APODICAIT CONSULTING
AI service-desk deflection · Fixed scope · 90 days

Deflection you can prove.

We raise your self-service resolution rate from a measured baseline to an agreed target within ninety days, counted in your own ticketing system on a query you keep. The final third of our fee is invoiced only if we reach it.

Book a baseline call Estimate your gap

No seats to license. No platform to buy.
No “tickets avoided” figure that only we can calculate.

90
Days, fixed
33%
Of the fee at risk
5
Guardrails, all must hold
1
Number, defined in writing
Why anyone should believe us

Most AI deflection projects end in a slide deck. Ours ends in a clause.

Everyone in this market sells capability. We sell a number — in writing, in your contract, with our own fee behind it. The scope is fixed before we start, the metric is defined before we start, and the last third of the fee moves only when the metric does.

Extract · standard statement of work 4.3 Contingent Fee. Thirty-three per cent (33%) of the Total Engagement Fee shall become invoiceable only upon Verified Attainment of the Target Self-Service Resolution Rate recorded in the Measurement Charter (§3), as computed by the Agreed Query against the Client’s system of record, and subject to all Guardrail Conditions (§4.4) holding at the Day 90 measurement date.

4.5 Non-Attainment. If the Target is not attained and the Client has met the Dependencies at §5.1, the contingent portion shall not be invoiced, and the Consultant shall perform a thirty (30) day Remediation Sprint at no additional fee.
Apodica Consulting LLC · SOW template v4 · §4
What we measure

One number, defined tightly enough to argue with.

Self-Service Resolution Rate is the share of in-scope contacts that reach a resolution without a human agent touching them. It is computed from your system of record, using a query we write in week one and hand over on day ninety.

Self-Service Resolution Rate SSRR = A ÷ (A + B)

A = in-scope contacts closed at Tier 0
    (article marked resolved, virtual-agent session
    ended with no ticket, catalog request fulfilled
    with zero agent touch)

B = in-scope contacts closed by a human agent at
    any tier, including escalations out of A
What counts as “in scope”
The intent taxonomy agreed in the Measurement Charter — typically your top 20 to 30 intents, covering 55 to 75 per cent of contact volume. Everything outside that set is excluded from both sides of the ratio, so we can neither win nor lose on work we never touched.
How the baseline is set
Computed over the trailing twelve months, frozen in writing on day 10, and never recomputed. Seasonality is handled with a same-period index rather than by re-picking the window, so a quiet December cannot be made to look like progress.
The 72-hour repeat-contact rule
A self-service session followed within 72 hours by a ticket on the same intent from the same person counts in B, not A. Deflection that bounces is not deflection. This rule typically reclassifies 8 to 15 per cent of apparent wins, and we disclose that figure on day 10 rather than at day 90.
Who runs the query
Your team, on your data, using our query. We ask to be in the room for the verification run; we never ask to be the source of truth. The query is version-controlled and handed to you, so you can re-run the result every quarter after we have gone.
Guardrails

A deflection number is easy to fake. Five conditions stop us.

Any consultant can raise a self-service rate by making a human harder to reach. All five of these must hold at day ninety, or the target counts as missed and the contingent third goes uninvoiced.

Guardrail 01

Demand cannot be suppressed

In-scope contact volume stays within ±20% of the seasonally indexed baseline. We cannot win by quietly hiding the intake channel.

Guardrail 02

Reopens cannot rise

Reopen rate stays at or below baseline plus two percentage points. A closed ticket that comes back was never a resolution.

Guardrail 03

Satisfaction cannot fall

Ticket CSAT stays within 0.2 of baseline, on your existing scale and at your existing sampling rate.

Guardrail 04

Majors cannot slip

P1 and P2 mean time to resolve must not degrade. Tier 0 work cannot consume the people who handle real incidents.

Guardrail 05

Escalations stay visible

Escalation rate out of Tier 0 is reported every week in the same readout, whether it flatters us or not.

Swipe to see all five →

Estimate

Roughly what is your gap worth?

Four inputs, and nothing leaves your device — this runs entirely in your browser. It is a sizing estimate rather than a proposal: a real target comes from your own ticket data and is written into a Measurement Charter before any work begins.

