Sample · Real engine output

Real engine output. A Truth Score Read on test data.

We ran our engine on a generated dataset built to behave like a mid-market B2B software company, Example Co: seven exports from its CRM, finance system and HR system, 72,837 records in all. Every number in the memo below is what the engine reported, unedited. The company is fictional, and nothing here is client data.

Get your Truth Score Read the memo

Starts with a 30-minute call. Commits you to nothing.

Company: Fictional mid-market B2B software company
Exports: 7: CRM accounts, contacts and deals; finance customers, invoices and subscriptions; HR employees
Records: 72,837
Truth Score™: 59 Band C, significant work needed
Verdict: Don’t sign yet
Status: Real engine output on generated test data, not a client.

Why test data?

Because we will never publish a client’s records.

And we have none to publish yet. So we built a dataset with the problems real companies have, ran it through the same engine your exports would go through, and printed the result unedited. On the call, we can walk you through how each number in this memo was produced.

Not a software company? The same method reads distributors, manufacturers, logistics firms and professional services. See the gaps typical in manufacturing, logistics and distribution, professional services, retail, healthcare or financial services.

The memo

One page. The verdict first.

This is the document a CEO and CFO receive, followed by a walk-through call. It carries the Truth Score™, a score out of 100 for how far the data can be trusted, and a Leverability read: which AI ideas you can use now, after fixes, or not at all.

1
The verdict
Sign, fix first, or walk away. One sentence on why, in the first line.
2
Truth Score™
A score out of 100 for how far the data can be trusted, pillar by pillar, with the cap shown.
3
Leverability read
Which AI ideas you can use now, after fixes, or not at all. Here, the one AI in question.
4
Red flags
The gaps no single system shows, counted and labelled with the systems involved.
5
Fix list
In order, with an owner and an effort estimate, ready to hand to your own team.
6
Scope line
What we looked at, and what we didn’t.
7
Sign-off
Checked by hand and signed off by DataTruth, then walked through on a call.

Truth Score Read · Sample · Example CoConfidential · Page 1 of 1

Decision: Proposed AI revenue forecasting tool
Prepared for: CEO and CFO
Compared: CRM, finance and HR exports · 72,837 records

Our verdict on the proposal

Don’t sign yet. Only 13% of deals and invoices reconcile between the CRM and finance, and 823 paying customers have no account in the CRM. A forecast trained on this would learn from deals finance can’t see. Link customers and trace deals to invoices first, then re-test.

Truth Score™

59/ 100Band C

Significant work needed

Accuracy22% 88
Completeness18% 97
Governance18% 69
Consistency16% 88
Connectivity14% 44
Lineagelimited signal12% 50

Weighted average 75. Capped at 59 because Connectivity, a critical pillar, scored 44: strong pillars can’t hide a weak foundation.

Leverability read · the AI in question

AI revenue forecasting

Blocked

Deals tied to invoices in finance13% reconcile
One customer across CRM and finance46% reconcile
Sales activity historyNot in exports
Deal close dates97% valid

Red flags between your systems

High
823 customers in finance with no account in the CRM. A further 2,978 CRM accounts have no finance record; some will be prospects.
CRM ↔ Finance
High
Only 13% of deals and invoices reconcile: 4,964 deals and 21,964 invoices with no match on the other side. A traceability gap, not yet a revenue figure.
CRM ↔ Finance
Medium
3,128 matched customers where the CRM and finance disagree on the city, and 421 where the country is written differently.
CRM ↔ Finance
Medium
1,004 invalid, 845 duplicated and 870 disposable email addresses across 23,116 contacts.
CRM
Medium
521 invoices and 280 deals carry invalid dates.
Finance · CRM

Fix first, in this order

1
Give every customer one ID across the CRM and finance
RevOps and finance · est. 2 to 3 weeks
2
Trace deals to invoices, and agree which deal stages should produce one
Finance and sales ops · est. 1 to 2 weeks
3
Clean contact emails, and standardise city and country, then re-test
Sales ops · est. 1 week

Scope CRM accounts, contacts and deals; finance customers, invoices and subscriptions; HR employees: 72,837 records in 7 exports. Export-based: live configuration, permissions and access logs not reviewed. Lineage scored on limited signal: the exports carry creation dates but no source-system field. Effort estimates are review judgement, not engine output.

