Step 11
Step 11: The Loop
What this step is for
A score is only worth taking if it changes what you do next. The Loop is where it does.
Every step so far has made a plan: the moments to win, the builds that lead them, the campaigns that bundle them. The Loop asks whether any of it worked, and changes the plan when it didn't. It answers one question: what changed, and what do we change?
The book sets out the whole method as six steps: measure, diagnose, write down the truth, prove it, publish it, measure again. Steps 1 to 10 of the Brand Journey are the middle four, done properly. This step is the first and the last: measuring, the same way every time, and recording what moved against what you expected to move.
Most teams run the first step and stop. Or they jump from a disappointing number straight to making more content. The Loop exists to stop both. It ties every fix to a prediction, and every prediction to a result, so the next quarter's plan is built on what actually happened.
What it reads from the chain
The Loop reads everything, and is the only step allowed to send changes back up the chain.
| From | It takes | What it can send back |
|---|---|---|
| The map and the Score (step 5) | The baseline: every moment's score, its range, and where it's lost | A new moment; a moment split or merged; a question reworded (which starts a new baseline) |
| GTM strategy (step 6) | Which moments to win first, next, or hold | A moment moved between tiers, with the evidence |
| Credentials (step 2) | Every claim and its proof | A claim that buyers or assistants keep getting wrong, to be fixed at the source |
| The fourteen builds | Each build's "Loop for this build" table and first-year targets | A job dropped, added or reshaped |
| The Campaign Engine (step 10) | Every campaign brief, with its measures filled in | A campaign repeated, reshaped or dropped |
| The deal record | Fit calls, proposals, wins and losses, and why | The mode weights and the moment weights, checked against real sales |
| New clients | What was happening when they first thought of calling, and where they'd heard of Kestrel | The moments buyers actually name |
Three clocks: monthly, quarterly, yearly
| Clock | What happens | Who | How long |
|---|---|---|---|
| Monthly | The machine scan: the same 101 questions, the same six assistants, three runs each. Corrections where an assistant misquotes a fact. New clients' answers logged | Being named's owner | Half a day to read |
| Quarterly | The review: the full rescore (machines, people and places, the deal record); each build's Loop table; each campaign's measures; the levers re-ranked; the next quarter's top three fixes chosen and predicted | The Score's owner, with every build owner | Half a day, prepared over a week |
| Yearly | The map refreshed: moments added, split or retired; the question set reviewed (a new baseline if it changes); weights checked against the deal record; the next year's campaign calendar | The Score's owner and the managing partner | Two days |
The book draws the quarter as four boxes in a circle: ask (week 1: re-run the same questions), fix (weeks 2 to 11: ship the top three fixes), ask again (week 12: the same sample, unchanged), and record (predicted against measured; drop what did nothing). Record is the step most teams skip, and the one this step exists for.
Step 1: measure again, unchanged
Run the same three instruments as the baseline in step 5, the same way.
| Instrument | Moments | What's asked | Unchanged means |
|---|---|---|---|
| Asked of machines | 61 | 101 questions, six assistants, three runs each: 1,818 answers | The same questions, word for word; the same assistants; the same number of runs |
| People and places | 38 | 114 searches and 40 places (directories, associations, review platforms, trade titles) | The same searches and places, checked the same way |
| Inside a deal | 50 | Read from the deal record: fit calls, proposals, wins and losses, and what buyers said | The same moments, read from the same fields |
Three rules keep the comparison fair:
Change nothing in the sample mid-year. A reworded question, a new assistant or a new place starts a new baseline. Make those changes once a year, record them, and rescore the old quarter on the new set if you can.
Read ranges, not just numbers. Kestrel's baseline is 18, with a range of 12 to 27. A quarter that reads 21 hasn't proved anything on its own. A quarter whose range sits wholly above the old score of 18 is starting to.
Read moments, not just the total. The total moves slowly, because most of Kestrel's moments are out in the market and change only when the market notices. Individual moments move faster, and say which fix worked.
Step 2: diagnose, and rank the levers
For every moment that moved, or didn't, find the stage where it's lost and why. Then split the total loss, as step 5 did, and see whether its shape has changed.