Day-90 target—
Points of lift—
Contacts off the queue, monthly—
Agent hours recovered, yearly—
The engagement

Ninety days, five phases, one readout every Friday.

The scope is fixed at signature. If we find more opportunity than the target requires, it goes into the roadmap you keep — not into a change order.

Days 1–10 · Instrument and freeze
Deliverable: Measurement Charter. Read access to the ITSM, knowledge base, portal analytics, chat and phone logs. Twelve months of ticket data extracted and profiled. Baseline SSRR computed, agreed and frozen, with the query placed under version control. Target, guardrails, exclusions and client dependencies signed by both sides. No build work starts until this document is signed.
Days 8–25 · Demand and gap analysis
Deliverable: Deflection Model. AI-assisted intent clustering across twelve months of ticket text and chat transcripts. Every intent scored on volume, handle time, variance and automatability. A knowledge gap map answering four questions per intent: does an article exist, can it be found, is it correct, and does it actually resolve the issue. Analysis of the portal searches that return nothing usable. The output is a ranked list of intents with the point contribution each is expected to make.
Days 20–55 · Build Tier 0
Deliverable: a published self-service estate. Knowledge articles drafted from your own resolved tickets, rewritten to a resolution-first template, and reviewed by your subject-matter experts before anything publishes. Search and retrieval tuning, with a synonym and acronym dictionary built from genuinely failed queries. Virtual agent intents built and conversation-designed, with a clean handoff to a human on every path. Portal information architecture rebuilt around your actual top tasks.
Days 40–75 · Automate the workflows
Deliverable: live automations. Five to eight of the highest-volume request types automated end to end — typically access requests, password and MFA resets, joiner-mover-leaver tasks, software entitlement, distribution lists, VPN and hardware refresh. Approval routing, integration triggers, auto-close and fulfilment SLAs configured in your platform. Every exception path documented, so nothing fails silently into a queue nobody watches.
Days 60–90 · Adoption and proof
Deliverable: a verified result and the handover pack. Agent enablement, with a deflect-to-knowledge habit built into ticket close. Portal-first routing and manager communications, scheduled rather than sprayed. A weekly SSRR readout against target, with the guardrails shown alongside. A day-90 verification run in your system, witnessed by both sides. Handover of the queries, runbooks, article template, governance cadence and a twelve-month roadmap.
What stays with you

Everything we build is yours, in your platform, without us.

Measurement Charter

The signed definition of metric, baseline, target, exclusions and guardrails. Four pages, and the whole argument.

Deflection Model

Ranked intents with volume, cost to serve, and the point contribution each is expected to make.

Knowledge estate

Typically 40 to 90 published articles on a resolution-first template, plus the template and authoring standard.

Search configuration

Tuned relevance, a synonym and acronym dictionary, and the failed-query report that keeps feeding it.

Virtual agent build

Configured intents and conversation flows in the platform you already license, with escalation paths that behave.

Automated workflows

Five to eight request types running end to end, documented, with exception handling and named owners.

The queries

Version-controlled SQL or platform reports for SSRR and every guardrail, so you can re-run the result after we leave.

Governance cadence

Who reviews what and how often, plus the twelve-month roadmap of intents we deliberately left on the table.

Qualification

We turn down more of these than we take on.

A contingent fee only works if the engagement is winnable. This is what we check on the first call — candidly, because a poor fit costs us a third of the fee.

A good fit

  • 200 to 2,000 employees, with an internal service desk of roughly 3 to 20 agents
  • 800 to 6,000 contacts a month flowing through one system of record
  • An established ITSM platform — ServiceNow, Jira Service Management, Freshservice, Ivanti, Halo, SysAid or Zendesk
  • A portal and knowledge base that exist and are underused
  • Current self-service resolution somewhere between 3% and 20%
  • A leader who can name two SMEs and get an article published within three days

Not a fit

  • Ticket data spread across email, spreadsheets and three tools with no shared identifier
  • A platform migration or ERP go-live landing inside the ninety days
  • Self-service already above 35%, where the remaining intents rarely justify the fee
  • Fewer than roughly 400 contacts a month, where the arithmetic becomes too thin to defend
  • Nobody able to approve knowledge content within a week
  • A mandate to reduce service desk headcount during the engagement, which ends the adoption work on day one
Investment

A fixed fee, set by ticket volume rather than by hours.

