In the enterprise

The systems I design.

• Continuous control monitoring that raises only the exceptions

  Comparing controls across systems and data took manual effort, time and judgement, long after the event.

  Data platform · Conversational AI · Evaluation and governance

  Knowledge is scattered across systems · The signal arrives too late · Every answer needs its evidence

• Agents that answer a team's questions from its own documents and records

  The answers a team needed lived in unstructured material, and the appetite for AI outran any agreement on where to start.

  Agentic systems · Retrieval and search · Conversational AI

  The knowledge lives in unstructured data · Nobody has agreed the first use case · The data can't go everywhere

• An AI-enabled configure-price-quote application that turns requirements into formal sales proposals with all supporting documents

  Turning a customer's requirements into a valid product specification, with prices and a delivery schedule, was slow specialist work repeated for every opportunity.

  AI-native applications · Systems integration · Data platform

  Knowledge is scattered across systems · It has to be priced before it's built · Nobody has agreed the first use case

• Plain-language analytics on governed definitions

  Teams wanted to ask questions of their data directly, without an assistant returning a different number from the dashboard.

  Conversational AI · Data platform · Evaluation and governance

  Every answer needs its evidence · Knowledge is scattered across systems · The people who own the work have to use it

• Spreadsheet decision models rebuilt as applications the owners still control

  Recurring decisions ran on spreadsheets that had hit their ceiling, fed by hand from separate systems, and the people who built them needed to keep control.

  AI-native applications · Predictive scoring · Data platform

  The spreadsheet is the system · The people who own the work have to use it · It has to be priced before it's built

• Agreements drafted from a clause library for a person to approve

  Agreements were assembled by hand from templates and records, and a change to the underlying rules meant reworking every one.

  Document and vision AI · Agentic systems · Systems integration

  The knowledge lives in unstructured data · The systems of record stay · Nobody has agreed the first use case

• Technical documents turned into structured data

  Documents held detail the business needed as data, and reading them by hand was both the bottleneck and the cost.

  Document and vision AI · Evaluation and governance · Data platform

  The knowledge lives in unstructured data · Nobody knows if it works until it meets real data · It has to be priced before it's built

• An agent that assembles the evidence behind an operating event

  The context needed to understand an operating event sat across several systems, and someone assembled it differently every time.

  Agentic systems · Systems integration · Retrieval and search

  The systems of record stay · Every answer needs its evidence · The signal arrives too late

• Long-form answers drafted from a governed document set

  Answering a long, standard question set meant finding the right passage in a large document library by hand, every time.

  Retrieval and search · Document and vision AI · Evaluation and governance

  The knowledge lives in unstructured data · Every answer needs its evidence · Nobody knows if it works until it meets real data

• Field work captured once on the device, straight into the core system

  Work done away from a desk was recorded on paper and transcribed later, so errors surfaced late and billing waited on all of it.

  AI-native applications · Systems integration · Hybrid and edge

  The pilot automates today's workflow · The people who own the work have to use it · The data can't go everywhere

• Inference placed on owned hardware, so the material never leaves the network it sits on

  Some of the material an assistant needed to read could not be sent to a service outside the network it sat on, and the work still had to run.

  Hybrid and edge · Agentic systems · Systems integration

  The data can't go everywhere · It has to be priced before it's built

• Enterprise search that keeps each source's permissions and leaves the data in place

  Finding an answer meant knowing which repository held it, what a person could see depended on where it lived, and some of it could not leave the country.

  Retrieval and search · Systems integration · Hybrid and edge · Agentic systems

  Knowledge is scattered across systems · The knowledge lives in unstructured data · The data can't go everywhere

• Inbound demand scored and routed on the systems already in place

  Inbound arrived from several sources into a system of record that was staying, and deciding who followed up was manual.

  Predictive scoring · Systems integration · Agentic systems

  The systems of record stay · The platform choice is still open · Nobody has agreed the first use case

• A memory layer rebuilt as one store with one writer, so it stops disagreeing with itself

  Several stores of the same knowledge, built at different times with no owner, contradicted each other, and the warning that mattered did not surface when it mattered.

  Retrieval and search · Evaluation and governance · Agentic systems

  Knowledge is scattered across systems · Every answer needs its evidence

• On-premises anomaly detection that sends on only the signals

  Continuous capture produced more material than anyone could review, on sites where the connection could not carry it in time.

  Hybrid and edge · Document and vision AI · AI-native applications · Evaluation and governance

  The data can't go everywhere · The signal arrives too late · Nobody knows if it works until it meets real data

• Customer onboarding redesigned, then automated with what was already owned

  New customers waited weeks for first value across email, re-keyed data and systems that did not talk to each other.

  Systems integration · AI-native applications · Data platform

  The pilot automates today's workflow · The systems of record stay · The people who own the work have to use it

• A planning side and a deterministic execution side, with one database as the only contract

  Judgement and execution ran in the same place, so a decision could not be reviewed apart from the mechanism that acted on it, and nothing stopped it acting.

  Agentic systems · Systems integration · Evaluation and governance

  The signal arrives too late · Every answer needs its evidence · Nobody knows if it works until it meets real data

• A data and AI platform chosen based on the use cases

  Several platform decisions had to be made together, each one constraining the others, before any use case could be built.

  Data platform · Systems integration · Evaluation and governance

  The platform choice is still open · Nobody has agreed the first use case · Knowledge is scattered across systems

• Outbound prospecting rebuilt as data enrichment and model-assisted fit scoring, landing in the system of record

  Deciding who to approach meant researching every account and contact by hand, and the reasons behind each choice sat in scattered lists nobody could sort or reuse.