At Kestrel's baseline, the Score lost 82 points, and three quarters of them at the gates:
| Stage | Points lost at baseline | What it meant |
|---|---|---|
| Found | 46 | Buyers and assistants who never come across Kestrel at their moment |
| Known | 16 | Choosers who've never heard of it |
| Believed | 5.8 | Claims with no outside voice |
| Answered | 5.1 | Questions no page answers |
| Preferred | 5.0 | No reason a buyer could repeat |
| Bought | 3.4 | Price "on application"; a contact form as the next step |
The levers come from that split: each fix names its moments and stages, and its predicted lift is worked out by rerunning the Score with the fix in place. At the baseline:
| Fix | Predicted lift (range) |
|---|---|
| Website: the pages that answer the trigger, framing and options moments | +4.2 (2 to 6) |
| The finance director's business case kit | +1.5 |
| The proposal: price, scope and the way out, side by side | +0.9 |
| References within a week | +0.9 |
| Procurement's pack | +0.3 |
| All five together | to about 26 (22 to 28) |
Each quarter, the levers are re-ranked from the new loss split. A fix that's landed drops off the list; the gap it closed shifts the shares, and the next biggest lever moves up.
Step 3: predict, fix, record
Pick the top three fixes for the quarter, no more. For each, write down before you start:
- The moments it should move, by name.
- The stage it should move them at.
- The predicted lift, as a range, from rerunning the Score with the fix in place.
- When the effect should show: this quarter's rescore, or later (being named takes a scan or two; Known can take a year).
- The build that owns it, and the date it ships.
At the end of the quarter, put the measured result beside the prediction. Four things can happen:
| Measured against predicted | What it usually means | What to do |
|---|---|---|
| Within or above the range | The fix worked, and the model of why was right | Keep it; look for the same pattern in other moments |
| Below the range, but moving | Partly right, or too early | Keep it; check again next quarter before adding to it |
| Nothing moved | Wrong diagnosis, or it never shipped as planned | Check it shipped. If it did, drop it and re-diagnose |
| Something else moved instead | A side effect, or a campaign, a launch or the market | Note what else changed that quarter before crediting anything |
The record is the asset. After a year, Kestrel knows which kinds of fix move which kinds of moment in its own market: something no benchmark or book can tell it. A fix that did nothing is dropped and recorded, not quietly continued.
Step 4: feed the map
The map in step 5 was the best guess at the time. The builds and campaigns test it every week, and they find things it missed. The Loop is where those findings go back in.
| Where the finding comes from | What it looks like | What the map does |
|---|---|---|
| New clients' answers | Buyers keep naming a situation the map doesn't have | A new moment, as a hypothesis, until interviews confirm it |
| Outbound replies and fit calls | Owners keep asking the same question at the trigger | A new question for an existing moment, or a new moment |
| The scan | Assistants answer a question with a rival the map didn't list | The rival added to the alternatives; the moment's question checked |
| Campaigns | A campaign needed a page the sitemap didn't have (the Website rule) | The moment it found added to the map, or the page given an end date |
| The deal record | Moments that never appear in won deals, however well they score | Weights checked; the moment may matter less than assumed |
| GTM | A "hold" moment confirmed by interviews | Moved to "win next", with a build to lead it |
Every change is logged with its source and its date, and the GTM strategy re-reads the map. A new moment means a new cell, a lead build and, at the next yearly reset, new questions in the sample.
The finished example: Kestrel's third-quarter review, in full
The review note the managing partner circulated at the end of September in Kestrel's first year, after the half-day review with every build owner. The firm, its clients and every figure are invented.
- Kestrel Advisory: the Loop, Q3 review
- 1. The Score
| Baseline (January) | End Q1 | End Q2 | End Q3 | |
|---|---|---|---|---|
| Score | 18 (12 to 27) | 19 (13 to 27) | 21 (15 to 28) | 24 (19 to 30) |
| Points lost at Found | 46 | 44 | 41 | 38 |
| Points lost at Known | 16 | 16 | 16 | 16 |
- The range now sits wholly above the January score of 18, for the first time. Known hasn't moved, as expected: it's read again at the year end, and most of the work that moves it (ads, social, outbound) only started in July.
- 2. This quarter's three fixes, predicted against measured
| Fix | Owner | Moments | Predicted | Measured | Verdict |
|---|---|---|---|---|---|
| Website wave 2: options and comparing pages, by sector | Website | COO options; comparing, by sector | +2.0 (1 to 3) | +1.6 | Within the range. Keep the pattern: answer in the first lines |
| The finance director's business case kit | The pitch | Finance director building the case | +1.5 (0.8 to 2.2) | +0.4 | Below. The board paper section shipped on 24 September, too late for this rescore. Re-measure in Q4 |
| The proposal: price, scope and the way out, side by side | The pitch | Finance director comparing; owner buying | +0.9 (0.4 to 1.4) | +1.1 | Within. Keep |
- 3. Moments worth noting
- COO · options · flat output: 0 to 6. Assistants now name Kestrel in 2 of 18 answers to "buy another line, or fix this one?", citing the "Kestrel vs new equipment" page and the benchmark's coverage.