One price, agreed before we start, covering the whole ninety days. No rate card, no timesheets, and no change orders for anything inside the signed scope.

Tier A
$58,000
200–500 staff · under 1,200 contacts a month
  • Intents in scope up to 15
  • Knowledge articles 40+
  • Automated workflows 5
  • Typical target lift +10 to +14 pts
Tier B · most common
$86,000
500–1,200 staff · 1,200–4,000 contacts a month
  • Intents in scope up to 25
  • Knowledge articles 60+
  • Automated workflows 7
  • Typical target lift +12 to +18 pts
Tier C
$124,000
1,200–2,000 staff · 4,000+ contacts a month
  • Intents in scope up to 35
  • Knowledge articles 90+
  • Automated workflows 8, multi-region
  • Typical target lift +14 to +20 pts

Swipe to compare tiers →

40%

At signature. Covers instrumentation, extraction and the Measurement Charter.

27%

At day 45, on acceptance of the Deflection Model and the first published build.

33%

At day 90 — only on verified attainment of target, with all five guardrails holding.

Start smaller: the Deflection Baseline Assessment
Two weeks, $9,500. We compute your real baseline, size the opportunity, rank your intents and tell you the target we would sign up to. It is credited in full against the engagement if you proceed, and it is yours to keep if you do not.
After day 90: Sustain
$4,500 to $9,000 a month. Monthly intent review, article refresh, new workflow builds and the SSRR readout, so the number holds once the novelty has worn off. Deflection decays without maintenance: an untended knowledge estate loses a few points a year as systems and policies change.
What we need from you
Read access within five business days. Two named subject-matter experts at roughly four hours a week between them. Article approval inside three business days. Change windows on the agreed release schedule. If a dependency slips, the clock extends day for day and we tell you in writing that week — never as a surprise at day eighty-five.
Straight answers

Questions we get on the first call.

Are you going to tell me to cut my service desk?
No, and we will say so in front of your team. Every engagement we have scoped has a backlog the desk never reaches — asset hygiene, problem management, project support, the requests that sit for a week. Deflection buys that capacity back. If the mandate is headcount reduction during the ninety days, we decline the work, because the people whose knowledge we need stop cooperating the moment they understand the brief.
We already pay for AI features in our ITSM. Why do we need you?
Most of our clients do, and that is usually the point. Licensing a capability and achieving a rate change are different problems. The gap is almost never the model: nobody has clustered twelve months of ticket text into intents, nobody has written articles that actually resolve, nobody has tuned search against the queries that fail, and nobody owns the number. We use what you already own wherever we can, and we tell you plainly when something you are paying for is not worth configuring.
What if we miss the target because of something on our side?
The dependencies are listed in the statement of work: access, SME hours, approval turnaround and change windows. If one slips, the timeline extends day for day and we issue a written delay notice that week. If the target is missed while you met the dependencies, the contingent third is never invoiced and we run a thirty-day remediation sprint at no fee.
Who owns the AI-generated knowledge content?
You do, on publication, without restriction. Drafts are generated from your own resolved tickets and reviewed by your subject-matter experts before anything goes live — no article publishes on our say-so. We also hand over the authoring template and standard, so your team keeps producing at the same quality after we leave.
Does our ticket data leave our environment?
Only what we agree in writing, and we would prefer the answer to be nothing. The default is analysis inside your tenancy, or on an isolated extract with identifiers stripped. We will sign your DPA, work to your data classification, and put retention and deletion terms in the statement of work rather than in a privacy policy nobody reads. No client data is ever placed in a consumer AI tool.
How much of our team’s time does this take?
Roughly four hours a week between two named subject-matter experts, plus article approvals inside three business days. Your reporting analyst runs the baseline query on day 10 and the verification query on day 90. Beyond that, the Friday readout takes twenty minutes and is the only standing meeting we ask for.
Next step

Start with the number you have today.

Forty-five minutes. Bring whoever runs the service desk. We will walk your ticket mix, name the intents we would target, and tell you on the call whether we would sign a contingent fee against your data — including when the answer is no.

Book a baseline call Request the one-pager

hello@apodicaconsulting.com
(602) 555-0148
Phoenix, Arizona · delivered remotely across the US and Canada