Impact Week findings
& conclusions.
Project overview
- →We've had working sessions with InnovateHR to understand the current state of the process, and which steps can be automated.
- →Our goal is to understand the processes and make the most adequate, achievable, and worthwhile automation recommendations.
- →Instrumentation plus the safest automation plays are the fastest way to start recovering hours and start measuring what's really happening.
- →From the first automation onward, we run measure‑learn‑iterate: that's how we find the next process worth automating.
What achieving your vision looks like (Core Objectives)
- Manual pre‑fill and research — the largest measured block of labour hours
- Wrong or incomplete documents — caught a week after the workbook was already built from them
- No measured baseline — every hours figure today is an estimate, not a fact
We started with documentation analysis and user interviews to map the current state of the process end to end. From there, process engineering shaped a brand‑new process with heavy automation built into the highest‑friction parts we identified. Our methodology also ranks every part by impact — pinpointing where time leaks or steps simply run too long — so we automate the highest‑impact parts first.
Radically reduce the time and labour the client‑onboarding process takes.
Five sections, in the order the case gets made.
Method
How the Current State and Future State documents were actually produced — what took your time, and what didn't.
Read everything first. Spend your time only on what reading couldn't answer.
Five steps, three of them off your calendar entirely. The draft map made the deep-dive productive; the deep-dive made the redesign rankable.
The five steps — left to right in time
Only two needed your team's calendar. time with your team Designli, off your calendar
Read everything you already had
Basecamp, 8 workbooks, the checklist, the timeline — exhausted before a single question was asked.
no meeting neededBusiness context, then field by field
Grace demoed Payroll, Timekeeping, Benefits — naming the source document behind every answer.
your team · 2 sessionsIndexed, graded, then mapped
107 files, one searchable base; 81 wastes registered and rated for whether automation can touch them.
107 filesThe two hardest workbooks, walked step by step
18 stickies came off the board; six corrections made on the spot; first measured labour hours captured.
your team · 1 sessionRescope, re‑rank, then redesign
Re‑ranked by hours lost, not weeks lost — into instrumentation, six automation plays, two process changes.
R0 – R8The method, in full
The complete diagram from the Method document — every card, every arrow. Scroll to see all five lanes.
Legend: amber = time with your team · white = Designli work, off your calendar · teal = what the step produced.
Four habits behind it — and the one thing still to be done
Four habits that did the work
- Read before asking. Three sessions were enough because none were spent establishing facts already in your own systems.
- The draft was validated on purpose. A wrong draft, inferences visibly marked, is a better agenda than a list of questions.
- Every claim carries its evidence. Stated, documented, observed, or inferred — recorded per claim, never quietly promoted.
- The goal was allowed to change. From elapsed weeks to labour hours and concurrent capacity — both documents re‑ranked around it.
Work still to come
- The rest of the workbook set. These two are two of a larger implementation — the same pass needs to run on the others before the labour‑hours case covers all of it.
- A measured baseline. Nearly every hours figure in Current State is an estimate, because the measurement never existed — which is why Future State's first move (R0) is instrumentation, not automation.
- Team composition — who runs the build. Covered in §05, still open.
Current State
How the Benefits and Payroll & Core workbooks get built today — measured where your team has measured it, honest about where nobody has.
What the numbers say about today's process
The engagement pace stays the client's choice. Inside that, the metric this document ranks everything by is labour hours per engagement — what consumes hours matters most; what consumes only calendar is real, but isn't the target.
Kickoff & collection, then the Benefits lane
Simplified from the full swimlane map. rework source legal / external gate
Kickoff call
15 minutes to receive everything Sales knows. Starts the clock.
15 min~50‑item flat checklist
No status view. Credentials, client docs and third‑party docs share one undifferentiated list.
rework sourceOn‑site visit
The strongest step today. Documents pulled at source; issues spotted on the spot.
1–3 daysDocuments arrive, validated by hand
Client and third‑party documents are often outdated, wrong, or incomplete — caught a week later.
rework sourceBroker & advisor intros
Email intros, one meeting — already light.
already lightPre‑fill to ~80%, by hand
Your team pre‑fills the larger workbook before the broker meeting.
