AI implementation firms · Discovery margin

How Much Profit Is Unpaid Discovery Costing Your AI Implementation Firm?

How loosely bounded discovery erodes project contribution margin, and how to control it without hurting delivery quality.

Leania · For AI implementation firms

DISCOVERY MARGIN

Structure should prevent rediscovery, not create bureaucracy.

All financial examples are synthetic illustrations of internal delivery economics, not verified client outcomes.

A £40,000 AI implementation looks profitable when the contract is signed.

Then discovery starts.

Another process walkthrough. An exception nobody wrote down. Requirements that need rewriting. A business rule nobody can agree on.

Each of these may be fair on its own. But when discovery is planned too thin, or keeps being reopened, the extra effort comes straight out of the project's contribution margin.

In smaller AI implementation firms, the people answering these questions are often your senior specialists. You also need them to scope new work, manage key accounts and run delivery. So the commercial question has two parts. How much discovery really improves the solution? And how much do you absorb because the engagement never had clear limits?

Why discovery costs more than the estimate suggests

Many projects start with a technology-shaped request. “We want an AI agent for customer enquiries.” “We need to automate invoice processing.” Those requests say little about the operating problem.

Take invoice processing. The client describes receiving an invoice, extracting the fields, checking them and posting to finance. A walkthrough reveals invoices with no purchase order, several invoices in one attachment, supplier records that don’t match, unwritten approval rules and rejected cases going back to shared mailboxes with no owner.

The technology may still work, but its scope, exception handling, integration and acceptance criteria have all changed. When details surface after the build starts, discovery comes back in other forms: design queries, change control, defect checks and rework. The hours sit under other project codes, but they still eat into delivery margin.

I would split discovery effort three ways. Essential discovery is the facts you need for a sound implementation decision. Avoidable repetition is asking for the same facts again, or reopening decisions nobody settled. Genuine changes are new requirements or constraints. Cutting essential discovery can increase risk. Avoidable repetition needs structure. Genuine changes may need a commercial decision on scope.

What a 50-hour discovery overrun does to project economics

Take a synthetic fixed-fee example:

  • Contracted implementation fee: £40,000.
  • Planned discovery: 30 hours at £75 an hour (fully loaded internal cost) = £2,250.
  • Actual discovery: 80 hours = £6,000.
  • Extra unrecovered cost: (80 − 30) × £75 = £3,750.

That £3,750 is 9.4% of the project fee. If the original £28,000 delivery budget included planned discovery, the unrecovered overrun cuts contribution margin from (£40,000 − £28,000) / £40,000 = 30% to (£40,000 − £31,750) / £40,000 = 20.625%, or approximately 20.6%. This assumes the fee and all other costs stay the same.

Repeat the same 50-hour overrun across 20 similar projects and it uses 1,000 additional hours and £75,000 in internal delivery cost. This is an illustrative scenario, not an industry benchmark or £75,000 in proven recoverable cash. Approved client changes, other cost shifts and avoided defects change the real numbers.

Synthetic example · Not client data or an industry benchmark

Contribution-margin bridge for one fixed-fee project

  1. Contracted fee£40,000
  2. Original delivery cost, including planned discovery£28,000
  3. Contribution before the overrun£12,000 · 30%
  4. Extra discovery cost− £3,750
  5. Contribution after the overrun£8,250 · 20.6%

Contribution margin before the overrun

30%

Contribution margin after the overrun

20.6%

Contribution margin = (fee − total delivery cost) ÷ fee. Assumes the fee and all other costs remain unchanged. Internal delivery cost, not recoverable cash.

The expensive discovery work may be happening after discovery

A team finishes requirements capture in 30 hours and declares discovery on budget. Then developers sort out conflicting requirements, testers find exceptions nobody wrote down and the client rejects a feature that doesn't match how the work is really handled. Some of this back and forth is fair. Some of it could have been prevented.

Measure three things together. Discovery efficiency is the effort it takes to understand enough to decide. Discovery quality is the evidence for exceptions, controls, decisions and acceptance criteria. Downstream rework is the delivery effort caused by missing or wrong discovery facts.

