How to Track Guidance Changes Across a Coverage List With AI
A source-first guidance tracking method with a normalized schema, quarter-over-quarter diff rules, a TJX worked example, and review controls.
Published August 28, 2026 · Updated August 30, 2026
In this article
Track guidance as versioned records, one row per company, metric, period, and accounting basis. Store the exact range, qualifiers, assumptions, source passage, and publication time. AI can locate language and draft a delta, but the comparison should run only after a rule confirms that the periods, units, definitions, and bases match. A changed assumption can matter more than an unchanged headline range.
This is a public-source workflow checked August 30, 2026. The worked example uses two TJX earnings-release exhibits filed with the SEC. It is an editorial analysis of public documents, not an AllMind product output. AllMind combines Grids and agents with built-in guidance and its revision history, model, actuals, consensus, filing and transcript data that can support this process, so the distinction between the worked example and the product is disclosed.
A guidance record needs more than a midpoint
| Field | Why it is required |
|---|---|
| Company and ticker | Entity key |
| Publication timestamp | Establishes which version was available |
| Guidance period | Quarter, full year, or long term |
| Metric and company label | Maps changing vocabulary |
| Lower and upper bound | Preserves a range |
| Unit | USD, percent, basis points, units, or other |
| Accounting basis | GAAP, adjusted, organic, constant currency, or other |
| Scope | Consolidated, segment, geography, product |
| Assumptions and exclusions | Captures what makes the range comparable |
| Status | Introduced, maintained, raised, lowered, widened, narrowed, withdrawn |
| Exact passage and source URL | Supports review |
| Reviewer and review time | Owns the judgment |
Keep the original words in addition to normalized values. “Continues to expect” and “increases” are useful classifications, but a reviewer should be able to inspect the source sentence.
Compare only compatible records
Before calculating a delta, require all of these to match:
- same fiscal period;
- same metric definition;
- same unit and currency;
- same accounting basis;
- same company or segment scope;
- same treatment of material assumptions and exclusions.
If one item differs, label the row basis changed and show both records without a numeric delta. Do not force the comparison by dropping an adjustment.
For compatible ranges, calculate several changes:
| Change | Calculation in ordinary terms |
|---|---|
| Lower bound | Subtract the prior lower bound from the current lower bound |
| Upper bound | Subtract the prior upper bound from the current upper bound |
| Midpoint | Compare the midpoint of the current range with the midpoint of the prior range |
| Range width | Compare the distance between the current bounds with the distance between the prior bounds |
Use code or a spreadsheet for these calculations. The language model should supply candidate inputs and source passages, with a person approving the mapping.
Worked example: TJX full-year fiscal 2027 guidance
TJX's first-quarter fiscal 2027 release, filed May 20, 2026, gave full-year comparable sales growth of 3% to 4%, pretax margin of 11.9% to 12.0%, diluted EPS of $5.08 to $5.15, and share repurchases of $2.75 billion to $3.0 billion. The release also said the company was not flowing through all of the first-quarter upside and assumed higher fuel cost for the rest of the year.
TJX's second-quarter fiscal 2027 release, filed August 19, 2026, maintained comparable sales growth at 3% to 4% and the repurchase expectation at approximately $2.75 billion to $3.0 billion. It increased headline pretax margin guidance to 12.3% to 12.4% and headline diluted EPS guidance to $5.31 to $5.36. The same release introduced expected tariff-refund effects and provided adjusted ranges excluding those benefits: 12.0% to 12.1% for pretax margin and $5.15 to $5.20 for diluted EPS.
| Metric | Q1 FY27 release | Q2 FY27 release | Defensible classification |
|---|---|---|---|
| Consolidated comparable sales growth | 3% to 4% | 3% to 4% | Maintained |
| Pretax margin, headline | 11.9% to 12.0% | 12.3% to 12.4% | Raised 0.4 percentage points at both bounds |
| Pretax margin, excluding expected tariff-refund benefit | Not separately stated | 12.0% to 12.1% | New adjusted basis; no clean prior-range delta |
| Diluted EPS, headline | $5.08 to $5.15 | $5.31 to $5.36 | Raised; midpoint increased by $0.22 |
| Diluted EPS, excluding expected tariff-refund benefit | Not separately stated | $5.15 to $5.20 | New adjusted basis; no clean prior-range delta |
| Share repurchases | $2.75B to $3.0B | Approximately $2.75B to $3.0B | Maintained, with approximation qualifier |
| Key assumption | Higher fuel cost for remainder of year | Tariff refunds and related compensation expense; amount and timing uncertain | Assumption set changed |
The example shows why one “guidance raised” label is incomplete. Headline EPS moved, comparable-sales guidance did not, and the new tariff-refund presentation changed the basis needed to interpret profitability. The adjusted Q2 ranges should not be compared as if TJX had published matching adjusted ranges in the Q1 release.