  Predictive scoring · Agentic systems · Systems integration

  Knowledge is scattered across systems · The spreadsheet is the system

• A read-only tool layer over a system with no interface, with everything it returns marked untrusted

  The material an agent had to read sat behind a system with no interface to build against, and anything it read could carry instructions of its own.

  Systems integration · Agentic systems · Retrieval and search

  The systems of record stay · The data can't go everywhere · Every answer needs its evidence

• A relationship system of record with a tool layer an agent can read and write

  The record of every relationship and open conversation ran in spreadsheets, kept by hand from several places, with no history and nothing an assistant could read.

  Systems integration · Agentic systems

  The spreadsheet is the system · Knowledge is scattered across systems · The data can't go everywhere

• Early warning of delivery risk, with the evidence attached

  The early signs of a slipping schedule sat in correspondence and reports weeks before they reached a status report.

  Predictive scoring · Retrieval and search · Evaluation and governance

  The signal arrives too late · Every answer needs its evidence · The knowledge lives in unstructured data

• Where AI belongs across existing technologies

  The business wanted to know where AI belonged across systems it did not intend to replace.

  Systems integration · Retrieval and search · Agentic systems · AI-native applications

  The systems of record stay · The platform choice is still open · Nobody has agreed the first use case

• An end-to-end workflow rebuilt as orchestrated agents

  Every handoff along a core workflow was re-keyed and re-interpreted, and the status of any item had to be asked for.

  Agentic systems · Systems integration · Data platform

  The pilot automates today's workflow · The spreadsheet is the system · It has to be priced before it's built

In the firm

The firms I help build.

• Certification made a condition of promotion

  Skill was asserted rather than evidenced, so the firm could not show depth to a buyer or find its own gaps before a pursuit exposed them.

  The people

• Career paths set by competency, and a growth plan for every person

  People could not see what the next role asked of them, and the firm only learned someone was leaving when they resigned.

  The people

• A repeatable way to win work alongside a platform vendor's field teams

  Joint pursuits depended on who was in the room, and the vendor people who decide which firms get invited sat across an organization nobody had mapped.

  The channel

• A delivery method defined and taught, so every engagement ran the same way

  Quality tracked whoever happened to be staffed, because how the firm ran an engagement was learned on the job and never written down.

  The practice

• A content and demand engine aimed at the people who refer the work

  Reaching the people who send work was done by hand, one relationship at a time, and stopped whenever the team got busy.

  The demand

• An architecture review board that scores designs before they are built

  Architectural decisions were made inside each engagement and only questioned once something broke, by which point changing them meant rebuilding.

  The practice

• An outbound play run end to end, from target list to qualified conversation

  Almost everything arrived through one channel, and the firm had never tested whether it could generate and qualify its own leads.

  The demand

• A paid discovery packaged so a partner's own sellers could use it

  A partner's sellers could not size their own deals without technical depth they did not have.

  The channel

• A single estimating method, and every scope change traced to its cause

  What was sold and what could be built kept drifting apart, and nobody could say whether the cause was the estimate, the design or the client.

  The sale

• A go-to-market and revenue model for a growing firm

  Revenue depended on a few leaders' relationships, pricing was a single blended rate, and the partner channel was informal.

  The position

• Past work packaged into offers that stack on one engagement

  The firm sold hours, so every deal was a custom negotiation, and buyers wanted a defined outcome before they would commit.

  The offer

• Partner-program credentials chosen for the deals they open

  The rating a partner carries decides which conversations it is shortlisted for, and the firm had not chosen which ones to go after.

  The channel

• Positioning rebuilt from market research, segment by segment

  The firm sold across ecosystems with no focus, so neither buyers nor technology vendors saw it as expert in any, and its message read like everyone else's.

  The position

• A practice run on numbers agreed before anyone was held to them

  The firm had no agreed measure of margin, capacity or how well it forecast, so a busy quarter and a profitable one looked the same until both were over.

  The practice

• A delivery practice scaled by dividing it along what each team owns

  Demand arrived faster than delivery could absorb it, and the customer, the platform and the build were each looked after by whoever was nearest.

  The practice

• A technical presales function stood up from scratch

  The firm architected its solutions after the contract was signed, in delivery.

  The sale

• Pricing moved off a single blended rate onto a role-based rate card

  A single blended rate meant every negotiation discounted the whole team, and margin moved without anyone deciding that it should.

  The offer

• A narrow product bet, argued from why the platform pitch kept losing

  The platform pitch everyone wanted to sell kept losing to cheaper, narrower products.

  The offer

• A satisfaction and reference loop that turned finished work into proof

  Good work ended quietly, and nobody asked for the evidence that would have won the next one.

  The demand

• Client retention designed as a motion, from first project to program

  Repeat work depended on somebody remembering to ask, long after the team that earned the right to ask had moved on.

  The practice

• A library of reusable assets harvested from finished engagements

  Every engagement rebuilt the same foundations, and the reusable parts stayed on the projects that first paid for them.

  The offer

• A commercial organization designed by role and sequenced by hire

  Everyone sold a bit of everything, so no part of the cycle had a specialist and hiring was a reaction to whoever left.

  The people

• A sales cycle defined end to end, owned stage by stage and taught to the team

  How work was won lived in habit and in a handful of people, so there was nothing to onboard a new seller with.

  The sale

• An operating model that names who owns what between selling and delivering

  Selling and delivering each assumed the other held the relationship, the forecast and the margin, and all three slipped in the gap.

  The practice

• A generalist firm repositioned around one specialty

  The firm competed on breadth, and the vendor field teams who decide which partners get invited did not know it existed.

  The position

• Wins and losses read back into how the next one is priced

  Pursuits were lost for reasons nobody wrote down, so the same causes kept recurring and instinct set the price.

  The sale