Portfolio / Case study

Case study · Sales technology

A sales-coaching expert had the method.
We built the product that sells it.

The starting point: a clickable prototype, a scoring method proven by hand, and a plan for an 8-week first version. Eight weeks later it was a working AI product on the client’s own accounts.

Sales technology · AI systems · Custom software · Analytics & attribution

Working product

8 weeks

From a clickable mockup with no data behind it to a working AI product: call scoring, pipeline warnings and a weekly briefing.

On the client’s own 8-week plan, on the client’s own accounts.

AI agents

6

Six AI agents released to the live product: call scoring, pipeline inspection, preparation, coaching, one-to-ones and diagnosis.

They hand work to each other: a pipeline warning leads to a diagnosis, which pre-fills the next one-to-one.

Live on your accounts

Week 5

The first version ran on the client’s own cloud and AI accounts from 1 May.

Nothing to migrate later.

Real calls

273

273 of the client’s real sales calls loaded and scored against the client’s own scorecard.

Every score backed by a quote from the call.

Customer onboarding

1 click

Customers connect their own CRM in one click, and the client connected theirs on the live system.

Read-only, with every access key encrypted and every permission recorded.

Pipeline watched

$518,000

Open pipeline checked across 21 deals, and 23 of 23 warnings were correct.

Proven end to end on a purpose-built test account before launch.

Billing protected

5×

Top-tier subscribers get the full usage they pay for. An error would have given them only a fifth of it.

Caught before a single invoice went out.

Client, people, product, tools and vendors withheld. Every figure is traced to the project’s own records: code history, weekly reports and work logs. Where a result was proven on test data rather than the client’s, it says so.

How we got there

Seven results,
each with its cause.

8 weeks

The core product, on the client’s 8-week plan

What held it back

The client had a prototype that looked right but ran on nothing. There was no scoring, no data and no cloud set-up.

What we did

We rebuilt it as a working product on the client’s own accounts. It had AI call scoring, 16 pipeline warning rules across 3 agents, and a weekly manager briefing. We told the client about a one-week slip in week 5 and recovered it by week 6. The core was complete in eight weeks, as planned.

How we write: every week the client got a plain-English report of what shipped, what slipped, and what we needed from them.

6

Six agents, one system

What held it back

  • Every agent existed only as a mockup screen with sample data

What we did

  • Call scoring: every point on the client’s scorecard gets a score, a reason and a quote from the call, and each section is marked WIN or FIX
  • Pipeline inspection: 16 checks across 3 agents on deals, leads and activity, plus custom agents built without code
  • Preparation: 5 briefings, from morning prep to the weekly plan
  • Coaching: one focus area per salesperson, and practice drills graded automatically
  • One-to-ones: an 8-section guided meeting that brings back last week’s commitments
  • Diagnosis: names the likely cause behind a drop across 10 measures, with the evidence and talking points for the one-to-one

How we build: the agents hand work to each other through 6 connections, so one finding becomes the next step instead of another report.

273

Real calls, scored against the client’s own method

What held it back

The client had proven their scorecard, with 12 criteria in 4 sections, by scoring calls by hand.

What we did

We loaded 273 of the client’s real sales calls and scored them in batches against that scorecard, with a fast rules-based method and with AI. Each score comes with quoted evidence from the call: 1,187 quotes in the first round. By 14 July, 84 calls had been scored.

1 click

Customers connect their own CRM

What held it back

Connecting a CRM meant copying an access key and walking through set-up with the customer’s administrator. That does not work for self-serve sign-up.

What we did

We built a one-click connection that only reads, and never writes to the customer’s CRM. Each customer’s access key is stored encrypted, and every permission they grant is recorded. The client approved it on their own account on the live system, and the first sync brought in 25 contacts, 25 companies, 4 salespeople and 31 activities. 35 updates went live the same day.

$518,000

A pipeline that flags its own risks

What held it back

The client’s CRM had no deals in it yet, so 15 of the 16 pipeline warning rules had nothing to check.

What we did

We built a test account with 24 sample deals and synced it end to end. The product found $518,000 of open pipeline across 21 deals. It flagged 9 overdue deals and 10 large deals that had gone quiet, and all 23 of 23 warnings matched what was planted. A deal leaves the list the day it gets a call. Along the way we fixed 15 defects, and 4 calls were scored with their own deal in view.

How we read data: we check a warning against a known answer before a manager sees it.

5×

Billing that charges what customers paid for

What held it back

The billing system worked out which plan a customer had bought from a price code, and quietly fell back to the smallest plan when it did not recognise the code. Until the client sets their real prices, no code would be recognised, so every subscriber would land on the wrong plan. A top-tier customer would have got a fifth of the usage they paid for.