1–2h + 4–5h · est.Broker meeting — verification
The broker verifies accuracy and helps complete what's left.
already a confirmationBroker approval
Required before anything goes to CentrallyHR. Cannot be removed.
legal gatePayroll & Core, then the send
Pull from the incumbent system
The majority of payroll data — if the login landed.
1–4hCollect what no system holds
EIN letter, state tax IDs, tax rates — also the documents most often sent wrong.
rework sourcePre‑fill, research included
State/federal law, codes, taxability rules — researched by hand. Largest measured labour block.
~½ dayRep‑only items & open decisions
Rate multipliers and some taxability exist in no document.
ceilingBoth workbooks, one email
A single send point already exists — but nothing checks the two workbooks first.
uncheckedCentrallyHR clarification loop
Questions relayed to the broker. Routine, quick, and fires on every engagement.
every engagementCentrallyHR configures UKG
The boundary of this map.
boundaryThe current-state map, in full
The complete diagram from the Current State document — every node, every rework loop. Scroll to see all four lanes.
Legend: rose = known rework source · amber = legal / external gate · slate = external wait · white = no waste flagged. Cross-lane: validated documents feed the Benefits pre-fill and the payroll collection; the on-site visit secures the incumbent-system login; both lanes converge on the one unchecked email.
Four things worth saying out loud
The incumbent payroll login is the single highest‑leverage item on the checklist. With it, most of the Payroll workbook comes straight out of the client's system in 1–4 hours. Without it, everything downstream degrades to manual entry.
The handbook is a smaller problem than it looked. 1–2 weeks, running in parallel from week 1 — and within this scope it gates only the Benefits eligibility section.
The confirmation‑meeting model already exists — your team built it. Benefits is pre‑filled to ~80% before the broker ever sees it. Payroll follows the same shape.
Both workbooks leave in one email, and nothing reconciles them first. A single send point already exists — the natural home for the checks it currently lacks.
The wastes, ranked by hours lost
| # | Waste | Where | Why it ranks here |
|---|---|---|---|
| 1 | Manual research during Payroll pre‑fill | P.5 | The most mechanical part of the largest measured labour block (~½ day) |
| 2 | Gap‑finding & question drafting, by hand | P.5 · B.5 | Repeated in both lanes to produce every confirmation meeting's question list |
| 3 | Wrong or incomplete documents at intake | 0.7 | A manual check, a week‑later rework loop, and the biggest time loss in the process |
| 4 | Manual entry beyond the incumbent reports | P.1 | The reports are structured; what remains is typed in by hand |
| 5 | Benefits pre‑fill runs entirely by hand | B.4 | 1–2h validation + 4–5h processing, on the larger of the two workbooks |
| 6 | Nothing reconciles the two workbooks before the send | S | An inconsistency surfaces at CentrallyHR instead of in‑house |
| 7 | Client stalls are invisible until they've cost days | 0.2 | Little direct labour lost — but a stall discovered late is what delays the pre‑fill start |
| 8 | The carrier‑access workaround runs silently | Lane B | No hours lost — pure risk, running unnoticed past go‑live |
| 9 | Policy answers live scattered outside the handbook | Handbook | Much smaller than it looked — within this scope the handbook feeds only Benefits eligibility |
Hard boundaries — fixed no matter what changes
| Boundary | Why it is fixed |
|---|---|
| ⚖ Broker approval before the send | Legally required. Brokers are the only experts on plan details and contribution structures |
| ⚖ Retirement advisor verification | Plan documents are too complex to interpret unaided |
| Rate multipliers & some taxability | They come from the payroll rep and exist in no document |
| Undecided client decisions | QuickBooks template, custom reports, bonus treatment — decisions the client hasn't made yet, not information sitting in any document |
| Credentials & authorizations | Roughly a third of the checklist is access and signatures, not documents |
| The clarification loop, entirely | CentrallyHR will always be able to ask questions. Reducible at the source; not removable |
Data necessary to measure the baseline
The business case is a labour‑hours case, and a baseline is what makes it provable. These aren't gaps in the mapping — they're figures no one has ever tracked. Future State's first recommendation, R0, is the mechanism that collects this list.