- Finance director · trigger: 0 to 2. The benchmark is being quoted.
- The mining mix-up: gone from every answer, as being named predicted, within three scans.
- Plant manager · framing: still 0. Ray Hollins's floor-day post went out on 19 September; we don't expect it in the Score before Q1. Two plant managers in live deals have mentioned it already.
- 4. Campaigns
| Campaign | Judged by (from its brief) | Result | Decision |
|---|---|---|---|
| The annual benchmark (May) | Coverage in the three cited trade titles; quoted by assistants on framing questions | Covered by 2 of 3 titles; quoted in 2 of the 7 framing questions | Repeat next May; pitch the third title earlier |
| "Before you buy another line" (Aug–Sep) | Name searches and fit calls, ad areas against holdout; fit calls mentioning equipment; outbound replies | Name searches up about 30% in ad areas, about 8% in holdout areas; 5 fit calls mentioning equipment against 1; 3 replies from 14 letters. Small numbers: a sketch, not proof | Repeat next year, in the second year's form, once Hartwell's results are confirmed |
| Retargeting holdout | Any difference, on against off | None measurable | Stays off |
- 5. Changes to the map
- New question, held: four owners replying to outbound letters asked whether a government grant could pay for a review. Added as a hypothesis to owner · options: "Can a grant pay for it?" Held until three interviews confirm it.
- New alternative: for the ramp-up question, two assistants named a large firm's "lean in 90 days" programme. Added to the alternatives; the comparison page goes on wave 3's list.
- Dropped from the levers: procurement's pack. It's worth 0.3 at most, and nobody in a deal has asked for it. Recorded, not continued.
- 6. Q4's three fixes
| Fix | Owner | Moments | Predicted | Shows by |
|---|---|---|---|---|
| The board paper, used in every live deal during board season | The pitch | Finance director building the case | +0.8 (0.3 to 1.3) | End Q4 |
| A food and drink sector case on the website and the fact sheet | Being named | Comparing, food and drink | +0.6 (0.2 to 1.0) | End Q4 or Q1 |
| The owners' page, and the first round table | Events | Owner options and checking | +0.5 (0.1 to 1.0) | Q1 |
What to notice. Every fix had a prediction before it started, and the note puts the measured result beside it, including the one that fell short and why. The total is read with its range, and the review says when the range first cleared the old score. The campaign results are called a sketch where the numbers are small. The map gained a question and an alternative from the builds' own findings, and lost a lever that wasn't worth its effort. And the next quarter's three fixes are chosen from the new loss split, each with a range and a date it should show.
Templates
1. The quarterly review agenda (half a day; papers out a week before)
1. The Score: total and range, against the last four quarters; the loss split by stage.
2. Last quarter's three fixes: predicted against measured, with a verdict each.
3. Moments that moved, or should have and didn't.
4. Each build's Loop table: one line per build, from its owner.
5. Campaigns: each brief's "judged by", with results.
6. Changes to the map: new moments, questions, alternatives; moments moved between
- tiers.
7. Next quarter's three fixes: moments, stage, predicted range, owner, date it ships, when
- it should show.
2. The fix log (one row per fix, kept for ever)
| Quarter | Fix | Owner | Moments | Stage | Predicted (range) | Shipped on | Measured | Verdict | What we learned |
|---|---|---|---|---|---|---|---|---|---|
3. The map change log
| Date | Change | Source | Evidence | Status (hypothesis / confirmed) | Effect on the sample |
|---|---|---|---|---|---|
4. The new-client question (asked at every first meeting, and logged)
- "What was happening when you first thought of talking to someone about this?" "Where had you come across Kestrel before you got in touch?" (Offer: a search, an AI assistant, a colleague or client, LinkedIn or YouTube, an event, a letter, an ad, somewhere else.)
Running it
Who owns what. One person owns the Score and the list of fixes, across marketing and sales, with the standing to say no to work that moves no moment. At Kestrel it's the managing partner. They don't own the builds; they decide which fix goes first and which build does it. Each build owner brings their Loop table to the review. Being named's owner runs the monthly scan. The marketing lead keeps the fix log and the map change log.
| When | What happens |
|---|---|
| Monthly | The scan; corrections; new clients' answers logged |
| Week 1 of each quarter | The rescore, on the unchanged sample |
| Week 2 | Papers out: the Score, the loss split, each build's Loop line, each campaign's results |
| Week 3 | The review: half a day, every build owner; the next three fixes chosen and predicted |
| Weeks 3 to 12 | The fixes ship, on the dates their owners gave |
| December | The yearly reset: the map refreshed, the sample reviewed, weights checked against the deal record, next year's campaign calendar |
Three fixes, not ten. A quarter with ten fixes can't tell which one worked. Three, each with a prediction, can.