Another synthetic comparison: Team A spends 30 hours on discovery and 45 hours fixing rework caused by requirements, 75 hours in total at £75 = £5,625. Team B spends 45 hours on discovery and 10 hours on rework, 55 hours in total = £4,125. Team B spent more on discovery and £1,500 less overall. Real teams should use a rate for each role.

The useful measure is total effort to an accepted working outcome, not the speed of the first workshop.

Where I'd look for avoidable discovery effort

Repeated fact-finding: the same rules come up in sales, workshops, design and UAT because nobody keeps a record the team can reuse. Ask which meetings add new evidence.

Undocumented exceptions: the happy path is clear, while missing details, failed validations, overrides and specialist judgements show up late. Sample failed and reworked cases as well as the ones that went smoothly.

Unclear decision ownership: engineers cannot settle disputed business rules. A named business owner, a decision record and a clear route for escalation may help more than another workshop.

Weak handover: the team has long notes, a map and meeting transcripts. Engineers still cannot find the triggers, decisions, exception paths, dependencies, controls or acceptance criteria.

Where avoidable discovery effort comes from

  1. Friction 01

    Repeated capture

    The same rules discussed in sales, workshops, design and UAT

  2. Friction 02

    Missing exceptions

    Failed validations, overrides and specialist judgements emerge late

  3. Friction 03

    Unclear ownership

    Engineers cannot settle disputed business rules

  4. Friction 04

    Weak handover

    Triggers, decisions, exception paths and acceptance criteria are hard to find

  1. Scope volatility
  2. Rework
  3. Margin exposure
Illustrative diagram. It describes no firm and carries no measured figures.

Separate qualification from bounded discovery

First-stage qualification asks whether there is a material operating problem, a plausible opportunity, a sponsor who owns the outcome and a reason to proceed. It needn't resolve every exception.

Bounded discovery sets out what the solution must do, the feasibility risks and the evidence for a reliable estimate. Complex engagements may need discovery priced on its own. Not every prospect needs a long paid assessment. The right commercial model depends on size, uncertainty and the client’s willingness to pay.

Five outputs, sized to the job:

  • 1. Operating problem and baseline.
  • 2. Current-state workflow, with decisions and exceptions.
  • 3. Evidence and uncertainty register.
  • 4. Intervention assessment, working through Stop → Simplify → Standardise → Workflow → Automate → AI.
  • 5. Delivery-ready brief with scope, exclusions, dependencies, acceptance criteria and open decisions.

A £15,000 project does not need the same paperwork as a regulated multi-system programme.

Structure should prevent rediscovery, not create bureaucracy.

Where I'd look before changing the way your team works

Review five to ten completed projects. Compare planned and actual discovery hours. Separate approved, client-paid changes from absorbed effort and genuine scope change. Look at rework caused by requirements in design, engineering and testing.

Does the overrun cluster in certain project types, levels of client maturity, integrations or delivery teams? Is the root cause pricing, decision ownership, unclear scope, patchy process evidence or the same artefacts produced again and again? Each cause needs its own response.

Rank what you find with three tests. How much money is at stake? How strong is the evidence of cause? Can you fix it without adding downstream rework? AI-assisted mapping may cut repetitive manual work, but exception paths, controls and rules still need an expert to check them.

Start with your last five projects

Before hiring more delivery staff or buying another AI tool, I'd want one answer. How much of your current capacity goes on avoidable discovery and requirements rework?

Start with your last five projects. The numbers may point towards better qualification, tighter scope, stronger process evidence or a new way to price discovery.

All financial examples are synthetic illustrations of internal delivery economics, not verified client outcomes.

For AI implementation firms

That's the kind of problem AI-PEX is designed to help implementation firms diagnose, quantify and address consistently.

Explore AI-PEX

Want to run your own project through the same arithmetic? The free AI Project Discovery & Margin Calculator shows the margin movement and builds a PDF diagnostic. Open the free calculator

← All insights