From your team
- Concurrent implementations per person, today — capacity is the stated payoff, and the baseline doesn't exist
- Benefits hours, tracked rather than estimated — the 1–2h validation and 4–5h processing figures are working estimates; a handful of tracked engagements would confirm them
- Hours in census verification and the remaining hand‑checks — the checks this map flags that still carry no figure at all
From the records
- Wrong‑document rates per type — retirement documents are ~25%; EIN, state tax and workers comp are named but unquantified
- Frequency and duration of the client back‑and‑forth — the biggest time loss in the process and the least measured
- Frequency and cost of the CentrallyHR loop — routine and usually quick, but it fires on every engagement
- Where the workbook lives between InnovateHR, the broker and CentrallyHR — no named system of record is an unsized version‑control risk
Beyond these two workbooks. This document maps two workbooks of the full set. The same analysis — the flow, the numbers, the wastes, the baseline — needs to be repeated for every other workbook before the labour‑hours case covers the whole implementation. The method is now established, so each additional workbook should be substantially faster to map than these first two.
Recommendation
The redesigned process — instrumentation first, then six automation plays and two process changes, ranked by hours recoverable.
The workbooks stop being the pacemaker
Documents are validated the moment they upload. Both workbooks pre‑fill continuously as inputs arrive. Research inside pre‑fill runs on Claude. Remaining gaps become one consolidated question list — turning every meeting into a confirmation pass. Nothing here removes broker approval, advisor verification, or the payroll‑rep consult.
*Design targets expressed as ranges, not measurements — arithmetic on your team's own figures. One target carries no range: zero follow‑up questions from CentrallyHR.
Kickoff & intake, then Benefits — redesigned
redesigned automated with Claudekept as‑is
Kickoff call
Kept. Structured intake goes out the same day.
keptStructured intake + live status
Three tracks, outstanding / received / rejected — visible to both sides.
R7 · redesignedOn‑site visit
Kept — the strongest step today. Dashboard is now the fallback.
1–3 days · keptEvery document validated at upload
A wrong upload bounces same‑day, with the specific reason.
R3 · redesignedPre‑fill to ~80% — continuous
The mechanism your team already runs, now fed continuously and measured.
redesignedGap list generated
Remaining blanks and conflicts become one grouped question list.
R2 · redesignedBroker approval
Legally required — never automated away. Lighter inputs make it easier.
legal gateCensus machine‑checked
Every dependent, SSN, enrollment — checked automatically; a person confirms the flags.
redesignedPayroll & Core, then a gated send
Pull from the incumbent system
Extraction deepened, with anomaly checks — wrong data flagged, not inherited.
R4 · 1–4hResearch automated
State/federal law, codes, taxability — sourced and cited on every value.
R1 · ~½d → 1–2hWeek‑3 meeting → confirmation pass
Walks the generated gap list, not the workbook. Client decisions made here, as today.
redesignedPayroll‑rep items
Rate multipliers and rep‑only taxability — the rep is asked only what only the rep knows.
ceiling · keptReconciliation + completeness gate
The two workbooks must agree; no required field ships blank without a logged reason.
R5+R6 · newOne email to CentrallyHR
Unchanged — both workbooks together, as today.
unchangedClarification loop — reduced at source
Follow‑ups shrink because the package arrives consistent and complete.
reduced, not removedThe redesigned map, in full
The complete diagram from the Future State document. Teal = redesigned; the Claude mark = automated; amber/slate = kept as-is.
Legend: teal = redesigned step · Claude mark = automated with Claude · amber = legal gate, kept · slate = external wait, unchanged · white = kept as-is. Cross-lane: continuous pre-fill (0.5) feeds both workbooks as inputs arrive; the on-site visit unlocks the payroll pull; the handbook gates only Benefits eligibility; both lanes converge at the reconciliation gate before the single send.
Nine changes, ranked by hours recoverable
Impact only — cost deliberately ignored in the ranking. Build estimate is one‑time engineering hours; the saving recurs on every engagement after.