Common mistakes
| The mistake | What it breaks | The bias it ignores | What it looks like | Instead |
|---|---|---|---|---|
| Measuring and stopping | The whole loop | Good enough beats best | A dashboard everyone looks at and nobody acts on | Every review ends with three fixes, predicted |
| Changing the questions mid-year | The comparison | Familiar feels safe | "The new questions are better", and the trend vanishes | Change the sample once a year, and record it |
| Reading the total, not the range | The decision | Feeling first, reasons after | Celebrating 18 to 21 | Read the range; read the moments |
| No prediction | The learning | Losses loom larger | Fixes chosen, results "interesting" | A range written down before the fix starts |
| Ten fixes at once | The attribution | A hard choice stalls | Everything changed; nobody knows what worked | Three a quarter |
| Crediting the fix for everything | The truth | We follow others | A launch and a fix in the same quarter, one credited | Note what else changed before crediting anything |
| More content as the answer to every gap | The diagnosis | Good enough beats best | A disappointing score, so more blog posts | Find the stage and the cause first |
| Keeping a fix that did nothing | The budget | Losses loom larger | "We've put so much into it" | Drop it, and record why |
The bias to watch in the team: the sunk cost fallacy. The more a team has put into a fix, the harder it is to drop when the rescore says it did nothing. The fix log exists so that dropping a fix is a recorded decision, not an admission.
What the Loop can't tell you
The Score looks forwards, at the choice. The measures a finance director already trusts (attribution, marketing mix models and experiments) look backwards, at sales. The Loop uses both: the Score says which fix to try, and an experiment such as Advertising's regional holdout says whether it paid.
Be clear about the limits:
- The weights are judgement. The mode weights and the moment weights need calibrating against real sales, mode by mode. The yearly reset checks them against the deal record, and that's a start, not a proof.
- Nobody has yet shown that the Score moves with market share. Until that's tested, read it as an index of your odds, not a forecast of sales.
- Some numbers aren't independent. Found and Preferred come from the same sample of answers, so the real margin is wider than the arithmetic says.
- It can't see a sale. Only your own data can. Where you have it, it outranks everything above.
And one measure costs nothing: ask every new client how they found you, and include "an AI assistant" as an answer. Over a year, those answers will tell Kestrel more about which moments matter than any model.
Shapes by business type
| Business type | Typical mode | How the Loop usually runs |
|---|---|---|
| B2B professional services | Committee or owner-led | Quarterly reviews; a slow total and faster moments; the deal record read deal by deal, because there are few of them |
| B2B software | Committee | Quarterly reviews, with product usage and pipeline data beside the Score; holdouts and, with enough years, a mix model |
| Considered consumer (finance, health, education) | Considered personal | Quarterly, timed around the seasons that start the search; the new-customer question matters most |
| Ecommerce | Quick choice or habitual | Monthly on the moments, quarterly on the total; sales data and holdouts carry more weight than for slower categories |
| Local services | Owner-led or considered personal | Five moments by hand each quarter, using the book's self-score worksheet; an afternoon to a day |
| Marketplaces | Varies on each side | A loop for each side, read together: a fix on one side often shows on the other |
Feeling a bit overwhelmed?
You should be, a little. One firm: 1,818 answers a month from the machines, 114 searches and 40 places, the deal record read moment by moment, a loss split by stage, levers re-ranked every quarter, three fixes predicted and recorded, every build's Loop table and every campaign's measures, a map that changes as the evidence comes in, and a yearly reset. Most firms look at a dashboard, feel good or bad about it, and make more content.
RocMode does part of the heavy lifting:
The measuring, the same way each time. RocMode asks your buyers' questions, without your name in them, of several AI assistants, several times each, and records who's named, which rivals appear and which sources are cited. Questions that name your brand don't count towards being found.
A range, not just a number. Each moment's score comes with its range, so you can tell a real move from noise.
What kind of moment, and the biggest lever. Each moment is sorted into won, hidden, borrowed or empty, and RocMode names the fix worth the most points, with the points it should add.
Writing down predictions, choosing the three fixes and running the review are still yours.