Build One Research Asset That Feeds a Quarter of Demand

The method comes first and the judgment is yours. Free tools give a first draft of parts of it, as the table alongside shows. You check each output. None of them picks the question, knows your buyer or tells you whether the quarter worked.

The skill's golden rule is the one I would put on the wall: "Every number in the asset traces to its own source checked in this session" and "is written the same way in every asset built from it".

My own example of one research base carrying several outputs is the India AI series on Offbeat AI Watch, 15 pages dated 16 January to 21 February 2026, with a LinkedIn article of 22 January 2026 on the Google/Ipsos survey: "So I went looking for the data." Its method page combines four sources, because "No single platform captures the full picture of a country's AI adoption". Step 2 asks the same of you: find out who measured a figure. That series is published analysis, not a demand case.

Step 1. Choose the objection the research must answer

About this example. FinBox is a real company. FinBox has not reviewed this guide. Facts about FinBox come from public sources, linked alongside with the date each was checked. Every number, product line, strength, rival, objection and offer marked illustrative is an input chosen to show how the tools work; none is FinBox's data or a statement of what FinBox sells, and the tool outputs are exercise results, not FinBox's real strategy. This is a worked example on public and illustrative inputs, not a client result.

What is public: FinBox is a Bengaluru company that offers "B2B credit infrastructure and risk intelligence solutions for banks, NBFCs, and fintechs". It raised a $40 million Series B in September 2025. In May 2026 it published "State of Digital Lending 2026", described on its page as a "21-chapter, 180-page report" across eight credit categories, written for Chief Risk Officers, Heads of Digital Lending and credit policy teams.

If you ran go-to-market at a company like FinBox, you would hold the file and the analyst. This exercise works from outside, on the public page; where an owner would do more, the text says so. Input to competitive intel, all illustrative and written for this exercise, not FinBox's: the product, "credit infrastructure for banks and NBFCs: a digital lending platform, a business rules engine and data APIs"; three rivals by category, "Legacy loan origination system vendor", "Credit bureau analytics add-on" and "In-house build"; the three objections alongside; SaaS as the model, fintech as the industry and sector. On your own company, add strengths, weaknesses, wins and losses from your CRM.

What came back: one battle card per category, and an approach for each objection. For "our delinquency looks fine so our book is fine" the approach was "a back test or pilot on the buyer's own applications".

How to read it: the tool repeats the objections you type and adds three sector ones. Its approaches are general, and one does not fit, so none of them picks the question. That pick is yours: tag recent lost deals by reason and start from the objection that cost most. Here the report came first, so the objection was matched to it; when a team commissions research, the objection comes first and sets the question.

Then test the match, because this is where the chain can break. The page files the rise in deeper buckets under "Old delinquency, not new" and calls it "the same borrowers sliding deeper into arrears". That speaks to the book a lender already holds, so it can answer "our book is fine". It cannot answer "our rules are fine", which is about decisions on new loans.

So the back test does not follow from this finding. The same line appears lower on the card as the sector's general proof point, and the tool's note says this answer "gives a pattern of answer instead of a fact". It tests new applications; the finding is about existing borrowers. The offer in this plan is mine, illustrative and not a FinBox service: a review of how the lender's existing book rolls into deeper buckets, by vintage.

Step 2. Check every number against the publisher, then lock it

Input to the evidence research asset skill: turn the report into a quarter of assets (a LinkedIn series, a one-page brief, a webinar invite) for Chief Risk Officers and Heads of Digital Lending at mid-sized NBFCs; the myth, "our delinquency looks fine, so our book is fine"; the boundary, the publisher's report page plus web search; and an illustrative rep's draft to check.

What came back: a scope block, a topline check, a lock sheet, a beat sheet and three draft posts. The myth is the objection from step 1. The truth is the page's line, "Top-line delinquency rates look broadly contained, but deeper buckets are expanding", with 91 to 360 day card overdue balances up 44% year on year. Dominance is personal loans under Rs 1 lakh at 89.3% of FY25 origination volume. The twist is micro-LAP under Rs 25 lakh at 3 to 3.5% delinquency against seasoned pools below 1%. The verdict, with no new number and as an exercise reading, not FinBox's view: read how existing borrowers roll into deeper buckets, by vintage, before trusting the top-line rate.

How to read it: the topline check is where the value is. The rep's draft failed on four counts, in the table alongside. One was a label: the page files the 44% under "Old delinquency, not new", so "defaults" says something it does not. Two were wrong digits. The fourth rounded 89.3% and swapped "FY25 origination volume" for "personal loans", which turns a share of one year's new loans into a statement about all personal loans. The page also never says whether volume means count or value, so keep the publisher's wording until the file settles it.

The check stops one step short of what a risk reader asks: whose data, and which month? The page names no bureau, regulator or dataset behind these figures; the file may. So lock the publisher's wording, then trace each figure to where it was measured. The 44% is growth in overdue balances, not a rate, and the page gives no card book figure to set it against. If the book grew, part of that rise is growth. And not every NBFC runs a card book, so ask each lender about the same bucket in a product it holds.

The file sits behind an email form, which was not submitted, so every figure was locked from the report page without page numbers; the file may state what the page leaves out. On that basis the sheet marks all five market figures directional: 3 to 3.5%, below 1% and 90% to 86.8% because the page gives no base or period, 44% because the year is not given, and 89.3% because of the volume question. The myth stays an illustration until sales confirms the words buyers use.