Instrument the process
Measure Benefits hours, validation hours, capacity per person — converts every claim below from extrapolated to measured.
Automated pre‑fill research
~½ day of research → a 1–2h review pass. Compounds hardest on multi‑state clients.
Question & gap generation
One consolidated question list — lets several workbooks resolve in a single meeting.
Document validation at intake
Kills the manual check and the week‑later rework loop — catches ~90% of bad docs at upload.
Deepen extraction
Hours saved, not coverage — plus catching faithfully‑wrong data before it's inherited.
Cross‑workbook reconciliation
Nothing checks the two workbooks against each other today — cheap to insert.
Completeness gate
Together with R5, targets zero clarification questions from CentrallyHR.
Intake form + status dashboard
Little direct labour saved, but removes the chase — survives every open question.
Carrier‑access check
No hours saved — pure risk removal on a workaround that runs silently past go‑live.
Seven of nine estimated ≈ ~62h. R2 and R3 carry two of the four headline numbers on their own and need a scoping pass before a figure goes against them.
What this design keeps, and why
| Kept as‑is | Why |
|---|---|
| The weekly one‑topic cadence — for any client who wants it | It stays available as a client‑care option, set by customer availability alone. The design's job is to guarantee that if an engagement runs eight weeks, the customer is the reason — the workbooks never are |
| The on‑site visit | The strongest single element of the current design — it kills document rework at the source |
| Broker approval & advisor verification | Legally required. Lightened by better inputs; never removed |
| The payroll‑rep consult | Rate multipliers and some taxability exist in no document. A permanent boundary — the design draws the line around it instead |
| The carrier‑access workaround | Access often lands after go‑live and your broker coordinates around it. R8 makes it visible; nothing here pretends to fix it |
| The single email to CentrallyHR | Already a good shape — one send point. The new gates are inserted at it, not instead of it |
| Client decisions with no documentary source | QuickBooks template, custom reports, bonus treatment — decisions nobody has made yet. They're made at the week‑3 meeting, as today |
| The handbook track | 1–2 weeks, runs in parallel, gates only Benefits eligibility. Already small; not worth touching in this scope |
Expected improvement, stated honestly
- Payroll & Core (measured): pre‑fill including research is ~half a day today, and the research is its most mechanical part. R1 + R2 + R4 plausibly reduce it to a 1–2 hour review pass — roughly 2–3 hours saved per engagement in this lane alone, plus whatever the manual document validation currently costs.
- Benefits (estimated): your team's estimates put the hands‑on work at 1–2 hours of document validation plus 4–5 hours of processing per engagement, with a couple of hours of broker and client meetings around it. The same mechanisms apply and the confirmation‑meeting model already works there, so R1–R3 plausibly compress those 5–7 hands‑on hours the way they compress Payroll's half‑day. R0 turns the estimates into tracked figures before the saving is claimed.
- Rework and waiting (unmeasured but named): R3, R5 and R6 attack the two known rework loops — wrong documents at intake (retirement documents arrive wrong roughly a quarter of the time) and the CentrallyHR clarification loop. The wait for corrected client information is the biggest time loss in the process today, and rework hours are the likeliest hiding place of the unexplained labour.
- Capacity: the payoff is more concurrent implementations per person. That baseline doesn't exist yet — R0 creates it. The headline ~10h → ~3h is assembled from your team's own figures, but it stays an estimate until R0 reports — we'd rather confirm it against a measured baseline in eight weeks than defend an invented precision today.
Where the four percentages come from. 60–70% is ~10h of hands‑on work falling to ~3–4h; 50–75% is a half‑day (~4h) pre‑fill falling to the 1–2h review pass; 40–60% is the 1–2h document‑validation pass once ~90% of uploads clear automatically and only exceptions are read by hand; 80–90% is the share of wrong or incomplete documents caught at intake rather than after the workbook was filled from them. The low end of every range is the number we would defend today; the high end is what the design is built to reach.
Assumptions
The Excel workbook format stays. All tooling writes into your existing .xlsx files, never replaces them — brokers and CentrallyHR keep working in the artifact they know.