What we did

We used every billing step as a customer would, on features nobody could test yet because the payment account is not switched on. That round found and fixed 31 of 31 problems, this one included, before a single invoice went out.

How we ship: we test the paths that have never run, not just the ones that look finished.

7

Reports a manager can trust

What held it back

When a customer’s CRM had more than one sales pipeline, deals outside the main one went missing from the close rate. One visible win and one hidden loss reads as 100%, when the truth is 50%. Seven features read that number.

What we did

We traced the error to one place and fixed it there, so all seven reports corrected together. The same week we fixed 17 other issues while connecting CRM deals to call scoring.

How we read data: a confident wrong number is worse than a missing one.

Month by month

From prototype
to release.

April 2026

The foundation.

AreaWhat we didThe number
FoundationRebuilt the client’s prototype as a real product9 updates
Call recordingsRead and cleaned the client’s call transcripts144 unique calls from 154 files
CRMBuilt a connection that only reads0 writes to the client’s CRM

The base for scoring was in place.

May 2026

Live, and scoring.

AreaWhat we didThe number
Cloud launchLive on the client’s own accounts from 1 May157 updates in the month
ScoringAI and rules-based scoring live1.6 seconds per call
Pipeline warnings3 agents watching the CRM16 rules

The 8-week core was complete, on plan.

June 2026

Tools for managers.

AreaWhat we didThe number
Manager toolsA guided one-to-one coaching assistant8 sections
Daily briefingsMorning, pre-call, post-call, end of day and weekly plan5 briefings
Data accuracyCredited calls to the right salesperson60 calls corrected

Managers had preparation and coaching tools on real calls.

July 2026

The commercial layer.

AreaWhat we didThe number
DiagnosisA sales-performance diagnosis engine10 measures, 4 on real data
Commercial featuresSelf-serve sign-up, billing, invitations, full data downloadDeployed 29 July
QualityAutomated checks, from zero0 → 287
AccountabilityA record of who changed what37 actions recorded

The commercial features were built; billing waits on the client’s payment keys and prices.

August 2026

Secure, and self-serve.

AreaWhat we didThe number
SecurityTightened every data access rule70 of 70
SeparationLive check on unsigned-in access101 of 101 refused
Hands-on reviewUsed the product as a customer would31 of 31 problems fixed
CRMOne-click connection, used by the client1 click

The client connected their own CRM on the live system.

September 2026

Numbers you can trust.

AreaWhat we didThe number
Smarter scoringCalls scored with their CRM deal in view18 issues fixed
Trustworthy numbersClose-rate error fixed at its source7 reports corrected
AccountabilityActions that left no record now leave one13 more
Stable releaseFull quality suite before release40 fixes

A stable release on 25 September.

October 1–4, 2026

Released.

AreaWhat we didThe number
PipelineEnd-to-end test with sample deals$518,000 across 21 deals
QualityAutomated checks943 of 943 pass
FixesDefects found during the CRM test15 fixed

Released to the live system on 4 October.

More results

Fast,
because it is checked.

Delivery pace

648

Updates shipped, an average of 24 a week.

The busiest week had 71.

Automated checks

943

Every release passes 943 automated checks before it ships.

Built from zero in 10 weeks.

Release gate

10

Full customer journeys re-run on every change; any failing check blocks the release.

Speed never breaks what customers use.

Hands-on review

125

Problems found and fixed by using the product as a customer would, before customers could find them.

Across 6 review rounds.

Scoring speed

75× faster

Scoring a call went from 120–180 seconds to 1.6 seconds.

Fixed within the week the slowdown appeared.

Data accuracy

60

Calls credited to the right salesperson by an AI classifier. 116 training recordings no longer count as sales calls.

On the client’s real calls.

Security

16

Security issues found by reviews built into delivery, and reported privately to the client.

From 6 security reviews.

Reporting

27

Plain-English progress reports to the client, one per planned week.

Backed by 58 detailed work logs.

Accessibility

100 / 100

Accessibility score on the main customer pages, up from 96.

Usable by more people, on more devices.

How we work

  1. Understand the operation

    We learned the client’s coaching method, scorecard and sales calls before writing code.

  2. Pick the right problem

    We built the core coaching loop first, and later weeks only after the core worked.

  3. Build beside the team

    We worked from the client’s own documents and feedback, and reported to them every week.

  4. Prove it before it ships

    Automated checks, hands-on review and a release gate, so problems are found before customers find them.

Have a method
worth turning into a product?

Tell us where your operation feels slow, repetitive or difficult. We will tell you what is worth building, and what the next practical step could be.

Prefer to write directly? enable JavaScript to see the address

Talk to an engineer

No sales theater. Tell us where your operation feels slow, repetitive, or difficult — an engineer reads every message.

Your message goes straight to our engineers at our address.