If you own the report, lock from the file: page number, base, period and original source on every row. One person owns the sheet, and a risk or data lead checks each row. A row that clears all four drops the directional label.

Step 3. Run the number lock script and fix what it finds

Input: the lock sheet as a CSV and four drafts, the three posts and the rep's draft, through the number lock script.

What came back: the first full run gave "Result: problems found", with the catches in the table alongside. After those fixes, the second run, on the posts and the corrected rep's draft, returned "every number is on the lock sheet and every weekday is right".

How to read it: a pass is narrow. The skill warns that "The script matches values, not bases". The rep's "almost 90%" matched the two-wheeler row, also 90%, so only the topline check caught it. The script reads digits, so check numbers written as words by eye. Reps will not run a script: marketing runs it and gives sales approved sentences (step 5).

Step 4. Build the series from locked numbers only

Input to the thought leadership series in CRAFT Content, all my wording for the exercise, not FinBox's view. The topic: "Reading credit risk below the top-line delinquency rate in Indian digital lending". The take: "A contained top-line delinquency rate does not show that your existing book is healthy; read how existing borrowers roll into deeper overdue buckets, by vintage, before you sign off the quarter". The reader: the step 2 reader, ending "in India". Four proof points, each starting "State of Digital Lending 2026 report page:", then the page's delinquency line; "91 to 360 day card overdue balances rose 44% year on year"; "sub-Rs 1 lakh personal loans were 89.3% of FY25 origination volume"; "micro-LAP under Rs 25 lakh at 3 to 3.5% delinquency while seasoned pools stay below 1%". Four articles in the contrarian style; fintech as the sector.

What came back: four articles of about 316 to 604 words, a post for each, a publishing order and sector notes. Every proof point came back as typed. Through the script the series gives "problems found" with 37 flags: 33 small structural numbers such as "Part 2 of 4", and four draft-length notes such as "about 604 words". Read each flag. None of these is a claim, and every other number it read matched the sheet.

How to read it: treat these as outlines. They set your take and figures inside a fixed frame, so someone who knows the buyer still has to write the argument. Budget four rewrites, not four edits. The tool labels every proof point a result, and articles 1 and 4 say the report's finding "shows that the idea has held up for someone in a real setting". A market statistic is not a result of your idea, so rewrite those lines, as in the edit of article 2's similar line alongside. Cut the lines on "a back test or pilot on your own applications" too: that is the sector line from step 1. The objections, such as "Can it explain a decline?", are what a buyer asks a vendor. Move them to the webinar and the sales follow-up.

Step 5. Lay the quarter out week by week

This step is my suggested plan, illustrative, not a tool output and not FinBox's. Before week 1, write down the named accounts, one owner, the cost of the writing, webinar and sales time, one measure per stage (named accounts reading, attending, replying, agreeing a review), and a stop rule. If at week 6 too few named accounts have shown up, by your own threshold, stop the remaining articles and move the time to named invitations. With a short list, a closed-door session may do more than reach.

The table alongside sets the weeks. Re-lock from the file first. Send the webinar invite three weeks ahead. Use the vendor-style objections from step 4 as webinar questions. Follow up with attendees at named accounts. A lender who wants more gets the review from step 1. That is a later stage: it needs a sponsor at the lender and agreed data terms first.

Two rules. A figure changes on the lock sheet first, then in every asset, and the script runs again. And a directional figure may open a question but never carries a slide headline, a target or a sales claim: say "the report page says", name what is missing, such as "period not stated", and never turn two directional figures into a multiple.

That rule sets a gate. On this sheet all five market figures are directional, so sales could claim only the report's size, and the twist could not headline a webinar. A figure confirmed from the file, with base, period and source, drops the label and may carry a claim. A figure that stays directional stays a question, in an approved sentence such as: "The report page says 91 to 360 day card overdue balances rose 44% year on year, without giving the year. How does that bucket look in your book?" If nothing clears the gate, the webinar title is a question and sales claims nothing. Do not ask attendees for their own bucket out loud; use an anonymous poll or a private benchmark.

This is the point where I would want a second pair of eyes: a quarter of demand has no agreed buyer objection or follow-up owner. Bring me your research brief, locked claims, channel plan and sales handoff, and we will work it through in 45 minutes.

1. Ask sales for the objections they hear most, and tag lost deals by reason.
2. Pick the one objection your research can answer in public, and check the finding answers it.
3. Name the target accounts, the owner, one measure per stage and the stop rule.
4. Run the evidence research asset skill on your report, then lock from the full file and trace each figure.
5. Run the number lock script on every draft and read every flag.
6. Feed only locked figures into the series, write the argument, and run the script on the result.

1. Choosing the research question in the content team. Start from the objection that stops deals, and check the finding answers it.
2. Taking figures from coverage of the report. Lock the publisher's page or file, then trace each figure to where it was measured.
3. Changing the label to make a number sound bigger. Keep the publisher's label and its own reading of the figure.
4. Trusting a script to catch a wrong base. It matches values; the topline check matches meaning.
5. Letting the series present market data as a customer result. Write the argument before it ships.

• How to Turn One Published Customer Result Into a Case Study, a LinkedIn Post and a Newsletter Without Changing the Number
• Auditing a B2B Page for AI Answer Engines: FITq, RACE and a BLUF Rewrite, Run on Our Own Page
• Webinars and Events That Create Pipeline, Not Just Attendees
• All 25 guides