The on‑site visit is (or becomes) standard practice. The design keeps it as the collection backbone.
A meaningful share of the pre‑fill research is genuinely researchable. The boundary between researchable taxability and payroll‑rep‑only taxability gets drawn field‑by‑field as R1 is built.
Benefits behaves like Payroll under the same mechanisms — supported by your team's estimates (the work has the same shape: gather on‑site, validate, process), confirmed once R0 reports.
The four headline percentages hold only if their inputs do. They rest on one tracked figure (Payroll's ~half‑day pre‑fill) and three estimates. If R0 shows the real hours are different, the percentages move with them — the ranges are wide for that reason.
Risks — and how the design answers them
Faithfully‑wrong extraction. A client's long‑standing payroll error, reproduced at machine speed, becomes inherited liability. Answer: validation is attached to extraction, and every automated fill is blank‑and‑flagged when uncertain, never guessed.
PII. The census is the highest‑PII step in the process. All processing stays inside the engagement's NDA boundary — no personal data in third‑party tooling without explicit sign‑off.
Third‑party friction. Brokers and the advisor work in the workbooks directly and their sign‑off is legally required. The design changes what they receive — more complete, fewer questions — never how they must work.
False rejections at intake. Validation starts warn‑only and flips to enforce per document type as precision proves out. The ~10% that slips through is caught by the downstream gates, not ignored.
Next Steps
Who runs it, and what Designli commits to. Team composition is the one piece of the case still open.
Additional analysis / process engineering time
Automations to fill in workbooks 1 & 2 automated.
Instrument the process
Measure Benefits hours, validation hours, capacity per person
Needs more analysisAutomated pre‑fill research
State/federal law, company size, earning codes, deduction codes, researchable taxability
10 hoursQuestion‑and‑gap generation
Diff what each workbook needs against what the inputs provide; emit one consolidated question list
Needs more analysisDocument validation at intake
Right document? Complete? — against the named failure list
Needs more analysisDeepen extraction
From incumbent payroll reports, with validation attached
12 hoursCross‑workbook reconciliation
At the single send — shared fields across the two workbooks must agree
12 hoursCompleteness gate
Before the email — no required field blank without a logged reason
5 hoursClient intake form + status dashboard
A form we develop: three tracks, live status per item
10 hoursCarrier‑access‑landed process check
Requested / approved / landed, flagged as go‑live approaches
1 hourAutomations to fill in workbooks 1 & 2 completed by Day 30, analysis for 3rd workbook started.
Or Month 2 is free.
Ship intake forms and AI automations.
Six recommendations ship in the first 30 days — the ones we can already estimate with confidence.
- →Automated pre‑fill research — state/federal law, company size, earning codes, deduction codes, researchable taxability
- →Deepen extraction — from incumbent payroll reports, with validation attached
- →Cross‑workbook reconciliation — at the single send, shared fields across the two workbooks must agree
- →Completeness gate — before the email, no required field blank without a logged reason
- →Client intake form + status dashboard — three tracks, live status per item
- →Carrier‑access‑landed process check — flagged automatically as go‑live approaches
This is a mix of what you'll see and what you won't: new software like the web‑form questionnaire, alongside background processes like AI validation and automated follow‑ups running behind the scenes.
The methodology keeps running — new automation targets identified, process engineering and metrics tracking continue.
Or Month 3 is free.
Keep finding what's worth automating next.
The commitment doesn't stop at Day 30 — it becomes an ongoing discipline.
- →We keep using our methodology to identify new automation targets across the rest of the workbook set
- →We keep doing process engineering on every newly identified target before any automation touches it
- →We keep measuring the metrics on the process — hours, error rates, capacity — so every claim stays grounded in data
- →By Day 60, we'll have delivered the automation for the third workbook, and possibly v2 versions of the first two
Team composition — draft, not final
This is the one piece of the case still open. Below is Designli's team builder, rebuilt live for this deck — click a role to add or remove it; allocation and the monthly total update as you go.
Illustrative roster from Designli's standard team builder — which roles this engagement needs, and at what allocation, is what §05 still has to answer.