synthesis specifically?
Because a synthesis reads as more certain than the evidence behind it. A clean summary hides how many sources supported each claim, who those sources were and what was missing. A pre-mortem forces each conclusion to be traced back to its support before it becomes a roadmap.
How long does a research pre-mortem take?
An hour or two for a solo founder, if the synthesis keeps citations attached to its claims. Most of the time goes on checking the three or four conclusions the whole plan depends on, not on reviewing everything.
What if the pre-mortem finds a serious weakness?
Then it has done its job. The usual outcomes are a narrower first segment, a smaller first version, or one targeted piece of extra research. Abandoning the idea entirely is rarer, and when it happens it has saved months.
Can I ask an AI to run the pre-mortem for me?
It is useful as a challenger, particularly for listing failure stories you would not have thought of and for finding which claims rest on few sources. The judgment about which risks matter for your situation, and what you are willing to bet on, stays with you.
How is this different from disqualifying an idea?
Disqualification happens early and asks whether the idea deserves research at all. A pre-mortem happens after research, when you believe the idea is good, and asks whether that belief would survive a sceptical reading of your own evidence.
What does it cost to re-check a synthesis?
As of October 2026, Hobby is $19 a month for 25 videos and 2 projects, Pro is $59 for 80 videos and 8 projects, and Studio is $199 for 250 videos, 20 projects and 3 seats, with a 7-day free trial on every plan. Questioning an existing project uses chat messages, not additional videos.