Checked by hand · Signed off by DataTruth Ref SAMPLE-01

Sample memo. Real engine output on generated test data for a fictional B2B software company. Not a client, and no client data.

What the engine found

Every system looks clean on its own. The truth is in the gaps between them.

Inside the systems, this data passes: completeness scores 97 and accuracy 88. Then we match the systems against each other. Only 13% of deals tie to an invoice, and 823 customers in finance have no account in the CRM. AI doesn’t read one system. It reads across them.

Inside the systems97
Completeness. Required and key fields are filled.
Inside the systems88
Accuracy. Values are mostly valid and well formed.
Between the systems13%
of deals and invoices reconcile between the CRM and finance.
Between the systems823
paying customers in finance with no account in the CRM.

How the score was reached

Strong pillars can’t hide a weak foundation.

Accuracy, Completeness and Connectivity are the critical pillars. If the weakest of them is below 80, the overall score is capped at 0.55 × that pillar + 35, rounded. Here Connectivity scored 44, so the cap is 59, below the weighted average of 75. Lineage scored 50 on limited signal: the exports carry creation dates but no source-system field, and the missing half counts as a gap, not a guess. The full rule

PillarWeightScore

Accuracy critical22%88

Completeness critical18%97

Governance18%69

Consistency16%88

Connectivity critical, weakest14%44

Lineage limited signal12%50

Weighted average75

Cap: 0.55 × 44 + 35 = 59.2, rounded59

Truth Score, the lower of the two59 · Band C

Leverability read

The same data, read against four AI plays.

A Truth Score Read focuses on the AI in front of you. An AI Investment Review reads every bet the same way. On this data, nothing is ready to deploy today, and the reason is specific to each play.

Revenue forecastingBlocked
Deals and invoices reconcile 13% of the time. The model would learn from revenue finance can’t trace.
Customer churn predictionBlocked
No activity or support-ticket history in the exports, and under half of customers link across systems.
Lead scoringNeeds work
Deal history is there and largely valid. Activity history is missing, and contact emails need cleaning first.
Customer segmentationNeeds work
Records are 97% complete, but customers disagree between systems on where they are.

If a fund commissioned this

The same findings, in an investment committee frame.

For an investor, the memo is addressed to the company’s leadership and the fund together, or shared by management as a board paper. Same pillars, same rules, same caps at every company. The plays below are an example value creation plan laid over the real findings.

IC summary · Example Co · for Example Fund1 page

Hold the AI lines in the plan until two fixes land. Truth Score 59, Band C, significant work needed. Two plays blocked, two after fixes. Customer linking and deal-to-invoice tracing come first, owned by the company.

Plan lineAI playLeverabilityCall

Forecast-led pricing and pipeline review13% of deals tie to an invoice Revenue forecasting Blocked Don’t fund yet

Retention programme on the installed baseNo activity history; under half of customers link Churn prediction Blocked Don’t fund yet

Sales productivityDeal history usable; contact emails need cleaning Lead scoring Needs work Fund after fixes

Segment-level go-to-market3,128 customers disagree on location Customer segmentation Needs work Fund after fixes

We use the targets in your plan. We don’t value the plays ourselves.

Findings: Real engine output on generated test data, not a client. Plan lines and fund name: illustrative example, not a client.

What this sample is

What it is, and what it isn’t.

It is

Real output from the engine your exports would go through
Every number unedited, including the ones that look harsh
The same one-page format a CEO and CFO receive

It isn’t

A client, or any client’s data
A benchmark: the score measures this data against fixed rules
The only answer a memo can give: the next one could say sign

How the score is builtThe pillars, the bands and the cap, in full What happens to your exportsNDA first, exports only, records deleted Every AI bet, read this wayThe AI Investment Review, in about 2 weeks

Want this memo on your own data?

A 30-minute call about the decision in front of you. Your memo arrives within 5 working days of complete exports.

Get your Truth Score

Starts with a 30-minute call. Commits you to nothing.