The $0.22 headline EPS midpoint change is calculated from the displayed ranges: $5.335 minus $5.115. It is an arithmetic description of company guidance, not an estimate or recommendation.
Scale the method across a coverage list
Create one row per company and guidance metric, then run these stages:
1. Ingest and version the source
Capture the release, filing exhibit, investor presentation, and transcript with publication times. The SEC's EDGAR APIs expose company submission history and XBRL facts. Preserve accession numbers and corrected versions.
2. Extract candidate guidance
Ask for exact guidance passages, not summaries. Require company, event, fiscal period, metric, range, unit, basis, scope, assumptions, and speaker. Mark items stated only in an analyst's question.
3. Normalize labels
Map company terms to stable internal IDs. Keep the original label. A change from “organic sales” to “comparable sales” may be a vocabulary update or a definition change; it needs review.
4. Run compatibility rules
Calculate deltas only for compatible records. Route basis changes, withdrawals, new metrics, and missing sources to an exception queue.
5. Review material changes
The covering analyst reads the exact passage and approves the status. Record the analyst's reason when the automated classification changes.
6. Publish a compact coverage grid
| Company | Metric | Prior range | Current range | Basis | Status | Assumption change | Source | Reviewed |
|---|---|---|---|---|---|---|---|---|
| TJX | Pretax margin, headline | 11.9% to 12.0% | 12.3% to 12.4% | Reported headline guidance | Raised at both bounds | Q2 introduced tariff-refund effects and separate adjusted guidance | Q1 and Q2 fiscal 2027 releases above | Yes |
Sort by material, analyst-approved changes. Do not sort by model confidence.
Detect changes in wording as a separate layer
Some companies keep a headline range while changing an assumption. Store and compare sentences about tariffs, currency, volume, pricing, mix, input costs, capital allocation, and macro conditions. Ask the model to highlight additions and removals, with both passages visible.
A wording change is a review trigger, not proof of a future outcome. The analyst decides whether it affects the thesis or model.
Tool requirements for this workflow
AlphaSense Transcript Summaries exposes a guidance-and-outlook section with links to transcript passages. FactSet Transcript Assistant supports transcript questions and summaries within its Workstation. Quartr AI Chat searches first-party IR documents. These can support one or more stages, but their public pages do not establish a shared guidance-diff benchmark.
AllMind's Grids can apply a fixed schema across a company list and keep a cited cell per result. The surrounding data layer is not limited to uploaded documents: AllMind spans 750M+ documents and 6,800+ premium data sources licensed from 100+ providers and partners. Data Viewer exposes company financials, LSEG I/B/E/S estimates, FactSet fundamentals, live prices, and filings. Authenticated research sources also include company KPIs and segment detail, company guidance with its revision history alongside actuals, consensus and analyst estimates from LSEG I/B/E/S, broker models through Visible Alpha, downloadable models, and earnings-history workbooks. Agent Studio can rerun the coverage task on a schedule or monitor. Public pages do not prove field accuracy, historical coverage, or correct treatment of changed bases, so use the TJX example plus several harder names as an answer-key trial.
Acceptance tests
- one unchanged range with a changed assumption;
- one raised lower bound and unchanged upper bound;
- one withdrawn metric;
- one new adjusted basis;
- one fiscal-year mismatch;
- one corrected release;
- one value stated in Q&A but absent from the release;
- one company with no guidance;
- one restricted source unavailable to the output recipient.
The system should abstain or route to review when comparison rules fail. Every exported status should retain both source passages and the approved calculation.
Worked-example sources and limits
The worked example is based on TJX exhibits filed with the SEC on May 20 and August 19, 2026, opened August 30. Calculations use the published ranges shown above and are shown explicitly. Product descriptions come from official AlphaSense, FactSet, Quartr, and AllMind pages. We did not run those products on the example. Contract-specific coverage, trigger timing, corrected-document handling, field accuracy, citations, and entitlements remain unverified.